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

Competitor analysis

Can the model find where a competitor can actually hurt you, from win/loss and churn data, rather than list features?

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. 63% were usable with at most a quick edit.

Reliably right

  1. Respects explicit constraints100% pass
    Delivers a competitive analysis memo to Dan within 900 words, addressing all requested topics without proposing a full feature-by-feature match.
    Opus 5.5 · Claude · The competitor isn't who you think
  2. Reads where Crestline is heading100% pass
    Uses Crestline's hiring (payroll, multi-entity, accountant partnerships) to infer direction, flags the partner-programme risk, and adjusts the response accordingly.
    GPT-6 Astra · ChatGPT · Free forever, funded
  3. Produces the required deliverable97% pass
    The memo is complete, in the right form for the CEO, within the word limit, and actionable with light edits.
    Opus 5.5 · Claude · The competitor isn't who you think

Where it slips

  1. Uses the supplied evidence correctly56% pass
    Two factual claims are wrong: the number of unconnected multi-user accounts going to Crestline is 40, not 42, and the number of connected 6-20 cancellations is 5, not 4.
    Sonnet 5.5 · API · Free forever, funded
  2. Gets the base of every number right61% pass
    Derived figures for unconnected multi-user cancellations (42 vs actual 40) and connected 6-20 cancellations (4 vs actual 5) are incorrect.
    Sonnet 5.5 · API · Free forever, funded
  3. Separates what a rival can copy66% pass
    Does not explicitly separate advantages a well-funded competitor could copy from those it couldn't, nor anchor the response in durable advantages with evidence.
    GPT-6 Luna · API · The competitor isn't who you think

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. Competes with what customers really choose

    Does the analysis treat the status quo (spreadsheets, manual work, doing nothing) as a competitor, and draw at least one implication from where competitors are investing rather than only from today's features?

    Passes when Analyses the status quo, strengths included, and gives a direction-of-travel insight that affects the recommendation.

    April DunfordLenny’s Podcast

  2. Separates what a rival can copy

    Does the analysis separate advantages a well-funded competitor could copy (features, price, being first) from ones it couldn't easily (switching costs, networks, relationships), and build the response on the latter?

    Passes when Names which advantages are copyable and which aren't, with evidence, and anchors the response in the durable ones.

    Hamilton HelmerLenny’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 PM for growth at Tallybook. Our CEO, Dan Whitlock, wants a feature-by-feature comparison against Billfront and a plan to match everything they have by Q2 (his note is below). Write him a competitive analysis of no more than 900 words: who we really lose to, where Billfront is a real threat and where it isn't, where it's heading, and what we should do instead of (or as well as) matching it. The files are attached.

What the model was given5 items: About Tallybook, ceo_note.md, deals_last_two_quarters.csv (trials that became deals, won or lost), customers_by_segment.md, billfront_public_info.md
About TallybookInvoicing and payments for freelancers and small businesses. Plans from $9 to $39 a month.
ceo_note.md1 lines · Download
From Dan (CEO): "Billfront is killing us. I want a full feature-by-feature comparison against them and a plan to match everything they have by Q2."
deals_last_two_quarters.csv (trials that became deals, won or lost)deal,segment,outcome,primary_reason,went_to T-900,Small business,won,, T-901,Freelancer,won,, T-902,Freelancer,won,, T-903,Small business,won,, T-904,Small business,lost,Accountant uses something else,Xero invoicing T-905,Small business,won,, T-906,Small business,lost,Didn't get round to switching,No decision T-907,Freelancer,won,, T-908,Freelancer,won,, T-909,Small business,won,, T-910,Freelancer,lost,Price,Billfront T-911,Freelancer,won,, T-912,Small business,lost,Price,Billfront T-913,Freelancer,won,, T-914,Freelancer,won,, T-915,Small business,won,, T-916,Freelancer,won,, T-917,Small business,won,, T-918,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-919,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-920,Freelancer,won,, T-921,Agency (multi-currency),won,, T-922,Small business,won,, T-923,Small business,won,, T-924,Small business,won,, T-925,Freelancer,won,, T-926,Freelancer,won,, T-927,Freelancer,won,, T-928,Small business,won,, T-929,Small business,lost,Didn't get round to switching,No decision T-930,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-931,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-932,Small business,won,, T-933,Small business,won,, T-934,Small business,won,, T-935,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-936,Agency (multi-currency),lost,Didn't get round to switching,No decision T-937,Small business,lost,Didn't get round to switching,No decision T-938,Agency (multi-currency),won,, T-939,Small business,lost,Didn't get round to switching,No decision T-940,Small business,won,, T-941,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-942,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-943,Freelancer,won,, T-944,Small business,won,, T-945,Small business,lost,Accountant uses something else,Xero invoicing T-946,Small business,won,, T-947,Small business,won,, T-948,Small business,won,, T-949,Small business,lost,Didn't get round to switching,No decision T-950,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-951,Freelancer,won,, T-952,Freelancer,won,, T-953,Freelancer,won,, T-954,Agency (multi-currency),won,, T-955,Small business,lost,Didn't get round to switching,No decision T-956,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-957,Freelancer,won,, T-958,Small business,won,, T-959,Freelancer,won,, T-960,Small business,lost,Didn't get round to switching,No decision T-961,Freelancer,won,, T-962,Small business,won,, T-963,Agency (multi-currency),won,, T-964,Small business,won,, T-965,Small business,won,, T-966,Small business,won,, T-967,Small business,lost,Didn't get round to switching,No decision T-968,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-969,Small business,won,, T-970,Freelancer,lost,Price,Billfront T-971,Small business,lost,Didn't get round to switching,No decision T-972,Small business,lost,Accountant uses something else,Xero invoicing T-973,Small business,lost,Accountant uses something else,Xero invoicing T-974,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-975,Small business,lost,Price,Billfront T-976,Freelancer,won,, T-977,Small business,lost,Didn't get round to switching,No decision T-978,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-979,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-980,Freelancer,won,, T-981,Small business,won,, T-982,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-983,Small business,won,, T-984,Small business,lost,Didn't get round to switching,No decision T-985,Freelancer,won,, T-986,Freelancer,won,, T-987,Small business,lost,Didn't get round to switching,No decision T-988,Small business,won,, T-989,Freelancer,won,, T-990,Freelancer,won,, T-991,Freelancer,won,, T-992,Agency (multi-currency),won,, T-993,Freelancer,won,, T-994,Freelancer,won,, T-995,Small business,won,, T-996,Small business,won,, T-997,Small business,won,, T-998,Small business,lost,Didn't get round to switching,No decision T-999,Freelancer,lost,Stayed on spreadsheets / Word templates,No decision T-1000,Agency (multi-currency),won,, T-1001,Small business,lost,Price,Billfront T-1002,Small business,won,, T-1003,Freelancer,won,, T-1004,Agency (multi-currency),lost,Didn't get round to switching,No decision T-1005,Freelancer,won,, T-1006,Small business,won,, T-1007,Freelancer,won,, T-1008,Small business,won,, T-1009,Freelancer,lost,Price,Billfront T-1010,Small business,won,, T-1011,Freelancer,won,, T-1012,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-1013,Freelancer,won,, T-1014,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-1015,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-1016,Freelancer,won,, T-1017,Freelancer,won,, T-1018,Freelancer,won,, T-1019,Small business,lost,Didn't get round to switching,No decision T-1020,Freelancer,won,, T-1021,Small business,won,, T-1022,Agency (multi-currency),won,, T-1023,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-1024,Agency (multi-currency),lost,Needs invoicing in several currencies with live rates,Billfront T-1025,Freelancer,won,, T-1026,Freelancer,won,, T-1027,Freelancer,won,, T-1028,Small business,won,, T-1029,Freelancer,won,,
customers_by_segment.md7 lines · Download
# Tallybook customers and revenue (September)

| Segment | Paying customers | Average revenue per month |
|---|---|---|
| Freelancer | 5,800 | $11 |
| Small business | 3,100 | $24 |
| Agency (multi-currency) | 240 | $46 |
billfront_public_info.md6 lines · Download
# Billfront (public pages, September)

- Pricing: $12/month Starter, $29/month Pro (multi-currency, live FX rates), $59/month Team.
- Homepage: "Invoicing for global teams". Case studies: three agencies, one consultancy.
- Recent launches: multi-currency invoicing (March), client portal (June).
- Job posts: 4 engineers for "global payments", 2 for "agency workflows".
What a strong answer doesThe answer key the graders mark against

Shows from the deals file that the biggest competitor is no decision: of 46 lost deals, 25 ended with the prospect staying on spreadsheets or never switching, 17 went to Billfront and 4 to Xero invoicing. Billfront's wins are concentrated in multi-currency agencies (11 of its 17), on a specific gap (invoicing in several currencies with live rates); elsewhere it won only 6 deals, mostly on price. Agencies are a small segment (240 customers at about $46 a month, about $11,000 MRR, against about $64,000 from freelancers and $74,000 from small businesses), and Billfront is heading further into them (global payments and agency workflow hiring). Recommends against matching everything: fix switching friction for small businesses who 'didn't get round to it' (imports, setup help, accountant tie-ins), make a deliberate call on multi-currency (build it if agencies matter strategically, or cede the segment), and keep the advantages Billfront can't easily copy. Gives a short comparison only where it matters.

Critical failures (cap the score at 40)
  • Recommends matching Billfront feature for feature
Task

Mid-level · v1.0

The outputs

8 setups answered this brief.

Opus 5.5Claude98Usable with a quick editGPT-6 LunaAPI94Usable with a quick editGPT-6 AstraChatGPT94Usable with a quick editSonnet 5.5API94Usable with a quick editGPT-6.1 SolAPI94Usable with a quick editHaiku 5.5API92Usable with a quick editGemini 3.8 FlashAPI71Usable with a quick editGemini 3.5 Flash-LiteGemini35Needs more work
Gemini 3.5 Flash-LiteGemini

Needs more than a quick edit2 of 11 checks passed

The core data analysis is wrong, making the entire competitive assessment unreliable. A PM would need to re-analyze the deals file from scratch.

Output

MEMORANDUM

TO: Dan Whitlock, CEO FROM: Growth PM DATE: October 24, 2023 SUBJECT: Competitive Analysis: Tallybook vs. Billfront & Q2 Growth Strategy

---

Dan,

You noted that "Billfront is killing us" and asked for a feature match by Q2. I’ve dug into our last two quarters of deal data (trials T-900 through T-1029) and analyzed Billfront’s public footprint.

The data tells a surprising story: Billfront is not killing us broadly. In fact, we are winning the vast majority of our trials. However, Billfront is systematically poaching a very specific, high-value segment.

Below is our competitive reality check, a feature comparison, and a smarter Q2 plan than simply copying their roadmap.

---

1. Who We Really Lose To (The Data)

Out of 130 trial outcomes analyzed from the last two quarters, 74 trials were won (57%) and 56 were lost (43%).

When we lose, where do those deals actually go? * No Decision / Inertia (38 lost / 68% of losses): * Stayed on spreadsheets / Word templates: 16 losses. * Didn't get round to switching: 22 losses. * Takeaway: Our biggest enemy isn't a competitor; it's friction and customer apathy. * Xero (4 losses / 7% of losses): Small businesses whose accountants forced them onto an enterprise ledger. * Billfront (14 losses / 25% of losses): * Price: 4 freelancer losses (lost to their $12 entry tier vs our $9–$11 range). * Needs invoicing in several currencies with live rates: 10 agency losses.

The Verdict: Billfront is only a threat in the Multi-Currency Agency segment. They are positioning themselves as a vertical tool for global agencies, while we are trying to be a generalist tool.

---

2. Feature-by-Feature Comparison: Tallybook vs. Billfront

Feature / CapabilityTallybook ($9–$39/mo)Billfront ($12–$59/mo)Where Billfront is a Real ThreatWhere It’s Noise
:---:---:---:---:---
Core InvoicingRobust, clean templatesStandard invoicingNeutralBoth handle basic billing effortlessly.
Pricing & Packaging$9 (Freelancer), $24 (SMB), $46 (Agency ARPU)$12 (Starter), $29 (Pro), $59 (Team)Real Threat: Their $12 tier undercuts our SMB/Freelancer fence for price-sensitive buyers.
Multi-Currency & Live FXBasic manual entry / limited supportNative live FX rates & multi-currencyLethal Threat: 100% of our agency losses cite this exact feature.
Client PortalBasic payment link viewAdvanced portal (invoices, history, downloads)MinorNice-to-have, but didn't cause any documented churn/loss.
Global Payments / TeamsSingle-user focus, standard gatewaysLocalization, team workflows, agency toolsFuture Threat: Their current hiring (4 engineers for global payments, 2 for agency workflows) proves they are doubling down here.

---

3. Where Billfront is Heading

Based on their job postings and recent launches (client portal in June, multi-currency in March), Billfront is moving upmarket into localized, multi-entity agency operations. They are deliberately abandoning the low-end freelancer market to capture high-value agencies paying $59+/month.

If we spend Q2 blindly matching their features (like building a complex client portal or team permissions), we will burn engineering cycles copying a product that is deliberately scaling away from our bread-and-butter.

---

4. Strategic Recommendation: What We Should Do Instead of Matching

Our revenue is anchored by 5,800 Freelancers ($11 ARPU) and 3,100 Small Businesses ($24 ARPU). Our 240 Agencies ($46 ARPU) are nice, but they represent a fraction of our base.

Instead of playing catch-up to Billfront in a market we aren't optimized to win, we should execute a three-part growth plan for Q2:

#### A. Surrender the Multi-Currency Agency Segment (Deliberately) Stop trying to win agencies that demand live FX rates. Building a multi-currency engine requires heavy compliance, banking integrations, and engineering overhead to chase 10 lost deals a quarter. Let Billfront have them.

#### B. Fix the Real Leaks: Inertia and Onboarding (High ROI) Since 68% of our losses are to "No Decision," our biggest growth lever is reducing time-to-first-invoice for people stuck on spreadsheets. * Action: Build a 1-click CSV importer for past invoices and client lists in Q2. If we can convert half of the "spreadsheet/word template" drop-offs, we instantly capture 8+ new accounts a quarter without touching our pricing.

#### C. Protect the SMB Flank with Smart Packaging Billfront is picking off price-sensitive freelancers with their $12 tier. * Action: Introduce a streamlined "Solo" tier at $10/mo with zero friction, and double down on Tallybook's core differentiator: speed and simplicity for domestic small businesses and freelancers who don't care about global FX rates, they just want to get paid fast.

Summary: Billfront isn't killing us—they are just taking the multi-currency headaches off our hands. Let’s focus Q2 engineering power on wiping out spreadsheet inertia rather than chasing Billfront's agency ghost.

Best, Growth PM, Tallybook

Claims the judge couldn’t find in the brief

  • Out of 130 trial outcomes, 74 trials were won (57%) and 56 were lost (43%).
  • No Decision / Inertia accounts for 38 lost deals (68% of losses): 16 stayed on spreadsheets, 22 didn't get round to switching.
  • Billfront took 14 losses (25% of losses): 4 on price, 10 on multi-currency.
  • Tallybook's core invoicing has robust, clean templates.
  • Tallybook's multi-currency support is basic manual entry / limited support.
  • Tallybook's client portal is a basic payment link view.
  • Tallybook has a single-user focus and standard payment gateways.
  • Building a multi-currency engine requires heavy compliance, banking integrations, and engineering overhead.
  • Billfront is deliberately abandoning the low-end freelancer market.
  • Billfront's $12 tier undercuts our SMB/Freelancer fence for price-sensitive buyers.

Grades and run details

Decision model 59 · LLM judge 17
Decision model checks
  • failUses the supplied evidence correctly51%
  • passAddresses the actual decision32%
  • passRespects explicit constraints12%
  • failIdentifies material uncertainty73%
  • failAvoids unsupported claims55%
  • passProduces the required deliverable56%
  • passFinds the real competitor97%
  • passPins down where Billfront wins59%
  • passCompetes with what customers really choose99%
  • failGets the base of every number right59%
  • partialSeparates what a rival can copy67%
Run
Run
#1
Time to output
7 s
Submitted
5 Oct 2026
Haiku 5.5API

Usable with a quick edit10 of 11 checks passed

The output provides a data-backed, actionable memo. The omission of a copyable vs. durable advantage distinction is a small gap that could be fixed with a quick edit without rechecking evidence or redoing reasoning.

Output

To: Dan Whitlock From: PM, Growth Re: Billfront competitive analysis

Bottom line: Billfront is beating us in one segment: multi-currency agencies. Elsewhere we mostly lose to inertia, not to them. Matching Billfront feature by feature would spend Q2 on the wrong problem. I recommend a scoped multi-currency build for agencies plus a push to convert stalled trials.

Who we actually lose to

Over the last two quarters we won 84 of 130 trials (65%) and lost 46:

Loss reasonLossesShare
No decision: "didn't get round to switching"1635%
No decision: stayed on spreadsheets / Word920%
Billfront: multi-currency with live rates1124%
Billfront: price613%
Accountant uses Xero49%

Fifty-four percent of our losses go to doing nothing. Billfront accounts for 37%.

Segment matters more than count. Agencies win only 7 of 20 trials (35%), against 77% for freelancers and 63% for small businesses. Eleven of our 13 agency losses went to Billfront. Agencies are 240 customers paying $46 a month, so weighting lost deals by segment revenue, Billfront-attributed losses come to roughly $610 a month, and about 83% of that is multi-currency agencies.

Where Billfront is a real threat

  • Multi-currency agencies. Their $29 Pro tier sells live FX rates, and we don't match that for this buyer. This is the one feature gap costing us repeatedly.
  • Their direction. The homepage says "Invoicing for global teams," three of their four case studies are agencies, and they launched multi-currency in March. Their job posts show four engineers on global payments and two on agency workflows. They're moving toward the segment where we're weakest.

Where it isn't

  • Freelancers and small businesses. Only 6 of our 46 losses went to Billfront on price, and none cite a missing feature. Most losses in these segments are status quo.
  • Price. Billfront's Starter is $12 against our $9 entry, so their cheapest plan costs more than ours. The price losses are probably at our mid tier against their Pro or Starter. I'd need the deal notes to confirm, but six deals don't justify a broad price cut.
  • Client portal. Billfront launched it in June, and no lost deal mentions it.

What to do instead of, or as well as, matching

  1. Scope live-FX multi-currency invoicing for agencies (Q2 target). Build what agencies actually need, not Billfront's full feature list. Integrating an FX rate feed will likely be faster than building rate infrastructure. Target: lift agency win rate from 35% to 50% or more.
  2. Work the 25 no-decision losses. "Didn't get round to switching" and "stayed on spreadsheets" point to trial follow-up and migration problems. Test a spreadsheet and Word-template importer, a personal check-in from sales on high-value trials around day 7, and a time-limited switch offer. Run these in two-week cycles and measure conversion.
  3. Address the accountant problem (4 losses). Let accountants view and collaborate on invoices, and check our Xero sync coverage.
  4. Don't chase parity. Skip the client portal and other Billfront features unless losses start citing them. Track loss reasons monthly so future roadmap decisions rest on data.

Caveats

  • 130 trials is a small sample, and loss reasons are rep-recorded. The agency figures rest on just 20 deals.
  • Agency ARPU of $46 is above our $39 top plan, so these customers are probably on add-ons or custom pricing. I'll confirm this before we size the multi-currency opportunity.
  • Billfront's pricing and job posts are public signals, not a confirmed roadmap.

Ask: Approve a two-week scoping sprint for agency multi-currency and a six-week trial-conversion test. I'll bring back scope, cost, and expected impact by the end of the sprint.

Grades and run details

Decision model 91 · LLM judge 92
Decision model checks
  • passUses the supplied evidence correctly25%
  • passAddresses the actual decision62%
  • passRespects explicit constraints67%
  • passIdentifies material uncertainty89%
  • partialAvoids unsupported claims27%
  • passProduces the required deliverable69%
  • passFinds the real competitor97%
  • passPins down where Billfront wins86%
  • passCompetes with what customers really choose100%
  • passGets the base of every number right9%
  • partialSeparates what a rival can copy73%
Run
Run
#1
API response time
32 s
Submitted
8 Oct 2026
GPT-6 LunaAPI

Usable with a quick edit10 of 11 checks passed

The memo is clear, evidence-based, and actionable; a PM could use it with minor wording edits without rechecking data or redoing reasoning.

Output

To: Dan Whitlock Subject: Billfront: where we’re losing—and what to do by Q2

Bottom line: Billfront is a real, focused threat in multi-currency agency invoicing, and a smaller price competitor elsewhere. But it is not the main explanation for our losses overall. A plan to match every feature by Q2 would spread effort across threats the deal data doesn’t support.

Who we really lose to: In 130 trial deals from the last two quarters, we won 84 and lost 46. Billfront was named in 17 losses (37%): 11 for multi-currency invoicing with live FX, and six for price. The other 29 losses were to no decision (16), spreadsheets or Word templates (9), and an accountant’s preference for Xero (4). In other words, stalled switching and familiar alternatives together account for more losses than Billfront.

These are recorded reasons from trial deals, not a complete view of the market. But they argue for fixing activation and switching friction alongside competitive product work—not treating every loss as a feature gap.

Where Billfront is a real threat: The signal is concentrated in agencies. Of 20 agency multi-currency deals, 11 were lost to Billfront specifically because prospects needed invoicing in several currencies with live rates. That’s a clear, repeated unmet need. The segment is small today—240 of our 9,140 customers—but its $46 average monthly revenue makes up about 7% of Tallybook MRR. It merits focused investment.

Where it isn’t: Billfront is not demonstrably beating us across freelancers and small businesses on a broad product advantage. Six price losses matter, but our public starting price is lower ($9 vs. Billfront’s $12); we don’t have enough plan-level detail to compare value or packaging. Nor should we assume we lack Billfront’s other features: the materials provided don’t document Tallybook’s feature set beyond invoicing and payments.

AreaWhat we can verify about BillfrontWhat we can verify about Tallybook
Pricing$12 Starter; $29 Pro; $59 TeamPlans from $9–$39; tier details unavailable
Multi-currency / live FXIncluded in Pro; cited in 11 lossesCapability not documented
Client portalLaunched in JuneCapability not documented
Global payments / agency workflowsHiring for both; these are signals of direction, not proof of shipped featuresCapability not documented

Where Billfront is heading: Its “invoicing for global teams” positioning, recent multi-currency and portal launches, and hiring across global payments and agency workflows point toward a broader agency platform. That makes the agency threat more likely to grow. Hiring is a roadmap signal, though—not confirmation those capabilities have shipped.

Recommendation: match the need, not the whole product.

  1. By Q2, validate and deliver a focused multi-currency offer for agencies. Start with discovery among lost and won agency prospects. If confirmed, prioritize multi-currency invoices and transparent, reliable live-rate handling; test packaging and willingness to pay before committing to a larger build. Track agency win rate and conversion, not feature count.
  2. Reduce the larger pool of avoidable losses. Improve trial onboarding, migration help, and prompts that get a first invoice sent. For Xero-driven losses, assess an integration or accountant-friendly workflow before building accounting features. These address observed losses without trying to replace established systems.
  3. Test price and packaging before cutting prices. The six Billfront price losses warrant follow-up on what customers compared and what value they perceived. Our lower entry price alone isn’t resolving those objections.
  4. Gate further parity work on evidence. Revisit portals, global payments, and agency workflows only if customer research, win/loss data, or retention shows they block adoption.

Decision: Approve a targeted agency multi-currency initiative plus switching and activation improvements—not a blanket commitment to match everything Billfront has by Q2.

Grades and run details

Decision model 95 · LLM judge 92
Decision model checks
  • passUses the supplied evidence correctly46%
  • passAddresses the actual decision86%
  • passRespects explicit constraints64%
  • passIdentifies material uncertainty89%
  • passAvoids unsupported claims69%
  • passProduces the required deliverable61%
  • passFinds the real competitor85%
  • passPins down where Billfront wins93%
  • passCompetes with what customers really choose100%
  • passGets the base of every number right45%
  • partialSeparates what a rival can copy77%
Run
Run
#1
API response time
24 s
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 8

Uses the supplied evidence correctlyWrongRightRight
Gemini 3.5 Flash-Lite · Gemini

Key numbers (wins, losses, breakdown by reason) are incorrect, and many feature claims about Tallybook are invented.

Haiku 5.5 · API

All facts and figures used to describe the current situation are drawn directly from the supplied context or arithmetic from that context, with no inventions.

GPT-6 Luna · API

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

Addresses the actual decisionMixedRightRight
Gemini 3.5 Flash-Lite · Gemini

Does not state what result or condition would change the recommendation.

Haiku 5.5 · API

The output unambiguously commits to a clear call (do not match Billfront feature-by-feature; instead build multi-currency for agencies and address no-decision losses) framed for the CEO, with an ask for approval.

GPT-6 Luna · API

Commits to a clear answer (targeted agency multi-currency plus switching improvements, not blanket matching) and states what would change it (further parity work gated on evidence).

Identifies material uncertaintyWrongRightRight
Gemini 3.5 Flash-Lite · Gemini

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

Haiku 5.5 · API

The output names small sample size, rep-recorded reasons, uncertain ARPU composition, and unconfirmed Billfront roadmap, and says it will confirm ARPU before sizing, thus identifying material uncertainties and how they will be resolved.

GPT-6 Luna · API

Names unknowns (lack of plan detail, Tallybook feature set, hiring as signal not shipped feature) and says how to resolve (discovery with agency prospects, gating further work on evidence).

Avoids unsupported claimsWrongRightRight
Gemini 3.5 Flash-Lite · Gemini

Presents interpretations (e.g., 'deliberately abandoning the low-end market') and feature details as fact without support.

Haiku 5.5 · API

The analysis presents hypotheses (e.g., trial follow-up and migration problems) as interpretations, not established facts, and caveats are provided for weaker signals.

GPT-6 Luna · API

Interpretations are labelled as such (e.g., 'argue for', 'not demonstrably beating us'), and confident claims are backed by data.

Produces the required deliverableMixedRightRight
Gemini 3.5 Flash-Lite · Gemini

The analysis contains wrong numbers, so a PM cannot act on it without redoing the data work.

Haiku 5.5 · API

The memo is in the requested form, addressed to Dan, within the word limit, and contains a complete set of recommendations that the CEO could act on with minimal edits.

GPT-6 Luna · API

Delivers a complete competitive analysis memo to the CEO within the word limit, with actionable recommendations.

Finds the real competitorMixedRightRight
Gemini 3.5 Flash-Lite · Gemini

Incorrectly counts no-decision losses as 38 instead of 25, and misstates the percentage.

Haiku 5.5 · API

The analysis counts losses by destination (25 no decision, 17 Billfront) and names no decision as the biggest competitor with the numbers.

GPT-6 Luna · API

Counts losses by destination, shows 16 'didn't get round to switching' and 9 spreadsheets/Word templates, and names no decision as the biggest loss category.

Pins down where Billfront winsMixedRightRight
Gemini 3.5 Flash-Lite · Gemini

Miscounts Billfront losses and misstates the concentration; does not correctly size the segment's revenue against others with accurate numbers.

Haiku 5.5 · API

It shows Billfront's wins are concentrated in multi-currency agencies (11 of 17), and sizes that segment's revenue ($11k MRR) against the larger freelancer and small business segments.

GPT-6 Luna · API

Shows Billfront wins concentrated in multi-currency agencies (11 of 17), and sizes the segment at 240 customers, ~7% of MRR.

Gets the base of every number rightWrongRightRight
Gemini 3.5 Flash-Lite · Gemini

All derived percentages and counts are based on wrong win/loss totals and wrong sub-counts.

Haiku 5.5 · API

All percentages and rates (e.g., 84/130 = 65%, 25/46 = 54%, win rates, revenue calculation) are computed correctly from the supplied data, with bases clear.

GPT-6 Luna · API

All derived figures (130 deals, 84 won, 46 lost, 17 Billfront, 11 multi-currency, 6 price, 29 other, 7% MRR) check out against the data.

All got wrong 1

Separates what a rival can copyWrongWrongWrong
Gemini 3.5 Flash-Lite · Gemini

Does not separate copyable advantages from durable ones; mentions 'speed and simplicity' but doesn't analyze its defensibility.

Haiku 5.5 · API

The analysis does not separate advantages a well-funded competitor could copy (features, price) from durable ones (switching costs, networks, relationships), nor anchor the response in the latter.

GPT-6 Luna · API

Does not explicitly separate advantages a well-funded competitor could copy from those it couldn't, nor anchor the response in durable advantages with evidence.

All got right 2

Respects explicit constraintsRightRightRight
Gemini 3.5 Flash-Lite · Gemini

The output is a memo under 900 words, addresses the CEO, and does not propose a feature-by-feature matching plan.

Haiku 5.5 · API

The output is a competitive analysis under 900 words, addresses the five requested areas, and does not propose a full feature-by-feature matching plan, respecting the explicit prohibitions.

GPT-6 Luna · API

Output is a competitive analysis memo under 900 words, does not propose a full feature-by-feature matching plan, and addresses the CEO as requested.

Competes with what customers really chooseRightRightRight
Gemini 3.5 Flash-Lite · Gemini

Treats spreadsheets and inertia as the main competitor and draws a direction-of-travel insight from Billfront's hiring.

Haiku 5.5 · API

The status quo (spreadsheets, not switching) is treated as a real competitor, and the analysis incorporates Billfront's direction of travel (global payments, agency workflows) to inform the recommendation.

GPT-6 Luna · API

Treats spreadsheets/doing nothing as a competitor, and gives a direction-of-travel insight (Billfront hiring signals broader agency platform, making threat likely to grow).

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
1Opus 5.5Claude86.4100.02None
2GPT-6 AstraChatGPT88.695.82None
3GPT-6.1 SolAPI88.695.82None
4GPT-6 LunaAPI93.283.32None
5Sonnet 5.5API86.483.32None
6Haiku 5.5API84.179.22None
7Gemini 3.8 FlashAPI68.254.22None
8Gemini 3.5 Flash-LiteGemini61.425.02None

What you get for the cost

GPT-6.1 Sol · API scores within a point of Opus 5.5 · Claude at a fifth of the cost.

Task score against relative cost for each setup. The table below has the same figures.304050607080901001×3×10×30×100×Cost per output, relative to the cheapest setup (log scale) →↑ Task scoreOpus 5.5 · Claude: task score 94, 75× the cheapestGPT-6 Astra · ChatGPT: task score 93, 87× the cheapestGPT-6.1 Sol · API: task score 93, 14× the cheapestGPT-6 Luna · API: task score 86, the cheapestSonnet 5.5 · API: task score 83, 44× the cheapestHaiku 5.5 · API: task score 80, 3.0× the cheapestGemini 3.8 Flash · API: task score 60, 12× the cheapestGemini 3.5 Flash-Lite · Gemini: task score 41, 2.9× the cheapestOpus 5.5 · ClaudeGPT-6 Astra · ChatGPTGPT-6.1 Sol · APIGPT-6 Luna · APISonnet 5.5 · APIHaiku 5.5 · APIGemini 3.8 Flash · APIGemini 3.5 Flash-Lite · Gemini
Best valueNobody 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 how to respond to competitors.

Why it matters

Feature grids hide the real competition, which is often the customer's current workaround, and the response that feels urgent is usually to copy everything.

What good looks like

  • Includes the status quo and non-obvious alternatives
  • Finds where losses really go, by segment
  • Separates advantages a rival can copy from ones it can't
  • Reads where competitors are heading

Deliberately not measured

  • Market sizing
Capability tested

Competitive insight

The failure we’re looking for

A feature checklist and a plan to match everything

Grading

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