Tasks / Growth

Write a GTM plan

Can the model pick the customers who value the difference most, position against what they do today, and build a go-to-market that lasts beyond launch day?

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

Positioned against the real alternative100% pass
It names the real alternatives (accountant-run payroll with mistakes, owner spreadsheet with risk) and ties the message to confidence and accuracy that those alternatives lack.
GPT-6 Astra · ChatGPT · Payroll, for the people who fear it
Protects the self-serve engine100% pass
Uses the 70% free-to-paid stat and product-qualified deal data to argue against capping the free plan, and keeps self-serve open.
GPT-6 Astra · ChatGPT · Enterprise, without killing the engine
Does the capacity sums100% pass
Works out deals needed (69-80), qualified opportunities (134), and AE-years (8) with checkable arithmetic.
GPT-6 Astra · ChatGPT · Enterprise, without killing the engine

Where it slips

Avoids unsupported claims47% pass
Presents the incorrect 'five of six no decision' figure as established fact, which is an unsupported claim about the current situation.
Opus 5.5 · Claude · Enterprise, without killing the engine
Uses the supplied evidence correctly52% pass
The claim 'Five of our six losses on existing-user deals ended in no decision' is not supported by the data (4 no-decision, 2 Miro), so a statement about the current situation is invented.
Opus 5.5 · Claude · Enterprise, without killing the engine
Identifies material uncertainty73% pass
The plan does not name specific unknowns that could change the decision, nor does it say how they would be resolved or what would trigger a pivot.
Gemini 3.8 Flash · API · Payroll, for the people who fear it

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.

Positioned against the real alternative
Does the plan name what the target customer does today without this product (including the status quo or doing nothing), and tie the message to the value that alternative lacks?
Passes when Names the real alternative and the differentiated value against it, and picks the segment that cares most about that value.
April DunfordLenny’s Podcast
Sells against a pain the buyer feels
Is the core message built on a pain or risk the target buyer already feels, with evidence from the material, rather than on generic upside like saving time?
Passes when Leads with the pain the evidence shows buyers care about, and uses that evidence to choose the message.
Jeanne DeWitt GrosserLenny’s Podcast
Diagnoses message or channel first
Does the plan work out whether the problem is the message (people reach it but don't convert) or the channels (the right people never see it), and spend accordingly?
Passes when Uses conversion and reach data to say which it is, and puts the effort there.
Emily KramerLenny’s Podcast
Plans beyond launch week
Does the plan say how demand keeps coming after launch, with the ongoing engine and what's measured over months, rather than treating launch day as the whole plan?
Passes when Gives the ongoing motion after launch and the measures that show whether it's working over time.
Arielle 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 · 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 PM for Ledgerly Payroll, a new add-on for our accounting product, launching to existing customers on 3 November. Marketing has drafted the launch plan below. Write the go-to-market plan you'd run instead (or the changes you'd make to theirs), in no more than 900 words: who we go after first, what we say to them, through which channels, and how we'll know it's working over the following three months. The files are attached.

What the model was given5 items: About Ledgerly, marketing_launch_draft.md, customers_by_size_and_payroll_method.csv (penalty notices: share who told us they got a payroll penalty notice last year), beta_headline_test.csv (two versions of the in-app payroll page, shown at random to beta customers), customer_calls.md

About LedgerlyAccounting software for small businesses: invoicing, bank feeds, bookkeeping. 8,000 paying customers. Payroll is a new add-on at $6 per employee per month.

marketing_launch_draft.md7 lines · Download

# Draft launch plan (Head of Marketing)

- Launch day 3 November: press release announcing Ledgerly as "the first Financial Operating System for small business".
- Paid social and search ads to new prospects: $60,000 in November.
- In-app banner to all 8,000 customers: "Run payroll in minutes".
- Webinar in launch week.
- Success measure: press mentions and sign-ups in launch week.

customers_by_size_and_payroll_method.csv (penalty notices: share who told us they got a payroll penalty notice last year)employees,current_payroll,customers,penalty_notices_last_year_pct 1-4,Owner does it in a spreadsheet,3140,4 1-4,Accountant runs it,1020,6 5-25,Owner does it in a spreadsheet,690,19 5-25,Accountant runs it,1480,23 5-25,Another payroll tool,1210,7 26-50,Another payroll tool,460,5

beta_headline_test.csv (two versions of the in-app payroll page, shown at random to beta customers)headline,landing_page_visits,started_setup,ran_first_payroll "Run payroll in minutes, right where you do your books",4210,92,38 Never get a payroll penalty again,4160,268,117

customer_calls.md10 lines · Download

# Customer calls about payroll (excerpts, September)

1. Bakery, 12 staff: "My accountant does payroll. Last spring they filed late and I got a $1,400 penalty. I don't know what I'm paying them for."
2. Plumbing firm, 18 staff: "Time isn't the issue. I'm scared of getting it wrong. The tax office letter was the worst week of my year."
3. Design studio, 3 staff: "It's me and two freelancers. A spreadsheet's fine."
4. Café group, 22 staff: "We use another tool, it's OK. Switching sounds like a nightmare mid-year."
5. Physio clinic, 9 staff: "I'd move tomorrow if it meant one less thing that can go wrong with the tax office."
…

What a strong answer doesThe answer key the graders mark against

Picks customers with 5 to 25 employees whose accountant runs payroll (1,480 customers, 23% had a penalty notice) and those doing it in a spreadsheet (690, 19%), not the smallest businesses (happy with a spreadsheet) or those on another payroll tool (switching mid-year is the barrier). Builds the message on fear of penalties and errors, not time: in the beta, 'Never get a payroll penalty again' got about three times as many setups as 'Run payroll in minutes' (268 against 92 from similar traffic, 6.4% against 2.2%) and more first payrolls (117 against 38), and the calls say the same. Reads this as a message problem, not a channel problem: the existing customer base already sees the page, so fix the message and reach existing customers in-app and through their accountants before spending $60,000 on ads to strangers. Drops the 'Financial Operating System' category claim. Plans beyond launch week: prompts around payroll and tax deadlines, accountant referrals, and monthly measures such as payroll attach in the target segment, first payrolls run, and penalties avoided.

Critical failures (cap the score at 40)

Keeps 'Run payroll in minutes' as the lead message despite the headline test

Task

Mid-level · v1.0

The outputs

8 setups answered this brief.

GPT-6 AstraChatGPT100Usable with a quick edit GPT-6 LunaAPI100Usable with a quick edit Opus 5.5Claude90Usable with a quick edit GPT-6.1 SolAPI87Usable with a quick edit Sonnet 5.5API85Needs more work Haiku 5.5API85Needs more work Gemini 3.8 FlashAPI83Usable with a quick edit Gemini 3.5 Flash-LiteGemini79Usable with a quick edit

Sonnet 5.5API

Needs more than a quick edit10 of 12 checks passed

The output incorrectly states that about half the base is a poor fit, which would require re-checking the data to correct; otherwise the plan is solid but that error undermines trust.

Output

Ledgerly Payroll: go-to-market plan

What I'd change in Marketing's draft

Draft	Problem	Change
$60k paid ads to new prospects	Payroll is an add-on, so a prospect would first have to adopt our accounting product. We have 8,000 customers and haven't yet proven the message with them.	Hold most of the budget until the message converts in-base (see Budget).
Banner to all 8,000	About half the base is a poor fit (details below).	Target about 2,170 customers first.
"Financial Operating System" press release	Customers never use this language. They talk about fear of penalties and the tax office.	Keep the press small and plain.
Success = press mentions and launch-week sign-ups	Neither measures payroll actually being run.	Measure first payroll run and repeat runs.

Who we go after first

Wave 1 (3 Nov): customers with 5–25 employees who run payroll via spreadsheet (690) or an accountant (1,480). That is 2,170 customers, 27% of the base.

They have the pain. 19% (spreadsheet) and 23% (accountant) got a payroll penalty notice last year, versus 4–6% for businesses with 1–4 staff and 5–7% for those already on a payroll tool.
They have the money. At $6 per employee, a 12-person business pays about $72 a month, against the $180 a month one landscaper pays their accountant.
The calls match the data. The bakery, plumber, physio, landscaper and dental practice all describe fear of mistakes, not lack of time.

Not now: - 1–4 employees (4,160 customers): The spreadsheet is "fine" and "free," and only 4–6% were penalized. Leave payroll available in the product but don't push it. - Other-tool users (1,670 customers): Penalty rates are low (5–7%), and switching mid-year "sounds like a nightmare." Revisit at the start of the next tax year with a migration offer.

What we say

The beta test was randomized. "Never get a payroll penalty again" beat "Run payroll in minutes" on setup starts (6.4% vs 2.2% of visits) and on first payroll runs (2.8% vs 0.9%), about 3x on both. Once people started, completion was similar (44% vs 41%), so the headline drives the decision to begin and the product handles the rest.

Core message: "Payroll filed right and on time, so the tax office letter never arrives."

Spreadsheet owners: "A spreadsheet won't tell you when you've got it wrong."
Accountant-run: "Stop paying $180 a month to worry about it." Use this to speak to cost and control without attacking accountants. Offer a way to invite their accountant in as a reviewer.

Accuracy promise: The dental practice said, "If you can promise it's right, I'm in." I'd like to offer a penalty-cover guarantee for errors caused by our software. Legal and Finance must approve it by 24 Oct. If they won't, the copy becomes "filed right and on time, with support to fix problems." We should not promise "never" without backing it. The beta headline did well, but a claim we can't stand behind creates risk.

Channels

Targeted in-app card and email to the Wave 1 list only, using the winning headline. This is the main channel, and it costs almost nothing.
Personal outreach from Customer Success to Wave 1 customers who told us they got a penalty notice (about 470 people by my estimate from the survey rates). Offer a 20-minute guided setup.
Webinar, reframed: "Penalty-proof your payroll" with a live setup walkthrough. Run it in launch week and again in week 4, since the second session catches people who weren't ready the first time.
Press: one plain release ("Ledgerly adds payroll, $6 per employee"), with no category claim. Follow with customer stories once we have them.
Paid: none to new prospects in November (see Budget).

Budget (the $60k)

~$15k: guided setup and migration help (contractor or CS time).
~$10k: webinars, email production, and case-study content.
~$35k: held until the January gate. If Wave 1 works, spend it on search ads for "payroll penalty" intent, aimed at businesses that would adopt accounting and payroll together.

How we'll know it's working (3 months)

Primary metric: paying accounts that have run at least two payrolls. This replaces sign-ups, press mentions and launch-week anything.

Targets (to be recalibrated after week 2, since the beta is my only baseline): - Month 1: at least 5% of Wave 1 has started setup. At least 45% of those run a first payroll, matching beta's 44%. - End of January: about 250 Wave 1 accounts running payroll (~11% of the segment). I'm assuming a 10-person average, which implies roughly $15k MRR. I'll replace that assumption with real headcount data. - Retention: at least 85% of those accounts run payroll in three consecutive cycles.

Guardrails (any breach pauses expansion): - Payroll errors or penalties caused by us: zero tolerance, reviewed weekly. - Support tickets per payroll run. - Churn of the core accounting subscription, which must not rise among adopters.

Measuring real impact: Hold out a random 10% of Wave 1 from all promotion. If adopters in the promoted group don't clearly exceed the holdout, we're only capturing people who would have found payroll anyway.

Decision gates: - Mid-December: If setup-to-first-payroll is below 35%, fix onboarding before pushing further. - Mid-January: If Wave 1 is at or above target and no guardrail is breached, release the held budget and start Wave 2 (other-tool users, ahead of the tax-year switch window). If it's below target, interview non-adopters before spending anything.

Caveats

Penalty rates are self-reported, and the eight calls are anecdotes.
Beta customers may be more enthusiastic than the base.
I haven't seen average headcount per segment, so the revenue estimates are rough.

Claims the judge couldn’t find in the brief

About half the base is a poor fit

Grades and run details

Decision model 100 · LLM judge 77

Decision model checks

passUses the supplied evidence correctly13%
passAddresses the actual decision99%
passRespects explicit constraints32%
passIdentifies material uncertainty99%
passAvoids unsupported claims38%
passProduces the required deliverable75%
passPicks the segment with the pain100%
passUses the headline test100%
passPositioned against the real alternative100%
passSells against a pain the buyer feels100%
passDiagnoses message or channel first100%
passPlans beyond launch week100%

Run

Run: #1
API response time: 37 s
Submitted: 6 Oct 2026

Check by check

Open a check to read the judge’s reasons. “Mixed” means the two graders disagreed.

Mixed 2

Uses the supplied evidence correctlyMixed

The claim 'About half the base is a poor fit' is not supported by the data; the target segment is 27%, not half.

Avoids unsupported claimsMixed

Presents 'About half the base is a poor fit' as fact when the data shows 73% are not in the target, not half.

Got right 10

Addresses the actual decisionRight

Commits to Wave 1 segment, message, channels, and decision gates with clear conditions for changing course.

Respects explicit constraintsRight

Delivers a go-to-market plan under 900 words, addressing the required elements, and respects the reader and form.

Identifies material uncertaintyRight

Names self-reported penalty data, anecdotal calls, beta enthusiasm bias, missing headcount data, and legal approval for guarantee, with resolution steps.

Produces the required deliverableRight

Complete plan with changes, segment, message, channels, measures, budget, and caveats; usable as-is with light edits.

Picks the segment with the painRight

Targets 5-25 employee customers with accountant or spreadsheet payroll, cites penalty rates, and explains why 1-4 and other-tool users are not first.

Uses the headline testRight

Uses the beta headline test results (3x setup lift) to build the message on penalties and accuracy.

Positioned against the real alternativeRight

Names the real alternatives (spreadsheet, accountant, other tool) and ties message to penalty avoidance and accuracy, targeting the segment that cares most.

Sells against a pain the buyer feelsRight

Leads with fear of penalties and tax office letters, supported by call quotes and headline test, not generic time-saving.

Diagnoses message or channel firstRight

Diagnoses message as the problem (beta conversion data) and focuses on in-app/email to existing base, holding paid ads.

Plans beyond launch weekRight

Plans ongoing motion with second webinar, monthly measures, retention, guardrails, holdout group, and Wave 2 decision gate.

Results

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

#	Model · Harness	Decision model	LLM judge	Runs	Critical failures
1	GPT-6 AstrawithChatGPT	100.0	100.0	2	None
2	Opus 5.5withClaude	96.0	85.2	2	None
3	GPT-6 LunawithAPI	94.2	89.3	2	None
4	GPT-6.1 SolwithAPI	93.8	78.0	2	None
5	Sonnet 5.5withAPI	98.1	77.7	2	None
6	Haiku 5.5withAPI	94.2	77.7	2	None
7	Gemini 3.8 FlashwithAPI	73.9	81.9	2	None
8	Gemini 3.5 Flash-LitewithGemini	74.2	63.5	2	None

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.

Nobody beats these on both score and cost

Task score and relative cost by setup
Setup	Task score	Relative cost (cheapest = 1)
GPT-6 Luna · API	90	1.0×
Haiku 5.5 · API	82	3.3×
Gemini 3.5 Flash-Lite · Gemini	68	3.5×
Gemini 3.8 Flash · API	77	11×
GPT-6.1 Sol · API	84	16×
Sonnet 5.5 · API	84	46×
Opus 5.5 · Claude	91	75×
GPT-6 Astra · ChatGPT	100	81×

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

Taking a product or add-on to market with marketing and sales.

Why it matters

Launch plans fail by trying to reach everyone, by selling upside nobody feels, and by treating launch day as the whole plan.

What good looks like

One primary segment: those who value the difference most
Positions against what customers do today
Sells against a pain the buyer feels, with evidence
Diagnoses whether the problem is the message or the channel
Plans beyond launch week

Deliberately not measured

Creative copywriting quality

Capability tested

Go-to-market judgement

The failure we’re looking for

Every channel, every segment, and nothing after launch day

Grading

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