Usable with a quick edit9 of 11 checks passed
The analysis is evidence-based, logically structured, and directly answers the board's question; a product manager could use it with minor wording tweaks and no rework of evidence or reasoning.
Output
Board memo: Crestline’s free plan—targeted threat, not a case for universal free
Recommendation: Do not launch a free plan for everyone. Crestline threatens customers buying standalone bookkeeping, including some larger businesses—not just solos. Our strongest protection is the accounting-firm relationship embedded in Harbourline, not account size or a feature lead. We should test a tightly bounded free offer, defend unconnected customers across sizes, and deepen our partner advantage before Crestline builds its own.
1. Where Crestline can hurt us—and where it is not yet winning
The threat is real, but headline cancellations overstate incremental damage
The cancellation file totals 250 accounts and $8,065 in lost MRR. Of these:
- 186 accounts, representing $5,581 MRR, went to Crestline.
- 38 returned to spreadsheets/no tool; 26 closed their businesses.
- Crestline-associated losses equal approximately 0.88% of our current $635,400 MRR.
Against the previous run rate of 95 cancellations monthly, 250 over August–September is about 60 additional cancellations—not 186. Destination tells us where customers went, not whether they would otherwise have stayed. The mid-August launch also complicates the comparison.
This is a meaningful warning, not evidence of an existential threat. Equally, cancellations do not capture prospects we failed to acquire or future pricing pressure.
Accounting-firm connection explains exposure better than size
Calculated from `cancellationsaugsep.csv` and `customer_base.md`:
| Account size | Unconnected exits to Crestline / estimated unconnected base | Connected exits to Crestline / estimated connected base | MRR lost to Crestline |
|---|---|---|---|
| --- | ---: | ---: | ---: |
| 1 user | 142 / 10,374 (1.37%) | 0 / 1,026 (0%) | $1,958 |
| 2–5 users | 31 / 2,646 (1.17%) | 3 / 2,254 (0.13%) | $1,850 |
| 6–20 users | 9 / 247 (3.64%) | 1 / 1,053 (0.09%) | $1,773 |
These are observed-period exit ratios using September-end balances and reported connection shares, not cohort churn rates.
Where Crestline can hurt us:
- Standalone solos: They account for 142 of 186 competitive exits. Free invoicing, expenses and bank feeds directly challenge a $14/month product. This segment already had 31% annual churn before launch.
- Unconnected larger businesses: These are not protected simply because they have more users. The highest observed competitive exit ratio is among unconnected 6–20-user accounts, albeit from only nine exits.
- Revenue beyond solos: Multi-user customers represent only 24% of Crestline exits but 65% of the associated lost MRR. A solo-only defensive strategy would miss most of the revenue leakage.
There is an important inconsistency to investigate: 44 multi-user Harbourline accounts report moving to a one-user free plan. They may have reduced usage, retained other tools, or reported their destination imprecisely. We should verify their actual workflows and Crestline plans before claiming full multi-user substitution. The revenue loss is nevertheless real.
Where Crestline is not yet winning materially:
Only four accounting-firm-connected accounts moved to Crestline, versus 182 unconnected accounts. Relative to estimated bases, that is approximately 0.09% versus 1.37%. Connected customers account for roughly 49% of our MRR.
An accountant reconciles, files and advises inside Harbourline. Replacing us therefore means changing a shared operating workflow, not just choosing cheaper software. Crestline currently has no accounting-firm partner programme. That is strong evidence of protection today—not immunity: four connected customers did leave, and customer selection may explain some of the retention difference.
2. Which advantages can it copy?
| Advantage | Assessment |
|---|---|
| Core bookkeeping, invoicing, expenses and bank-feed functionality | Copyable; partly already matched. These are not a sufficient basis for defending subscription pricing. |
| Collaboration features and accountant-facing software | Copyable with investment. The $120M raise gives Crestline resources, although funding alone does not prove execution. |
| Larger-business functionality | Contestable. Its hiring signals an intention to close gaps; size is not a durable barrier. |
| Our 2,100-firm network and embedded client–accountant workflows | Not quickly reproducible by shipping features. Crestline must recruit firms, earn trust, change working practices and migrate client relationships. It can build an alternative over time, but cannot instantly copy our existing relationships. |
Our moat is therefore the working network, not merely the existence of an accountant portal. We should strengthen productive firm–customer connections rather than assume the partner count itself guarantees defensibility. We have no evidence here for additional advantages in brand, data, accuracy or cost.
3. Where Crestline is heading
The public hiring mix is more informative than “free forever” alone:
- 14 payroll engineers: a substantial signal toward an adjacent recurring workflow.
- Six mid-market multi-entity engineers: evidence of ambition beyond solos.
- Three accountant-partnership roles, labelled “exploring”: early recognition of the channel where we are strongest.
The likely strategy is free acquisition at the bottom, paid expansion through greater complexity and adjacent products. Paid plans already start at $25/month. Payroll and multi-entity could create monetisation opportunities while preserving the solo-free promise.
This is a directional inference, not a confirmed roadmap. Hiring does not establish launch dates or product quality. Nevertheless, we should expect a broader competitor, not permanently dismiss Crestline as a free solo tool.
4. What we should do
1. Reject universal free; authorise a controlled test. Our solo segment alone generates approximately $159,600 MRR, versus $1,958 lost to Crestline in the file. Those figures are not a forecast, but show the cannibalisation risk. Test a limited one-user offer for new self-serve prospects and selected at-risk customers, with clear usage boundaries and paid upgrade paths. Use holdouts to measure incremental activation, retained contribution, conversion, support costs and paid-plan downgrades—not free registrations.
2. Defend unconnected customers across all sizes immediately. Interview the multi-user switchers first. Establish whether price, unused seats, workflow simplicity or missing value drove departure. Improve onboarding and retention offers accordingly; do not default to discounts for every customer.
3. Invest primarily in the partner workflow. Make firm connection, shared reconciliation, filing handoffs and multi-client work easier. Help existing firms connect more suitable clients. Measure active shared workflows and cohort retention, not merely partner sign-ups.
4. Review in 90 days against explicit evidence. Track competitive losses and acquisition win rates by size and firm connection, lost MRR, partner recruitment and Crestline’s product releases. Reconcile the historical annual-churn figures with the separate 95-per-month baseline before making causal claims.
Decision requested: Approve a bounded free-plan experiment and partner-led defence—not a universal price reset. Reconsider broader action if connected-customer losses rise materially or controlled tests show that free creates sustainable incremental value.
Grades and run details
Decision model 82 · LLM judge 100
Decision model checks
- failUses the supplied evidence correctly69%
- passAddresses the actual decision91%
- passRespects explicit constraints42%
- passIdentifies material uncertainty90%
- passAvoids unsupported claims40%
- passProduces the required deliverable76%
- passFinds who's actually leaving84%
- passReads where Crestline is heading100%
- passCompetes with what customers really choose73%
- failGets the base of every number right59%
- passSeparates what a rival can copy100%
Run
- Run
- #1
- Time to output
- 87 s
- Submitted
- 5 Oct 2026