Humanity’s Last PRD
Latest drop: Haiku 5.5 · 8 Oct
Tasks / Strategy

Call product-market fit

Can the model tell real product-market fit from a blended survey, a launch spike or a flattering metric, and say where to scale?

Task type v1.1 · 2 tasksLast changed 6 Oct 2026 · ChangelogDifficulty

What AI gets right here, and what you’ll still have to catch

From 16 graded outputs by 8 models. 50% were usable with at most a quick edit.

Reliably right

  1. Respects explicit constraints100% pass
    The output is a memo addressed to Sam, within 1,200 words, and respects the form and length requirements.
    GPT-6 Astra · ChatGPT · 38%, and the retention that isn't
  2. Finds who it's a must-have for100% pass
    It identifies recruiters as the must-have segment, noting what they value (candidate summaries, ATS scorecards, focus on the candidate) and builds the plan around them.
    GPT-6 Astra · ChatGPT · 38%, and the retention that isn't
  3. Spots false fit100% pass
    The memo calls out the August promotional MRR spike, the active-definition change, and the early erosion as misleading signs of fit.
    GPT-6 Astra · ChatGPT · 38%, and the retention that isn't

Where it slips

  1. Uses the supplied evidence correctly28% pass
    The claim that the cohort jumps line up with the August 3 definition change is not supported; the jumps occur in the week of July 27–August 2, one week before the change, so the evidence is misused.
    GPT-6.1 Sol · API · 38%, and the retention that isn't
  2. Gets the base of every number right44% pass
    The month-6 retention percentages are computed from incorrect sums and denominators, so the base is wrong.
    GPT-6 Luna · API · Do we have fit, or a spike?
  3. Unpicks the 38%53% pass
    It separates by persona but does not break down the 38% by tenure into ~11% very disappointed for users under two weeks and ~47% for established users, which the criterion explicitly requires.
    Haiku 5.5 · API · 38%, and the retention that isn't

How it’s graded

Each check is built on how world-class product and growth leaders do this work, as they’ve described it on Lenny’s Podcast, and names the leader it comes from.

  1. Judges fit by retention, not sign-ups

    Does the call on product-market fit rest on whether retention curves flatten, segment by segment, rather than on sign-ups, revenue growth or a survey alone?

    Passes when Reads retention by segment from the data and lets it decide where fit exists, with the survey as supporting evidence.

    Todd JacksonLenny’s PodcastSean EllisLenny’s Podcast

  2. Finds who it's a must-have for

    Does the output profile the users who would be very disappointed without the product, and aim the plan at them and at the users closest to them?

    Passes when Names the segment or persona where the product is a must-have, what they value, and builds the plan around them.

    Rahul VohraLenny’s Podcast

  3. Spots false fit

    Does the output catch evidence that looks like fit but isn't, such as revenue won through custom work, a launch spike, a promotion, or a metric whose definition changed?

    Passes when Names each piece of flattering evidence in the material and explains why it isn't fit.

    Product and growth leadersOn Lenny’s Podcast

  4. Doesn't scale before fit

    Does the recommendation avoid pouring spend or headcount into segments that haven't shown fit, and say what would justify scaling them later?

    Passes when Scales only where fit is shown, and sets the evidence that would justify widening.

    Todd JacksonLenny’s Podcast

Plus our standard checks

Uses the supplied evidence correctly · Addresses the actual decision · Respects explicit constraints · Identifies material uncertainty · Avoids unsupported claims · Produces the required deliverable · Gets the base of every number right · and 2 written for each task, which you’ll see in the tasks below.

The tasks

Read the brief, then compare up to three outputs, each with the LLM judge’s verdict on every check. Highlights mark what a PM had to fix.

The brief

You're the Head of Product at Relay. Our CEO, Dana Kowalski, wants to triple the sales team next quarter, "because we clearly have product-market fit", and has asked for your view before Thursday's leadership meeting. Write a memo for her of no more than 900 words: do we have product-market fit, for whom, and what should we do next quarter instead of (or as well as) tripling sales? The files are attached below. Work from the data, not from anyone's summary of it.

What the model was given5 items: About Relay, pmf_survey_responses.csv (sent in September to every paying account's admin), logo_retention_by_cohort.csv (paying accounts still active, by sign-up month and segment), signups_by_month.csv, q3_new_business.md
About RelayShift scheduling software for teams that work in shifts: healthcare clinics, hospitality, retail and others. Sold per location per month. 1,120 paying accounts.
pmf_survey_responses.csv (sent in September to every paying account's admin)response_id,segment,company_size,role,months_using,answer,main_benefit R1001,Other,10-50,Operations manager,12,very disappointed,Fine for now R1002,Hospitality,50-200,Area manager,12,somewhat disappointed,Staff like it R1003,Retail,50-200,Store manager,7,not disappointed,Staff can see the rota R1004,Healthcare clinic,50-200,Office manager,8,very disappointed,On-call swaps that follow our rules R1005,Retail,50-200,Regional manager,8,not disappointed,Shift swaps R1006,Hospitality,200-500,Owner,2,very disappointed,Not sure yet R1007,Retail,50-200,Store manager,4,very disappointed,Easier than paper R1008,Other,50-200,Owner,12,very disappointed,Fine for now R1009,Retail,10-50,Store manager,1,not disappointed,Fine for now R1010,Hospitality,10-50,Head chef,11,not disappointed,Shift swaps R1011,Hospitality,50-200,Owner,8,somewhat disappointed,Shift swaps R1012,Retail,50-200,Store manager,1,very disappointed,Shift swaps R1013,Healthcare clinic,50-200,Operations manager,12,very disappointed,"Rota done in an hour, not a day" R1014,Hospitality,50-200,Owner,12,somewhat disappointed,Easier than the spreadsheet R1015,Hospitality,200-500,Area manager,5,not disappointed,Nice app R1016,Healthcare clinic,200-500,Office manager,9,somewhat disappointed,Staff stop texting me at night R1017,Hospitality,10-50,Head chef,5,somewhat disappointed,Staff like it R1018,Hospitality,200-500,Area manager,12,somewhat disappointed,Not sure yet R1019,Healthcare clinic,50-200,Practice manager,14,very disappointed,Staff stop texting me at night R1020,Other,10-50,Team lead,6,not disappointed,Shift swaps R1021,Healthcare clinic,200-500,Operations manager,3,somewhat disappointed,Covering sick calls fast R1022,Retail,50-200,Owner,3,somewhat disappointed,Shift swaps R1023,Healthcare clinic,200-500,Office manager,4,very disappointed,Covering sick calls fast R1024,Hospitality,200-500,Owner,1,somewhat disappointed,Staff like it R1025,Other,50-200,Team lead,11,somewhat disappointed,Shift swaps R1026,Hospitality,10-50,General manager,13,very disappointed,Easier than the spreadsheet R1027,Hospitality,10-50,General manager,4,somewhat disappointed,Shift swaps R1028,Hospitality,10-50,Head chef,6,somewhat disappointed,Not sure yet R1029,Hospitality,10-50,Owner,6,somewhat disappointed,Easier than the spreadsheet R1030,Healthcare clinic,50-200,Office manager,2,not disappointed,On-call swaps that follow our rules R1031,Retail,50-200,Regional manager,14,very disappointed,Staff can see the rota R1032,Healthcare clinic,50-200,Office manager,2,somewhat disappointed,On-call swaps that follow our rules R1033,Retail,10-50,Store manager,3,somewhat disappointed,Shift swaps R1034,Hospitality,10-50,Area manager,13,not disappointed,Easier than the spreadsheet R1035,Hospitality,10-50,Area manager,9,not disappointed,Shift swaps R1036,Other,50-200,Owner,13,not disappointed,Saves time R1037,Retail,10-50,Store manager,11,very disappointed,Staff can see the rota R1038,Retail,50-200,Store manager,10,somewhat disappointed,Easier than paper R1039,Other,50-200,Owner,5,very disappointed,Shift swaps R1040,Other,50-200,Team lead,3,not disappointed,Saves time R1041,Retail,10-50,Owner,10,not disappointed,Shift swaps R1042,Other,10-50,Owner,6,somewhat disappointed,Saves time R1043,Healthcare clinic,200-500,Clinical lead,6,very disappointed,On-call swaps that follow our rules R1044,Healthcare clinic,50-200,Office manager,12,very disappointed,Credential checks before a shift R1045,Hospitality,200-500,Area manager,12,not disappointed,Not sure yet R1046,Healthcare clinic,50-200,Office manager,5,somewhat disappointed,Covering sick calls fast R1047,Healthcare clinic,50-200,Operations manager,10,somewhat disappointed,Covering sick calls fast R1048,Healthcare clinic,200-500,Clinical lead,12,not disappointed,Covering sick calls fast R1049,Hospitality,10-50,Area manager,11,somewhat disappointed,Nice app R1050,Other,50-200,Team lead,6,somewhat disappointed,Saves time R1051,Hospitality,10-50,Head chef,9,somewhat disappointed,Shift swaps R1052,Hospitality,10-50,Area manager,3,not disappointed,Shift swaps R1053,Retail,50-200,Regional manager,7,not disappointed,Fine for now R1054,Retail,10-50,Regional manager,13,somewhat disappointed,Fine for now R1055,Retail,50-200,Store manager,4,very disappointed,Fine for now R1056,Healthcare clinic,50-200,Clinical lead,4,very disappointed,Covering sick calls fast R1057,Healthcare clinic,50-200,Clinical lead,6,very disappointed,On-call swaps that follow our rules R1058,Hospitality,10-50,Area manager,11,somewhat disappointed,Nice app R1059,Healthcare clinic,200-500,Operations manager,14,somewhat disappointed,Credential checks before a shift R1060,Hospitality,10-50,Owner,5,very disappointed,Staff like it R1061,Healthcare clinic,50-200,Clinical lead,1,very disappointed,Covering sick calls fast R1062,Retail,50-200,Regional manager,6,somewhat disappointed,Easier than paper R1063,Hospitality,200-500,Area manager,7,somewhat disappointed,Nice app R1064,Other,10-50,Team lead,7,somewhat disappointed,Saves time R1065,Healthcare clinic,50-200,Clinical lead,10,very disappointed,Staff stop texting me at night R1066,Retail,10-50,Store manager,4,not disappointed,Fine for now R1067,Hospitality,10-50,General manager,8,very disappointed,Not sure yet R1068,Other,50-200,Owner,4,somewhat disappointed,Fine for now R1069,Hospitality,50-200,General manager,5,not disappointed,Nice app R1070,Healthcare clinic,50-200,Office manager,5,somewhat disappointed,Covering sick calls fast R1071,Retail,50-200,Store manager,11,not disappointed,Staff can see the rota R1072,Hospitality,10-50,Head chef,14,somewhat disappointed,Easier than the spreadsheet R1073,Healthcare clinic,50-200,Operations manager,14,somewhat disappointed,"Rota done in an hour, not a day" R1074,Hospitality,10-50,General manager,13,not disappointed,Easier than the spreadsheet R1075,Retail,50-200,Owner,7,not disappointed,Staff can see the rota R1076,Hospitality,200-500,Area manager,1,very disappointed,Shift swaps R1077,Retail,50-200,Owner,1,not disappointed,Shift swaps R1078,Healthcare clinic,50-200,Practice manager,12,somewhat disappointed,Credential checks before a shift R1079,Healthcare clinic,200-500,Operations manager,9,very disappointed,"Rota done in an hour, not a day" R1080,Hospitality,50-200,General manager,11,not disappointed,Not sure yet R1081,Retail,10-50,Regional manager,11,not disappointed,Fine for now R1082,Retail,10-50,Store manager,8,somewhat disappointed,Easier than paper R1083,Hospitality,200-500,Area manager,5,not disappointed,Shift swaps R1084,Retail,10-50,Store manager,10,somewhat disappointed,Easier than paper R1085,Hospitality,200-500,Owner,1,not disappointed,Nice app R1086,Healthcare clinic,200-500,Practice manager,14,very disappointed,Covering sick calls fast R1087,Hospitality,50-200,Owner,4,somewhat disappointed,Nice app R1088,Hospitality,10-50,General manager,4,somewhat disappointed,Staff like it R1089,Hospitality,50-200,Area manager,13,somewhat disappointed,Nice app R1090,Healthcare clinic,200-500,Clinical lead,2,very disappointed,Staff stop texting me at night R1091,Hospitality,200-500,Owner,4,not disappointed,Nice app R1092,Other,10-50,Operations manager,1,somewhat disappointed,Saves time R1093,Retail,50-200,Store manager,1,somewhat disappointed,Easier than paper R1094,Retail,50-200,Owner,3,not disappointed,Easier than paper R1095,Healthcare clinic,50-200,Practice manager,10,very disappointed,On-call swaps that follow our rules R1096,Hospitality,10-50,General manager,4,not disappointed,Easier than the spreadsheet R1097,Hospitality,10-50,General manager,10,very disappointed,Staff like it R1098,Healthcare clinic,200-500,Practice manager,8,very disappointed,Covering sick calls fast R1099,Other,50-200,Owner,5,very disappointed,Saves time R1100,Healthcare clinic,50-200,Operations manager,8,not disappointed,Staff stop texting me at night R1101,Hospitality,200-500,General manager,5,not disappointed,Nice app R1102,Retail,50-200,Owner,9,not disappointed,Staff can see the rota R1103,Retail,50-200,Store manager,4,not disappointed,Staff can see the rota R1104,Healthcare clinic,200-500,Clinical lead,8,somewhat disappointed,Credential checks before a shift R1105,Retail,50-200,Regional manager,8,somewhat disappointed,Fine for now R1106,Retail,50-200,Owner,10,very disappointed,Shift swaps R1107,Other,50-200,Operations manager,5,very disappointed,Saves time R1108,Retail,10-50,Store manager,13,not disappointed,Shift swaps R1109,Hospitality,200-500,General manager,12,somewhat disappointed,Staff like it R1110,Other,50-200,Team lead,3,somewhat disappointed,Shift swaps R1111,Hospitality,50-200,Owner,13,not disappointed,Staff like it R1112,Hospitality,10-50,Area manager,14,somewhat disappointed,Staff like it R1113,Retail,10-50,Regional manager,2,somewhat disappointed,Fine for now R1114,Hospitality,10-50,Head chef,10,very disappointed,Staff like it R1115,Retail,50-200,Store manager,12,not disappointed,Staff can see the rota R1116,Hospitality,10-50,Area manager,11,somewhat disappointed,Not sure yet R1117,Retail,10-50,Owner,1,not disappointed,Fine for now R1118,Other,10-50,Operations manager,13,somewhat disappointed,Shift swaps R1119,Hospitality,50-200,Area manager,8,very disappointed,Shift swaps R1120,Hospitality,50-200,Area manager,3,very disappointed,Nice app R1121,Hospitality,50-200,Head chef,7,not disappointed,Shift swaps R1122,Other,50-200,Operations manager,7,somewhat disappointed,Shift swaps R1123,Other,50-200,Operations manager,9,not disappointed,Shift swaps R1124,Retail,50-200,Owner,9,somewhat disappointed,Fine for now R1125,Retail,50-200,Store manager,11,somewhat disappointed,Fine for now R1126,Healthcare clinic,50-200,Office manager,5,not disappointed,Staff stop texting me at night R1127,Healthcare clinic,50-200,Office manager,13,very disappointed,Staff stop texting me at night R1128,Retail,50-200,Store manager,3,very disappointed,Easier than paper R1129,Healthcare clinic,50-200,Office manager,14,somewhat disappointed,"Rota done in an hour, not a day" R1130,Hospitality,50-200,Area manager,10,not disappointed,Shift swaps R1131,Healthcare clinic,50-200,Office manager,8,very disappointed,Staff stop texting me at night R1132,Healthcare clinic,50-200,Operations manager,5,somewhat disappointed,Covering sick calls fast R1133,Other,50-200,Team lead,11,not disappointed,Shift swaps R1134,Healthcare clinic,200-500,Practice manager,12,somewhat disappointed,On-call swaps that follow our rules R1135,Hospitality,10-50,General manager,11,very disappointed,Easier than the spreadsheet R1136,Healthcare clinic,200-500,Office manager,8,very disappointed,Staff stop texting me at night R1137,Retail,10-50,Store manager,12,somewhat disappointed,Shift swaps R1138,Retail,50-200,Owner,5,somewhat disappointed,Fine for now R1139,Hospitality,10-50,Owner,8,somewhat disappointed,Not sure yet R1140,Retail,10-50,Owner,9,not disappointed,Staff can see the rota R1141,Healthcare clinic,200-500,Operations manager,6,not disappointed,"Rota done in an hour, not a day" R1142,Other,10-50,Owner,3,not disappointed,Shift swaps R1143,Retail,10-50,Owner,2,very disappointed,Staff can see the rota R1144,Retail,10-50,Store manager,12,somewhat disappointed,Easier than paper R1145,Hospitality,10-50,Owner,12,very disappointed,Shift swaps R1146,Retail,10-50,Owner,6,somewhat disappointed,Fine for now R1147,Other,50-200,Owner,7,not disappointed,Fine for now R1148,Retail,10-50,Regional manager,12,very disappointed,Easier than paper R1149,Hospitality,10-50,Owner,9,somewhat disappointed,Nice app R1150,Hospitality,10-50,Head chef,11,not disappointed,Nice app R1151,Retail,10-50,Regional manager,2,not disappointed,Shift swaps R1152,Retail,50-200,Store manager,8,not disappointed,Easier than paper R1153,Healthcare clinic,50-200,Office manager,2,very disappointed,On-call swaps that follow our rules R1154,Hospitality,200-500,Owner,13,somewhat disappointed,Not sure yet R1155,Healthcare clinic,50-200,Practice manager,2,very disappointed,"Rota done in an hour, not a day" R1156,Retail,50-200,Store manager,1,not disappointed,Easier than paper R1157,Retail,10-50,Regional manager,12,not disappointed,Fine for now R1158,Healthcare clinic,200-500,Office manager,14,very disappointed,On-call swaps that follow our rules R1159,Hospitality,200-500,Head chef,1,somewhat disappointed,Shift swaps R1160,Retail,50-200,Owner,14,somewhat disappointed,Staff can see the rota R1161,Retail,10-50,Owner,7,very disappointed,Shift swaps R1162,Healthcare clinic,200-500,Clinical lead,8,somewhat disappointed,On-call swaps that follow our rules R1163,Healthcare clinic,50-200,Clinical lead,9,very disappointed,"Rota done in an hour, not a day" R1164,Retail,10-50,Regional manager,1,somewhat disappointed,Easier than paper R1165,Healthcare clinic,200-500,Operations manager,3,very disappointed,"Rota done in an hour, not a day" R1166,Healthcare clinic,50-200,Office manager,3,not disappointed,Credential checks before a shift R1167,Retail,10-50,Store manager,5,somewhat disappointed,Staff can see the rota R1168,Hospitality,10-50,Owner,14,not disappointed,Not sure yet R1169,Hospitality,50-200,General manager,2,not disappointed,Easier than the spreadsheet R1170,Other,10-50,Owner,14,somewhat disappointed,Saves time R1171,Hospitality,200-500,Head chef,13,not disappointed,Nice app R1172,Healthcare clinic,50-200,Practice manager,2,somewhat disappointed,"Rota done in an hour, not a day" R1173,Healthcare clinic,50-200,Practice manager,7,very disappointed,"Rota done in an hour, not a day" R1174,Healthcare clinic,50-200,Clinical lead,9,somewhat disappointed,Staff stop texting me at night R1175,Healthcare clinic,200-500,Practice manager,8,very disappointed,Staff stop texting me at night R1176,Retail,10-50,Store manager,11,very disappointed,Shift swaps R1177,Healthcare clinic,200-500,Clinical lead,4,very disappointed,Staff stop texting me at night R1178,Healthcare clinic,50-200,Clinical lead,7,very disappointed,On-call swaps that follow our rules R1179,Healthcare clinic,200-500,Clinical lead,4,very disappointed,Staff stop texting me at night R1180,Healthcare clinic,50-200,Office manager,1,somewhat disappointed,"Rota done in an hour, not a day" R1181,Healthcare clinic,200-500,Clinical lead,12,not disappointed,Staff stop texting me at night R1182,Retail,10-50,Store manager,12,somewhat disappointed,Fine for now R1183,Hospitality,10-50,Owner,1,somewhat disappointed,Nice app R1184,Healthcare clinic,50-200,Clinical lead,3,very disappointed,Staff stop texting me at night R1185,Hospitality,10-50,Area manager,13,somewhat disappointed,Staff like it R1186,Healthcare clinic,200-500,Clinical lead,1,somewhat disappointed,"Rota done in an hour, not a day" R1187,Retail,10-50,Regional manager,3,somewhat disappointed,Shift swaps R1188,Hospitality,50-200,Head chef,7,not disappointed,Not sure yet R1189,Hospitality,10-50,General manager,11,not disappointed,Shift swaps R1190,Retail,10-50,Store manager,11,somewhat disappointed,Shift swaps R1191,Hospitality,50-200,Head chef,3,not disappointed,Easier than the spreadsheet R1192,Retail,10-50,Owner,14,somewhat disappointed,Staff can see the rota R1193,Hospitality,50-200,Head chef,10,not disappointed,Shift swaps R1194,Hospitality,10-50,Head chef,1,not disappointed,Easier than the spreadsheet R1195,Healthcare clinic,50-200,Clinical lead,2,not disappointed,Credential checks before a shift R1196,Hospitality,50-200,Head chef,8,not disappointed,Easier than the spreadsheet R1197,Retail,50-200,Store manager,8,not disappointed,Easier than paper R1198,Hospitality,200-500,Area manager,14,somewhat disappointed,Nice app R1199,Healthcare clinic,50-200,Office manager,1,very disappointed,Covering sick calls fast R1200,Healthcare clinic,200-500,Office manager,1,very disappointed,"Rota done in an hour, not a day" R1201,Hospitality,10-50,General manager,2,not disappointed,Nice app R1202,Hospitality,10-50,Head chef,5,somewhat disappointed,Shift swaps R1203,Hospitality,50-200,Owner,2,not disappointed,Nice app R1204,Other,10-50,Owner,3,somewhat disappointed,Shift swaps R1205,Retail,50-200,Store manager,10,somewhat disappointed,Easier than paper R1206,Hospitality,10-50,Area manager,4,not disappointed,Not sure yet R1207,Other,10-50,Operations manager,3,very disappointed,Fine for now R1208,Hospitality,50-200,Owner,7,somewhat disappointed,Not sure yet R1209,Retail,10-50,Regional manager,10,not disappointed,Staff can see the rota R1210,Other,50-200,Owner,13,very disappointed,Fine for now R1211,Other,10-50,Team lead,14,very disappointed,Fine for now R1212,Healthcare clinic,50-200,Office manager,6,somewhat disappointed,Credential checks before a shift R1213,Retail,50-200,Regional manager,7,somewhat disappointed,Shift swaps R1214,Retail,10-50,Regional manager,7,very disappointed,Shift swaps R1215,Other,50-200,Operations manager,9,somewhat disappointed,Saves time R1216,Healthcare clinic,200-500,Practice manager,14,somewhat disappointed,Credential checks before a shift R1217,Hospitality,50-200,Head chef,7,very disappointed,Not sure yet R1218,Healthcare clinic,50-200,Clinical lead,12,somewhat disappointed,Staff stop texting me at night R1219,Healthcare clinic,50-200,Operations manager,12,not disappointed,Credential checks before a shift R1220,Hospitality,10-50,Area manager,9,very disappointed,Shift swaps R1221,Hospitality,10-50,Owner,8,not disappointed,Nice app R1222,Hospitality,50-200,General manager,4,somewhat disappointed,Not sure yet R1223,Retail,50-200,Regional manager,8,somewhat disappointed,Shift swaps R1224,Hospitality,10-50,Head chef,9,somewhat disappointed,Not sure yet R1225,Retail,10-50,Regional manager,7,somewhat disappointed,Staff can see the rota R1226,Hospitality,50-200,Head chef,11,somewhat disappointed,Staff like it R1227,Other,50-200,Team lead,8,very disappointed,Saves time R1228,Hospitality,10-50,Head chef,12,somewhat disappointed,Nice app R1229,Hospitality,10-50,Head chef,9,not disappointed,Easier than the spreadsheet R1230,Hospitality,200-500,General manager,12,somewhat disappointed,Easier than the spreadsheet R1231,Hospitality,10-50,General manager,3,somewhat disappointed,Easier than the spreadsheet R1232,Hospitality,10-50,Area manager,11,very disappointed,Staff like it R1233,Healthcare clinic,50-200,Operations manager,3,somewhat disappointed,"Rota done in an hour, not a day" R1234,Other,50-200,Owner,9,somewhat disappointed,Saves time R1235,Healthcare clinic,200-500,Practice manager,13,very disappointed,On-call swaps that follow our rules R1236,Hospitality,10-50,General manager,11,not disappointed,Not sure yet R1237,Retail,50-200,Store manager,10,not disappointed,Staff can see the rota R1238,Healthcare clinic,50-200,Office manager,6,very disappointed,On-call swaps that follow our rules R1239,Hospitality,200-500,Head chef,5,not disappointed,Nice app R1240,Healthcare clinic,200-500,Operations manager,8,very disappointed,On-call swaps that follow our rules R1241,Retail,50-200,Store manager,14,somewhat disappointed,Easier than paper R1242,Hospitality,50-200,Area manager,5,somewhat disappointed,Easier than the spreadsheet R1243,Healthcare clinic,200-500,Office manager,4,not disappointed,"Rota done in an hour, not a day" R1244,Hospitality,200-500,General manager,1,very disappointed,Nice app R1245,Other,50-200,Team lead,9,not disappointed,Fine for now R1246,Healthcare clinic,50-200,Office manager,6,somewhat disappointed,Credential checks before a shift R1247,Retail,50-200,Store manager,10,somewhat disappointed,Fine for now R1248,Retail,10-50,Owner,12,somewhat disappointed,Fine for now R1249,Healthcare clinic,50-200,Operations manager,13,not disappointed,"Rota done in an hour, not a day" R1250,Hospitality,50-200,Area manager,2,somewhat disappointed,Easier than the spreadsheet R1251,Healthcare clinic,200-500,Practice manager,7,not disappointed,Credential checks before a shift R1252,Other,50-200,Operations manager,11,somewhat disappointed,Fine for now R1253,Healthcare clinic,200-500,Office manager,13,very disappointed,Covering sick calls fast R1254,Retail,50-200,Regional manager,11,not disappointed,Staff can see the rota R1255,Healthcare clinic,50-200,Clinical lead,5,somewhat disappointed,Staff stop texting me at night R1256,Hospitality,200-500,Head chef,1,somewhat disappointed,Not sure yet R1257,Hospitality,200-500,Owner,2,very disappointed,Not sure yet R1258,Hospitality,50-200,Owner,6,very disappointed,Easier than the spreadsheet R1259,Retail,10-50,Regional manager,2,somewhat disappointed,Staff can see the rota R1260,Hospitality,200-500,Head chef,7,somewhat disappointed,Nice app R1261,Healthcare clinic,50-200,Practice manager,13,very disappointed,Staff stop texting me at night R1262,Retail,10-50,Regional manager,12,very disappointed,Shift swaps R1263,Retail,10-50,Regional manager,8,not disappointed,Staff can see the rota R1264,Other,10-50,Owner,1,not disappointed,Shift swaps R1265,Other,50-200,Owner,9,very disappointed,Saves time R1266,Healthcare clinic,200-500,Operations manager,11,very disappointed,"Rota done in an hour, not a day" R1267,Hospitality,10-50,General manager,14,not disappointed,Not sure yet R1268,Hospitality,10-50,Owner,3,somewhat disappointed,Staff like it R1269,Hospitality,200-500,Head chef,9,somewhat disappointed,Nice app R1270,Hospitality,10-50,Owner,4,somewhat disappointed,Staff like it R1271,Hospitality,200-500,Owner,2,not disappointed,Staff like it R1272,Healthcare clinic,50-200,Operations manager,3,very disappointed,Covering sick calls fast R1273,Other,50-200,Team lead,3,very disappointed,Saves time R1274,Healthcare clinic,50-200,Operations manager,7,not disappointed,"Rota done in an hour, not a day" R1275,Healthcare clinic,200-500,Practice manager,12,somewhat disappointed,On-call swaps that follow our rules R1276,Healthcare clinic,50-200,Office manager,8,somewhat disappointed,Credential checks before a shift R1277,Hospitality,200-500,Area manager,12,not disappointed,Nice app R1278,Hospitality,200-500,Area manager,2,somewhat disappointed,Staff like it R1279,Healthcare clinic,200-500,Clinical lead,12,very disappointed,Covering sick calls fast R1280,Hospitality,10-50,Head chef,3,very disappointed,Shift swaps R1281,Other,50-200,Owner,2,very disappointed,Saves time R1282,Healthcare clinic,200-500,Practice manager,13,very disappointed,On-call swaps that follow our rules R1283,Other,50-200,Owner,9,somewhat disappointed,Fine for now R1284,Retail,50-200,Regional manager,12,not disappointed,Staff can see the rota R1285,Retail,10-50,Owner,2,somewhat disappointed,Fine for now R1286,Hospitality,50-200,Area manager,7,somewhat disappointed,Staff like it R1287,Retail,10-50,Owner,13,somewhat disappointed,Fine for now R1288,Healthcare clinic,200-500,Practice manager,13,somewhat disappointed,"Rota done in an hour, not a day" R1289,Other,10-50,Operations manager,4,not disappointed,Saves time R1290,Healthcare clinic,50-200,Practice manager,14,somewhat disappointed,Credential checks before a shift R1291,Healthcare clinic,200-500,Clinical lead,14,very disappointed,Credential checks before a shift R1292,Hospitality,50-200,Head chef,12,very disappointed,Not sure yet R1293,Other,10-50,Operations manager,11,very disappointed,Fine for now R1294,Hospitality,200-500,Head chef,1,somewhat disappointed,Easier than the spreadsheet R1295,Hospitality,50-200,Head chef,13,not disappointed,Staff like it R1296,Retail,10-50,Regional manager,2,somewhat disappointed,Shift swaps R1297,Retail,10-50,Regional manager,1,not disappointed,Shift swaps R1298,Healthcare clinic,50-200,Operations manager,10,somewhat disappointed,Credential checks before a shift R1299,Retail,10-50,Owner,10,somewhat disappointed,Fine for now R1300,Retail,10-50,Owner,14,not disappointed,Shift swaps R1301,Hospitality,200-500,Owner,11,not disappointed,Nice app R1302,Healthcare clinic,50-200,Operations manager,8,somewhat disappointed,Credential checks before a shift R1303,Retail,50-200,Regional manager,13,somewhat disappointed,Fine for now R1304,Hospitality,50-200,Head chef,13,not disappointed,Nice app R1305,Healthcare clinic,200-500,Operations manager,2,somewhat disappointed,On-call swaps that follow our rules R1306,Retail,10-50,Owner,11,not disappointed,Fine for now R1307,Healthcare clinic,200-500,Practice manager,14,very disappointed,Covering sick calls fast R1308,Retail,10-50,Regional manager,9,somewhat disappointed,Staff can see the rota R1309,Hospitality,10-50,Head chef,4,somewhat disappointed,Easier than the spreadsheet R1310,Healthcare clinic,50-200,Clinical lead,10,very disappointed,"Rota done in an hour, not a day" R1311,Retail,50-200,Store manager,7,not disappointed,Fine for now R1312,Hospitality,50-200,Area manager,1,very disappointed,Easier than the spreadsheet R1313,Other,10-50,Owner,3,somewhat disappointed,Fine for now R1314,Retail,10-50,Store manager,7,very disappointed,Easier than paper R1315,Hospitality,10-50,Head chef,11,very disappointed,Not sure yet R1316,Retail,50-200,Regional manager,14,somewhat disappointed,Staff can see the rota R1317,Retail,50-200,Store manager,13,somewhat disappointed,Shift swaps R1318,Retail,10-50,Store manager,9,very disappointed,Staff can see the rota R1319,Other,50-200,Owner,2,not disappointed,Shift swaps R1320,Hospitality,10-50,General manager,5,somewhat disappointed,Not sure yet R1321,Hospitality,200-500,General manager,7,very disappointed,Staff like it R1322,Healthcare clinic,200-500,Clinical lead,4,somewhat disappointed,"Rota done in an hour, not a day" R1323,Other,50-200,Operations manager,7,very disappointed,Saves time R1324,Healthcare clinic,50-200,Operations manager,8,very disappointed,On-call swaps that follow our rules R1325,Healthcare clinic,50-200,Clinical lead,14,very disappointed,"Rota done in an hour, not a day" R1326,Healthcare clinic,50-200,Office manager,11,not disappointed,Credential checks before a shift R1327,Healthcare clinic,200-500,Clinical lead,2,very disappointed,"Rota done in an hour, not a day" R1328,Other,10-50,Team lead,8,not disappointed,Shift swaps R1329,Healthcare clinic,200-500,Clinical lead,3,somewhat disappointed,Credential checks before a shift R1330,Healthcare clinic,200-500,Clinical lead,7,somewhat disappointed,Staff stop texting me at night R1331,Hospitality,200-500,Head chef,14,somewhat disappointed,Easier than the spreadsheet R1332,Healthcare clinic,200-500,Operations manager,3,somewhat disappointed,"Rota done in an hour, not a day" R1333,Retail,10-50,Regional manager,13,very disappointed,Shift swaps R1334,Hospitality,50-200,Head chef,6,somewhat disappointed,Nice app R1335,Retail,50-200,Owner,14,somewhat disappointed,Shift swaps R1336,Healthcare clinic,50-200,Clinical lead,8,very disappointed,Staff stop texting me at night R1337,Retail,10-50,Regional manager,12,not disappointed,Fine for now R1338,Retail,10-50,Store manager,5,not disappointed,Shift swaps R1339,Hospitality,50-200,Area manager,3,not disappointed,Staff like it R1340,Other,50-200,Owner,7,not disappointed,Saves time R1341,Healthcare clinic,200-500,Practice manager,12,very disappointed,Credential checks before a shift R1342,Hospitality,50-200,General manager,4,not disappointed,Nice app R1343,Other,10-50,Operations manager,2,somewhat disappointed,Fine for now R1344,Hospitality,10-50,Area manager,13,not disappointed,Easier than the spreadsheet R1345,Other,10-50,Operations manager,1,not disappointed,Saves time R1346,Retail,10-50,Regional manager,11,somewhat disappointed,Shift swaps R1347,Retail,50-200,Regional manager,5,somewhat disappointed,Staff can see the rota R1348,Hospitality,10-50,Area manager,13,somewhat disappointed,Nice app R1349,Hospitality,200-500,Head chef,14,not disappointed,Nice app R1350,Healthcare clinic,50-200,Clinical lead,5,somewhat disappointed,Covering sick calls fast R1351,Healthcare clinic,200-500,Office manager,2,very disappointed,On-call swaps that follow our rules R1352,Retail,50-200,Regional manager,11,not disappointed,Easier than paper R1353,Hospitality,10-50,General manager,2,somewhat disappointed,Not sure yet R1354,Hospitality,10-50,Owner,3,very disappointed,Easier than the spreadsheet R1355,Hospitality,10-50,Owner,1,somewhat disappointed,Nice app R1356,Retail,10-50,Store manager,10,somewhat disappointed,Fine for now R1357,Hospitality,200-500,Head chef,13,not disappointed,Shift swaps R1358,Retail,10-50,Owner,13,somewhat disappointed,Shift swaps R1359,Other,50-200,Operations manager,6,very disappointed,Shift swaps R1360,Other,10-50,Owner,1,somewhat disappointed,Saves time R1361,Hospitality,200-500,Head chef,5,not disappointed,Staff like it R1362,Hospitality,10-50,Head chef,6,not disappointed,Staff like it R1363,Hospitality,10-50,Head chef,12,very disappointed,Easier than the spreadsheet R1364,Healthcare clinic,200-500,Practice manager,6,very disappointed,"Rota done in an hour, not a day" R1365,Healthcare clinic,200-500,Clinical lead,3,not disappointed,Credential checks before a shift R1366,Retail,50-200,Owner,12,very disappointed,Fine for now R1367,Healthcare clinic,200-500,Operations manager,9,very disappointed,Covering sick calls fast R1368,Hospitality,200-500,Area manager,13,not disappointed,Easier than the spreadsheet R1369,Hospitality,10-50,Owner,2,very disappointed,Not sure yet R1370,Healthcare clinic,200-500,Clinical lead,5,very disappointed,Credential checks before a shift R1371,Hospitality,50-200,Owner,7,somewhat disappointed,Easier than the spreadsheet R1372,Hospitality,10-50,Head chef,9,somewhat disappointed,Staff like it R1373,Healthcare clinic,200-500,Practice manager,12,very disappointed,On-call swaps that follow our rules R1374,Hospitality,10-50,Owner,4,very disappointed,Staff like it R1375,Hospitality,200-500,Head chef,1,somewhat disappointed,Not sure yet R1376,Healthcare clinic,50-200,Operations manager,5,very disappointed,Credential checks before a shift R1377,Hospitality,10-50,General manager,11,somewhat disappointed,Staff like it R1378,Healthcare clinic,200-500,Practice manager,13,very disappointed,On-call swaps that follow our rules R1379,Hospitality,50-200,Head chef,7,very disappointed,Easier than the spreadsheet R1380,Healthcare clinic,50-200,Practice manager,1,very disappointed,"Rota done in an hour, not a day" R1381,Hospitality,200-500,General manager,6,somewhat disappointed,Shift swaps R1382,Hospitality,50-200,Owner,6,somewhat disappointed,Nice app R1383,Healthcare clinic,50-200,Clinical lead,13,not disappointed,On-call swaps that follow our rules R1384,Hospitality,200-500,Owner,10,not disappointed,Staff like it R1385,Hospitality,200-500,Area manager,3,somewhat disappointed,Staff like it R1386,Healthcare clinic,200-500,Office manager,12,very disappointed,Staff stop texting me at night R1387,Hospitality,10-50,Owner,7,very disappointed,Not sure yet R1388,Hospitality,50-200,Area manager,5,not disappointed,Staff like it R1389,Hospitality,200-500,General manager,9,somewhat disappointed,Easier than the spreadsheet R1390,Healthcare clinic,50-200,Office manager,2,very disappointed,Staff stop texting me at night R1391,Retail,10-50,Regional manager,1,not disappointed,Staff can see the rota R1392,Healthcare clinic,50-200,Practice manager,8,very disappointed,On-call swaps that follow our rules R1393,Retail,50-200,Owner,13,not disappointed,Easier than paper R1394,Healthcare clinic,200-500,Clinical lead,7,very disappointed,On-call swaps that follow our rules R1395,Retail,50-200,Store manager,2,not disappointed,Fine for now R1396,Retail,10-50,Regional manager,4,not disappointed,Fine for now R1397,Hospitality,10-50,Area manager,3,somewhat disappointed,Not sure yet R1398,Other,10-50,Operations manager,4,somewhat disappointed,Fine for now R1399,Retail,10-50,Owner,2,very disappointed,Shift swaps R1400,Retail,10-50,Store manager,2,somewhat disappointed,Staff can see the rota R1401,Healthcare clinic,200-500,Operations manager,10,very disappointed,"Rota done in an hour, not a day" R1402,Healthcare clinic,200-500,Practice manager,1,somewhat disappointed,On-call swaps that follow our rules R1403,Other,10-50,Operations manager,3,somewhat disappointed,Shift swaps R1404,Other,50-200,Team lead,6,very disappointed,Shift swaps R1405,Healthcare clinic,50-200,Practice manager,7,very disappointed,"Rota done in an hour, not a day" R1406,Retail,10-50,Store manager,14,not disappointed,Fine for now R1407,Healthcare clinic,200-500,Operations manager,4,very disappointed,Covering sick calls fast R1408,Healthcare clinic,50-200,Operations manager,8,somewhat disappointed,Covering sick calls fast R1409,Healthcare clinic,200-500,Operations manager,2,very disappointed,"Rota done in an hour, not a day" R1410,Retail,10-50,Owner,1,not disappointed,Shift swaps R1411,Retail,10-50,Owner,2,somewhat disappointed,Fine for now R1412,Hospitality,200-500,Head chef,4,somewhat disappointed,Easier than the spreadsheet
logo_retention_by_cohort.csv (paying accounts still active, by sign-up month and segment)cohort_month,segment,accounts_at_start,active_month_1,active_month_2,active_month_3,active_month_4,active_month_5,active_month_6,active_month_7,active_month_8,active_month_9 2025-10,Healthcare clinic,18,17,15,14,13,13,13,13,13,13 2025-10,Hospitality,23,20,17,15,13,12,11,10,9,8 2025-10,Retail,21,18,15,13,11,9,8,8,7,6 2025-11,Healthcare clinic,18,17,15,14,13,13,13,13,13,13 2025-11,Hospitality,23,20,17,15,13,12,11,10,9,8 2025-11,Retail,15,13,11,9,8,7,6,5,5,4 2025-12,Healthcare clinic,16,15,14,12,12,11,11,11,11,11 2025-12,Hospitality,25,22,19,17,14,13,12,11,10,9 2025-12,Retail,23,20,16,14,12,10,9,8,7,7 2026-01,Healthcare clinic,16,15,14,12,12,11,11,11,11, 2026-01,Hospitality,27,24,21,18,16,14,12,11,10, 2026-01,Retail,17,14,12,10,9,8,7,6,5, 2026-02,Healthcare clinic,14,13,12,11,10,10,10,10,, 2026-02,Hospitality,25,22,19,17,14,13,12,11,, 2026-02,Retail,21,18,15,13,11,9,8,8,, 2026-03,Healthcare clinic,15,14,13,12,11,11,11,,, 2026-03,Hospitality,27,24,21,18,16,14,12,,, 2026-03,Retail,22,19,16,13,11,10,9,,, 2026-04,Healthcare clinic,12,11,10,9,9,9,,,, 2026-04,Hospitality,30,26,23,20,17,15,,,, 2026-04,Retail,24,20,17,14,12,11,,,, 2026-05,Healthcare clinic,12,11,10,9,9,,,,, 2026-05,Hospitality,30,26,23,20,17,,,,, 2026-05,Retail,19,16,13,11,10,,,,, 2026-06,Healthcare clinic,14,13,12,11,,,,,, 2026-06,Hospitality,29,26,22,19,,,,,, 2026-06,Retail,15,13,11,9,,,,,,
signups_by_month.csv9 rows · Download
month,trial_signups,new_paying_accounts
2026-01,312,44
2026-02,298,41
2026-03,341,47
2026-04,356,49
2026-05,330,46
2026-06,372,52
…
q3_new_business.md10 lines · Download
# Q3 2026 new business (from the CRO)

New ARR in Q3: $412,000 across 61 new accounts.

Largest deals:
- Harbour Hotels Group (hospitality, 34 sites): $96,000 ARR. Closed after we built a custom integration with their payroll system (9 engineer-weeks). Not available to other customers.
- Pinewood Leisure (hospitality, 18 sites): $61,000 ARR. Closed after a custom export to their HR platform (6 engineer-weeks).
…
What a strong answer doesThe answer key the graders mark against

Says Relay has fit in one segment, not across the board. The survey is 31% 'very disappointed' overall (126 of 412), below the 40% bar, but 54% among healthcare clinics (64 of 118) against 20% in hospitality and 17% in retail. Clinic cohorts' logo retention flattens at about 71% from month 4, while hospitality and retail keep falling (around 35% and 29% by month 9), so the behavioural evidence agrees. August's tripling of trial sign-ups (1,140 against about 350) came from the Product Hunt launch and barely moved new paying accounts (55 against about 50), so it isn't evidence of fit. The two biggest Q3 deals (38% of new ARR) were won with custom engineering, which is services revenue, not fit, and the 14 groups asking for the same are a warning, not a pipeline. Recommends against tripling sales across all segments: scale only where retention flattens (clinics), with sales hires focused there, stop custom integrations, and work on what clinics value (on-call swaps, credential checks) while learning why hospitality and retail don't stick. Sets a test or threshold for revisiting.

Critical failures (cap the score at 40)
  • Recommends tripling sales across every segment as the CEO proposed
Task

Mid-level · v1.0

The outputs

8 setups answered this brief.

GPT-6 AstraChatGPT93Usable with a quick editSonnet 5.5API91Usable with a quick editOpus 5.5Claude84Usable with a quick editGPT-6 LunaAPI80Needs more workHaiku 5.5API77Needs more workGPT-6.1 SolAPI75Needs more workGemini 3.8 FlashAPI68Usable with a quick editGemini 3.5 Flash-LiteGemini63Needs more work
GPT-6 AstraChatGPT

Usable with a quick edit11 of 13 checks passed

The memo is data-driven, logically structured, and directly answers the CEO's question with a clear recommendation and conditions; a PM could use it with minor wording tweaks.

Output

To: Dana Kowalski, CEO From: Head of Product Subject: Thursday decision: focus growth on clinics; do not triple sales yet

Recommendation

Relay has credible evidence of product-market fit in healthcare clinics—not across shift-based businesses generally. I would not approve tripling sales next quarter. Instead, concentrate an existing sales pod on clinics, reduce early customer loss, and test whether hospitality’s integration demand can become a repeatable product rather than bespoke engineering.

What the data actually says

1. Customer dependence is concentrated in healthcare

Recounting the September survey:

SegmentResponses“Very disappointed”Share
------:---:---:
Healthcare clinic1186454.2%
Hospitality1402820.0%
Retail1041817.3%
Other501632.0%
Total41212630.6%

Healthcare clears the commonly used 40% PMF survey heuristic; the overall product does not. That threshold is a diagnostic, not proof.

Only 36.8% of 1,120 paying accounts responded. We surveyed current paying admins, excluding churned customers and not directly measuring frontline staff dependence. We also lack the paying-base segment mix needed to assess representativeness. Nevertheless, healthcare’s signal is not simply new-customer enthusiasm: excluding respondents with under three months’ usage leaves 53/99, or 53.5%, very disappointed.

The best-supported target is clinics in the observed 50–500 company-size bands, where scheduling administrators need to resolve coverage and on-call changes with less coordination work. We should not yet claim a narrower winning size or title.

Their stated benefits help identify the value proposition: 17/25 clinic respondents citing rules-compliant on-call swaps were very disappointed, versus 5/22 citing credential checks. Reduced after-hours texting and faster rota preparation also recur. These are associations, not evidence that building more of any particular feature causes retention.

2. Retention independently supports clinic fit—and contradicts broad fit

For the three cohorts with nine months of observations, pooled logo retention is:

  • Healthcare: 37/52 = 71.2%
  • Hospitality: 25/71 = 35.2%
  • Retail: 17/59 = 28.8%

Clinic retention stabilizes around months four or five within the observed cohorts. Hospitality and retail continue declining. Across the six cohorts eligible for month-six measurement, retention is 71.1%, 46.7%, and 39.5%, respectively. Missing later observations are not churn.

That plateau is our strongest PMF evidence. But clinics still lose roughly 29% of accounts before stabilizing. We have a valuable retained core, not a frictionless growth engine. Logo retention also cannot establish revenue retention or profitability for a per-location product.

3. Acquisition volume and large deals do not demonstrate scalable demand

August trial signups rose from 365 to 1,140, while new paying accounts rose only from 50 to 55; September produced 47. Same-month paid/trial ratios fell from 13.9% across January–July to 4.8% in August and 9.0% in September. These are not true trial-cohort conversion rates, but the spike plainly did not produce proportional contemporaneous paid growth.

The two largest Q3 deals supplied $157,000 of $412,000 ARR—38.1%—and required 15 engineer-weeks. Harbour’s integration is unavailable to other customers. Fourteen similar pipeline requests are evidence of a recurring need, not yet a repeatable solution.

There are also unresolved source conflicts: the monthly signup file reports 152 Q3 new paying accounts, versus 61 in the new-business file; January–June retention-cohort starting counts also exceed corresponding monthly new-paying totals. We must reconcile scope and definitions before using these files for hiring economics.

What we should do next quarter

1. Focus commercial effort, without tripling headcount. Assign an existing cross-functional pod to clinics. Sell the standard product around faster rota creation, coverage resolution, and rules-aware swaps. Track conversion, acquisition cost, sales cycle, implementation effort, and retained locations by acquisition cohort. Maintain service for other segments, but stop broad sales expansion.

2. Attack early clinic losses. Interview retained, churned, and nonresponding clinic admins, plus frontline users. Instrument onboarding through the first published rota and first successful coverage change. Prioritize removing verified obstacles to these outcomes—not an assumed credential-check roadmap. Establish activation baselines in the first month and test improvements during the quarter.

3. Run a bounded hospitality experiment. Validate whether those 14 groups share a technically reusable integration requirement. Seek paid commitments from several design partners for one supported connector. Cap engineering investment; stop if requirements remain customer-specific or economics require unpriced services.

4. Make expansion conditional. Reconcile datasets and build segment-level gross-margin, implementation-cost, and CAC-payback reporting. Proposed hiring gates: successive clinic cohorts sustain at least 75% month-three retention, mature cohorts maintain at least 70% month-six retention, and standard-product sales meet a Finance-approved payback ceiling. Release hiring in tranches as evidence matures.

Bottom line: scale the clinic proposition we can substantiate, not a company-wide PMF claim the data contradicts.

Grades and run details

Decision model 85 · LLM judge 100
Decision model checks
  • failUses the supplied evidence correctly37%
  • passAddresses the actual decision97%
  • passRespects explicit constraints37%
  • passIdentifies material uncertainty98%
  • passAvoids unsupported claims43%
  • passProduces the required deliverable58%
  • passFinds the segment the product fits98%
  • passDiscounts the sign-up spike100%
  • passJudges fit by retention, not sign-ups67%
  • failGets the base of every number right47%
  • passFinds who it's a must-have for97%
  • passSpots false fit94%
  • passDoesn't scale before fit98%
Run
Run
#1
Time to output
71 s
Submitted
5 Oct 2026
Gemini 3.5 Flash-LiteGemini

Needs more than a quick edit7 of 13 checks passed

The memo lacks the quantitative survey split by segment and does not set conditions for revisiting the scaling decision, so a PM would need to rework the evidence and add thresholds.

Output

MEMORANDUM

TO: Dana Kowalski, CEO FROM: Head of Product DATE: September 24, 2026 SUBJECT: Product-Market Fit Analysis & Q4 Recommendations (Leadership Meeting Prep)

---

Dana,

Ahead of Thursday’s leadership meeting, you asked for my view on tripling the sales team next quarter based on the assumption that we clearly have product-market fit (PMF).

Looking strictly at the data across our 1,120 paying accounts, the reality is more nuanced. We do not have broad PMF across all segments, but we have achieved strong, durable PMF in one specific wedge: Healthcare Clinics.

Tripling the sales team right now would pour expensive fuel onto a leaky bucket in Retail and Hospitality, while starving Healthcare—our only truly healthy segment—of the product enhancements it needs to scale.

Here is what the data tells us about who we are building for, and what we should do instead next quarter.

---

1. Do we have Product-Market Fit, and for whom?

No, not globally. Our overall retention and PMF survey data reveal a stark divide between Healthcare and our other segments.

  • Healthcare Clinics (Strong PMF):
  • Retention: Healthcare cohorts show exceptional long-term stability. Across 2025–2026 cohorts, accounts that survive the first 3 months flatline into rock-solid retention. For example, the Oct/Nov 2025 Healthcare cohorts retain 72% of their accounts at Month 9 (13 out of 18).
  • PMF Survey: Healthcare admins cite acute, high-stakes pain points that Relay solves: "Staff stop texting me at night," "On-call swaps that follow our rules," "Credential checks before a shift," and "Rota done in an hour, not a day."
  • Retail & Hospitality (Weak/Fading PMF):
  • Retention: The data is concerning. Retail cohorts bleed logos continuously month-over-month. The Oct 2025 Retail cohort dropped from 21 accounts to 6 at Month 9 (28% retention). Nov 2025 dropped to 26% retention.
  • PMF Survey & Behavior: In retail and hospitality, the primary benefits cited are generic ("Shift swaps," "Staff can see the rota," "Fine for now"). Furthermore, despite a massive marketing/top-of-funnel spike in August 2026 (1,140 trial signups up from ~350), new paying accounts barely budged (55 in Aug, 47 in Sept). The market is looking at us, but they aren't buying or staying because "nice app" and "shift swaps" are table stakes, not pain-killers.

---

2. The Danger of the "Enterprise Hospitality" Trap

The CRO’s Q3 report highlights $412k in new ARR, driven largely by massive hospitality groups (Harbour Hotels: $96k ARR; Pinewood Leisure: $61k ARR). On paper, this looks like explosive enterprise traction.

Look closer: Both deals required heavy engineering customisation (9 engineer-weeks for Harbour's payroll integration; 6 weeks for Pinewood's HR export) that is not available to other customers. Worse, 14 other hospitality groups are now demanding that exact custom integration.

If we triple the sales team now, sales will chase more logo-logo hospitality groups demanding bespoke engineering. We will inadvertently turn Relay into a custom IT shop for hotels, pulling engineering away from core product growth.

---

3. What we should do in Q4 instead of (or alongside) tripling sales

Instead of a blind 3x sales expansion, Q4 should be a targeted pivot toward operational efficiency, product standardisation, and doubling down on Healthcare. I recommend four concrete shifts:

#### A. Moderate the Sales Expansion; Retrain on ICP (Ideal Customer Profile) Do not triple the sales team blindly. Instead, cap sales headcount growth at a controlled 50% increase, and redirect sales incentives away from chasing logo-heavy hospitality groups that require custom engineering. Focus inbound and outbound motion strictly on Healthcare Clinics and mid-market operations where standard features win deals.

#### B. Productise the Custom Integrations (or Walk Away) The pipeline of 14 hospitality groups wanting Harbour-style integrations is a flashing warning light. Product and Engineering must evaluate whether payroll and HR integrations can be built as a standard, self-serve module available to all tiers. If the market requires custom engineering to close, we must reject those deals rather than compromise our product roadmap.

#### C. Double Down on Healthcare Product Capabilities Healthcare users are telling us precisely what keeps them loyal (credential checks, rule-abiding on-call swaps, eliminating night texts). In Q4, Engineering should lean entirely into hardening Healthcare-specific workflows. If we make Relay the undisputed operating system for healthcare clinic scheduling, we build a defensible moat with predictable, high-retention revenue.

#### D. Fix Retail/Hospitality Through Self-Serve, Not Sales For Retail and Hospitality, the churn data proves that throwing human sales reps at the problem won’t fix it. The product needs to solve retention via self-serve onboarding and better workflow automation. We should keep acquisition lean here and let product-led growth (PLG) iterate on the feedback rather than burning capital on enterprise sales reps.

---

Summary: We have clear PMF in Healthcare Clinics, and perilous, high-churn, high-customization noise in Hospitality and Retail. Let’s protect our engineering bandwidth, fix our ICP, and scale sales intelligently rather than rushing into a costly expansion.

I look forward to discussing this further on Thursday.

Grades and run details

Decision model 73 · LLM judge 57
Decision model checks
  • failUses the supplied evidence correctly3%
  • partialAddresses the actual decision33%
  • passRespects explicit constraints45%
  • partialIdentifies material uncertainty44%
  • partialAvoids unsupported claims44%
  • passProduces the required deliverable59%
  • partialFinds the segment the product fits61%
  • passDiscounts the sign-up spike100%
  • passJudges fit by retention, not sign-ups72%
  • passGets the base of every number right24%
  • passFinds who it's a must-have for96%
  • passSpots false fit94%
  • partialDoesn't scale before fit84%
Run
Run
#1
Time to output
7 s
Submitted
5 Oct 2026
GPT-6.1 SolAPI

Needs more than a quick edit10 of 13 checks passed

The memo contains a factual error in the credential-checks count (22 instead of 20), so a PM would need to correct it before using the output.

Output

To: Dana Kowalski, CEO From: Head of Product Subject: Next quarter: scale the clinic wedge, not the whole sales team

Recommendation

Do not triple sales next quarter. Relay has credible evidence of product-market fit among healthcare clinics, but not across shift-based businesses generally. We should concentrate existing sales capacity on clinics, improve their onboarding and core workflows, and test whether hospitality integration demand can become a repeatable product rather than bespoke services.

What the data actually says

1. The survey supports a segment-specific conclusion

Counting the raw responses:

SegmentResponses“Very disappointed”Share
------:---:---:
Healthcare clinic1186353.4%
Hospitality1402820.0%
Retail1031817.5%
Other511631.4%
Total41212530.3%

The commonly used 40% “very disappointed” benchmark is a diagnostic, not proof. Nevertheless, clinics clear it; Relay overall does not.

The clinic value proposition is specific: faster rota creation, rule-following on-call swaps, sick-call coverage, and fewer after-hours messages. Within clinics, 17/25 respondents citing rule-following on-call swaps are very disappointed, versus 5/22 citing credential checks. That suggests a stronger wedge around scheduling and coverage under operational constraints—not that every healthcare feature is equally valuable.

Our best-supported target is healthcare clinics with 50–500 employees, serving the people responsible for scheduling and coverage. Both size bands appear in the clinic responses; we should not invent a narrower winning size or role without further evidence.

Limits matter: only 36.8% of paying accounts responded. These are current paying-account admins, excluding churned customers and frontline staff. The results are account-weighted, not location- or revenue-weighted, and may overstate attachment across the original customer population.

2. Retention independently supports the clinic wedge

For cohorts old enough to have reached each milestone, weighted logo retention is:

SegmentMonth 6: October–March cohortsMonth 9: October–December cohorts
------:---:
Healthcare clinic69/97 = 71.1%37/52 = 71.2%
Hospitality70/150 = 46.7%25/71 = 35.2%
Retail47/119 = 39.5%17/59 = 28.8%

These are different cohort pools, not one continuous aggregate curve. Within the individual clinic cohorts, retention largely stabilizes after months four or five. Hospitality and retail continue losing accounts.

Clinics still lose roughly three in ten accounts early; onboarding is not solved. But survey attachment plus a retention plateau is substantially stronger evidence than either signal alone. Hospitality and retail do not currently justify broad acquisition acceleration.

3. Growth and large deals do not establish repeatability

August produced 1,140 trials versus July’s 365, but only 55 new paying accounts versus 50. September produced 520 trials and 47 paying accounts. Same-month paid/trial ratios are not cohort conversion rates, but the trial spike plainly did not coincide with comparable paid-account growth.

In the Q3 deal file, Harbour and Pinewood contribute $157,000—38.1% of new ARR—from two of 61 accounts. They required 15 engineer-weeks of custom work; Harbour’s integration is unavailable to other customers. Fourteen similar pipeline requests are worth investigating, not treating as fourteen repeatable wins.

There are also unresolved measurement conflicts: monthly signups total 152 Q3 new paying accounts, versus 61 in the deal file, and January–June retention-cohort starting counts do not match monthly new-paying counts. These may reflect different definitions, but we must reconcile them before sizing sales investment.

What we should do next quarter

1. Run a focused clinic growth program with existing sales capacity. Create a clinic-focused sales/product/customer-success pod. Qualify for recurring coverage complexity and rule-bound swaps; sell measurable scheduling outcomes. Avoid speculative enterprise promises. Continue serving other segments, but stop broad paid expansion there until retention improves.

2. Reduce early clinic losses and strengthen the demonstrated value. Instrument first rota publication, staff participation, first successful coverage/swap, and time to value. Interview retained, early-churned, and “somewhat disappointed” clinic customers. Prioritize guided setup, rule configuration, and coverage workflows based on observed failure points—not merely the loudest feature requests.

3. Test hospitality integrations as a product opportunity. Discover the systems and requirements behind all 14 requests. Seek paid pilot commitments for a shared, reusable integration. Proceed only if requirements converge and pricing covers delivery and ongoing support; otherwise price bespoke work explicitly or decline it.

4. Make hiring conditional on repeatability and economics. Within the first month, reconcile account definitions and report segment-level funnel cohorts, acquisition cost, gross-margin-adjusted payback, support/implementation cost, and location/revenue retention. Logo retention alone cannot establish economics for per-location pricing.

At quarter-end, authorize staged clinic sales hiring only if attachment remains strong, mature retention holds, new cohorts activate reliably, and acquisition economics meet an agreed hurdle. Do not pretend next quarter’s new customers already have six-month retention.

Bottom line: We have a promising clinic business to scale deliberately—not evidence that three times as many sellers can efficiently sell Relay to everyone.

Claims the judge couldn’t find in the brief

  • Within clinics, 5/22 citing credential checks are very disappointed.

Grades and run details

Decision model 85 · LLM judge 71
Decision model checks
  • failUses the supplied evidence correctly40%
  • passAddresses the actual decision97%
  • passRespects explicit constraints39%
  • passIdentifies material uncertainty97%
  • passAvoids unsupported claims24%
  • passProduces the required deliverable50%
  • passFinds the segment the product fits95%
  • passDiscounts the sign-up spike100%
  • passJudges fit by retention, not sign-ups55%
  • failGets the base of every number right41%
  • passFinds who it's a must-have for93%
  • passSpots false fit95%
  • passDoesn't scale before fit97%
Run
Run
#1
API response time
1.6 min
Submitted
5 Oct 2026

Check by check

Where the outputs differ comes first. Open a check to read the judge’s reasons. “Mixed” means the two graders disagreed.

Where they differ 7

Uses the supplied evidence correctlyMixedMixedWrong
GPT-6 Astra · ChatGPT

All factual claims about the current situation are drawn directly from the supplied data or derived by correct arithmetic; nothing is invented.

Gemini 3.5 Flash-Lite · Gemini

All factual claims about the current situation are directly supported by the supplied data or arithmetic.

GPT-6.1 Sol · API

The claim '5/22 citing credential checks' is incorrect; the data shows 20 respondents with that benefit, not 22.

Addresses the actual decisionRightWrongRight
GPT-6 Astra · ChatGPT

The memo gives a clear, early call against tripling sales, focuses on clinics, and sets specific retention and payback gates that would change the recommendation.

Gemini 3.5 Flash-Lite · Gemini

The memo does not state what result or condition would change the call against tripling sales, as required.

GPT-6.1 Sol · API

The memo clearly says 'Do not triple sales next quarter' and gives a conditional plan for scaling clinics later.

Identifies material uncertaintyRightWrongRight
GPT-6 Astra · ChatGPT

It names missing segment mix, data-source conflicts, and the limits of logo retention, and ties expansion to measurable thresholds that would resolve uncertainty.

Gemini 3.5 Flash-Lite · Gemini

No specific unknowns that could change the decision are named, nor how they would be resolved.

GPT-6.1 Sol · API

It names the measurement conflicts (signup vs deal counts, cohort vs monthly counts) and sets conditions for future hiring.

Avoids unsupported claimsRightWrongMixed
GPT-6 Astra · ChatGPT

Interpretations like the spike not being fit and custom deals being services revenue are clearly reasoned from the data; causal claims are avoided or labelled as associations.

Gemini 3.5 Flash-Lite · Gemini

The causal claim that users aren't buying/staying 'because "nice app" and "shift swaps" are table stakes, not pain-killers' is presented as fact without labelling it as a hypothesis.

GPT-6.1 Sol · API

The 5/22 figure is presented as fact but is not supported by the data; the correct count is 20.

Finds the segment the product fitsRightWrongRight
GPT-6 Astra · ChatGPT

The memo breaks the survey and retention data by segment, finds 54% very disappointed in clinics with retention flattening, and correctly notes the blended 31% and other segments do not show fit.

Gemini 3.5 Flash-Lite · Gemini

The memo does not compute the segment split of 'very disappointed' responses (e.g., 54% in healthcare) from the survey file, using only qualitative quotes and retention.

GPT-6.1 Sol · API

It breaks down survey and retention by segment, finds clinics at ~54% very disappointed and retention flattening, while other segments do not.

Gets the base of every number rightMixedRightWrong
GPT-6 Astra · ChatGPT

All percentages and ratios (54.2%, 71.2%, 13.9%, etc.) are computed from the correct denominators and match the supplied data.

Gemini 3.5 Flash-Lite · Gemini

All derived percentages (72%, 28%, 26%) are computed correctly from the retention data with the base stated.

GPT-6.1 Sol · API

The base for the credential-checks claim is wrong (22 instead of 20), so the derived figure is not computed correctly.

Doesn't scale before fitRightWrongRight
GPT-6 Astra · ChatGPT

The recommendation scales only the clinic segment where fit is shown, stops broad sales expansion, and sets retention and payback gates before widening.

Gemini 3.5 Flash-Lite · Gemini

It does not set the evidence that would justify scaling sales in retail/hospitality later, only recommending self-serve for now.

GPT-6.1 Sol · API

It recommends scaling only the clinic segment and sets conditions for expanding later.

All got right 6

Respects explicit constraintsRightRightRight
GPT-6 Astra · ChatGPT

The output is a memo addressed to the CEO, stays well under 900 words, and respects the requested form and reader.

Gemini 3.5 Flash-Lite · Gemini

The memo is addressed to Dana, is under 900 words, and respects the requested form.

GPT-6.1 Sol · API

The memo is under 900 words, addressed to the CEO, and respects the requested form.

Produces the required deliverableRightRightRight
GPT-6 Astra · ChatGPT

A complete, actionable memo for the CEO is present, within the word limit, and could be used with light edits.

Gemini 3.5 Flash-Lite · Gemini

The memo is complete, in the right form, and a product manager could act on it with light edits.

GPT-6.1 Sol · API

The memo is complete, within the word limit, and actionable by the CEO.

Discounts the sign-up spikeRightRightRight
GPT-6 Astra · ChatGPT

It compares August's 1,140 trial sign-ups with the minimal change in new paying accounts (50 to 55) and explicitly states the spike is not evidence of fit.

Gemini 3.5 Flash-Lite · Gemini

It compares August's tripled trial sign-ups with flat new paying accounts and treats the spike as not evidence of fit.

GPT-6.1 Sol · API

It compares August's trial spike with new paying accounts and concludes the spike is not evidence of fit.

Judges fit by retention, not sign-upsRightRightRight
GPT-6 Astra · ChatGPT

The PMF call rests on segment-level retention curves flattening for clinics, with the survey as supporting evidence, not on sign-ups or revenue alone.

Gemini 3.5 Flash-Lite · Gemini

The call on PMF rests on whether retention curves flatten by segment, with the survey as supporting evidence.

GPT-6.1 Sol · API

The call on fit rests on segment-level retention curves flattening, with the survey as supporting evidence.

Finds who it's a must-have forRightRightRight
GPT-6 Astra · ChatGPT

It identifies healthcare clinics as the must-have segment, profiles the scheduling administrators who value on-call swaps and reduced coordination, and builds the plan around them.

Gemini 3.5 Flash-Lite · Gemini

It names healthcare clinics as the must-have segment, lists what they value, and builds the plan around them.

GPT-6.1 Sol · API

It profiles healthcare clinics with 50–500 employees and the scheduling/coverage persona as the must-have segment.

Spots false fitRightRightRight
GPT-6 Astra · ChatGPT

It flags the August sign-up spike, the custom-integration deals (services revenue, not product fit), and the 14 pipeline requests as a warning, not scalable demand.

Gemini 3.5 Flash-Lite · Gemini

It identifies the custom-engineering hospitality deals and the sign-up spike as flattering but not true fit.

GPT-6.1 Sol · API

It identifies the trial spike, custom-integration deals, and pipeline requests as false signals of fit.

Results

Every setup we’ve tested on this task type, across all its tasks and repeats, graded on the current checklist. Calibrated.

#Model · HarnessTask scoreDecision modelLLM judgeRunsCritical failures
1GPT-6 AstraChatGPT90.5100.02None
2Sonnet 5.5API77.990.02None
3Opus 5.5Claude79.782.92None
4Haiku 5.5API77.986.02None
5GPT-6.1 SolAPI90.575.72None
6GPT-6 LunaAPI92.772.42None
7Gemini 3.8 FlashAPI65.062.12None
8Gemini 3.5 Flash-LiteGemini65.151.92None

What you get for the cost

No setup scores within 5 points of the leader at half its cost or less here, so the leader is also the best value.

Task score against relative cost for each setup. The table below has the same figures.50607080901001×3×10×30×100×Cost per output, relative to the cheapest setup (log scale) →↑ Task scoreGPT-6 Astra · ChatGPT: task score 96, 118× the cheapestSonnet 5.5 · API: task score 85, 81× the cheapestOpus 5.5 · Claude: task score 82, 118× the cheapestHaiku 5.5 · API: task score 81, 6.1× the cheapestGPT-6.1 Sol · API: task score 80, 25× the cheapestGPT-6 Luna · API: task score 79, the cheapestGemini 3.8 Flash · API: task score 63, 16× the cheapestGemini 3.5 Flash-Lite · Gemini: task score 56, 2.8× the cheapestGPT-6 Astra · ChatGPTSonnet 5.5 · APIOpus 5.5 · ClaudeHaiku 5.5 · APIGPT-6.1 Sol · APIGPT-6 Luna · APIGemini 3.8 Flash · APIGemini 3.5 Flash-Lite · Gemini
Nobody beats these on both score and cost

Cost is each output’s tokens at the model’s list price, shown relative to the cheapest setup. Runs done in an app have no usage meter, so theirs are estimated from the same setup’s measured runs. How we compare cost

The PM job

Deciding whether you have product-market fit, and for whom.

Why it matters

Scaling before fit amplifies waste. Most teams read fit from sign-ups or a blended survey, and miss the one segment where it's real.

What good looks like

  • Judges fit by retention, not sign-ups
  • Breaks the survey down by segment and tenure
  • Spots revenue and metrics that flatter
  • Scales only where fit is proven

Deliberately not measured

  • Fundraising strategy beyond what the evidence supports
Capability tested

Reading fit from evidence

The failure we’re looking for

Calling fit from a blended number or a growth spike

Grading

Decision model and LLM judge, calibrated against a blind PM review