Chief Financial Officer / United States
Four decisions that cost something before they paid
- Revenue, 2025
- $612M
- 61 per cent subscriptions, 32 per cent payments, 7 per cent implementation and training
- Net annual dilution, 2025
- 1.1%
- Down from 4.6 per cent in 2021 under a board-approved share budget
- Rule-of-40 score after stock compensation, 2025
- 26
- 36 on the usual basis: 17 per cent growth plus a 19 per cent free-cash-flow margin
Decisions & consequences
01Charging for the AI notes assistant before it ate the margin
Brindlewood launched its AI notes assistant in April 2024 inside our top subscription tier, at no extra charge. Veterinarians liked it: a consult that used to end with fifteen minutes of typing now ended with a draft record to check and sign. By the first quarter of 2025, 3,100 hospitals were producing 2.6 million notes a month. Each note cost us $0.31 in transcription and inference charges, about $800,000 a month and rising, and subscription gross margin had fallen from 81.2 per cent a year earlier to 77.4 per cent. Two competitors were offering similar assistants free.
Our chief product officer, Elliot Castellane, and our chief executive wanted to keep the feature free and unlimited for at least another year to win hospitals from those competitors. Our head of sales feared any charge would help rivals. I argued that a feature whose cost rises with every use cannot be priced like software that costs nothing to serve twice, and that the longer we waited, the more customers would treat unlimited use as an entitlement. The decision was the chief executive's. She chose a middle course, and I accepted that it would slow adoption.
From July 2025 the top tier included 250 notes per veterinarian per month, enough for about four in five clinicians, with further notes at $0.40 each. Lower tiers could buy the assistant for $149 per veterinarian per month. Engineering, with finance alongside, routed simpler consults to a smaller model, trimmed silence from audio before transcription and cached common text. About 140 hospitals left or downgraded, citing the free alternatives, and took roughly $3.1 million of annual recurring revenue with them. Adoption growth slowed for two quarters.
By mid-2026 the cost per note was $0.11, and 2,050 hospitals on lower tiers had bought the add-on, worth about $14 million of annual recurring revenue. Subscription gross margin was back to 80.4 per cent. Every AI feature on our roadmap now carries a cost per use and a pricing plan before anyone builds it.
I would do two things differently. I would have priced the feature at launch, when nobody had yet been promised unlimited use. And I would have shown hospitals a usage dashboard a quarter before metering began. About 300 were surprised by their first invoice, and we issued $0.9 million of credits.
Replacing dollar grants with a share budget after the price fell
In November 2021 Brindlewood's shares traded at $88. By October 2022 they were at $31. Our equity awards were set in dollars, so the refresh grants planned for 2023 would have required almost three times as many shares as the year before. On the draft plan, gross grants would have reached 6.1 per cent of shares outstanding, with stock compensation already at 19 per cent of revenue. Shareholders who had stayed with us through the fall would have been diluted by the people we were paying to reverse it.
Our chief people officer and chief technology officer proposed topping up every engineer's unvested awards to restore their original value. Their concern was real. About 1,100 engineers held grants worth far less than they had been told to expect, and competitors were recruiting them. The chair of the Compensation Committee was initially sympathetic. I proposed a share budget instead: gross grants capped at 3 per cent of shares outstanding in 2023, falling to 2.25 per cent by 2025; refresh grants in fixed share bands by level, so that employees carried some of the price risk every owner carries; more of junior engineers' pay in cash; and retention grants concentrated on about 260 engineers who ran the payments platform, the medical record and our data infrastructure.
My paper set out what this would cost. The shift to cash added about $14 million a year. Engineering attrition rose from 11 per cent in 2022 to 16 per cent in 2023, and we lost two senior payments engineers to a competitor. Some managers felt finance had overruled them on their own people, and in part we had.
By 2025 stock compensation was 10 per cent of revenue, gross grants were 2.1 per cent of shares and net dilution was 1.1 per cent, down from 4.6 per cent in 2021. Engineering attrition fell back to 10 per cent. The committee now approves the share budget before the salary budget, and I report gross burn and net dilution each quarter alongside the expense.
I underestimated how the change would feel to mid-level engineers whose awards had already lost value. Next time I would publish the per-level model to employees before the first grant cycle rather than after it, and I would give managers a small discretionary pool from the start instead of adding one a year later.
Declining a larger deal for a small acquisition and a measured buyback
In early 2025 Brindlewood held $610 million of cash and investments, with $287 million of convertible notes due in March 2026. Three proposals reached the executive team. Corporate development, with our chief executive's initial support, recommended buying Brackenfold Agrisoft, a software company serving rural and livestock practices, for about $240 million. It had $48 million of revenue growing 6 per cent a year and customers who rarely overlapped with ours. One of our directors, a former investment banker, wanted a $300 million accelerated repurchase while our shares traded near $38. A smaller team proposed buying Cobbleway, a pharmacy-fulfilment software company with $11 million of revenue growing at 40 per cent, for $64 million.
I put all three, plus holding the cash, into one paper with one test: the return each would earn against our cost of capital, judged on our own estimate of what Brindlewood shares were worth rather than on the market's mood. Brackenfold failed it. Its growth and margins could not support the price without savings that depended on moving rural practices onto our platform, which our own product team doubted was possible. The large buyback passed on value but would have left too little cash to repay the notes without borrowing. Cobbleway passed comfortably, because its prescriptions would run through records our hospitals already kept with us.
The board approved Cobbleway, a $120 million buyback over two years sized first to offset dilution and then to reduce the share count modestly, and repayment of the notes in cash. Our chief executive accepted the recommendation after a difficult meeting. Six months later a sponsor-backed competitor bought Brackenfold for $255 million, which gives that competitor a route into rural practices we do not serve.
Cobbleway closed in June 2025 and is now live in 1,900 of our hospitals, at about $19 million of annual revenue by mid-2026. By September 2026 we had repurchased 2.3 million shares for $96 million, an average of $41.70. The notes were repaid in cash in March 2026.
I would change one thing. We bought about a third of those shares in a quarter when the price stood above our own estimate of value. I would now write a price ceiling into the authorisation itself, so that the discipline we apply to acquisitions also applies to our own stock.
Delaying a listing to fix revenue controls first
In the spring of 2017 Veyrholt Security was preparing to list in the autumn. Revenue was on course for $142 million, growing more than 40 per cent, and our bankers believed the market would value us generously. During the audit of three years of financial statements for the registration statement, our auditors found that some multi-year subscription contracts carried side letters, agreed by sales staff, that gave customers extra termination or refund rights finance had never seen. They concluded that we had a material weakness in internal control over revenue, which the registration statement would have to disclose.
I was vice president of finance and corporate controller, reporting to our CFO, Joanna Brannagh. Our chief executive and our largest venture investor wanted to keep the autumn date, disclose the weakness and remediate after listing, as other companies had done. I argued that a security company asking customers to trust its controls could not list with a weakness in its own, and that we did not yet know how much revenue the side letters touched. Joanna backed a delay. The board agreed to defer by at least two quarters and accepted the risk that the market window would close.
We reviewed all 2,600 active multi-year contracts and found 38 side letters. Eleven changed the timing of revenue, and the audited financial statements were corrected before we filed. We created a deal desk that had to approve any non-standard term, made unapproved side letters a disciplinary matter, adopted the new revenue standard in the same year and documented every revenue control. Our auditors tested the remediation before filing.
We listed on the New York Stock Exchange in June 2018 at $16 a share, raising $172 million. Bankers had spoken of a higher range the previous autumn, and software valuations had fallen in between. Two senior sales leaders left over commission adjustments on the affected contracts. The registration statement described the weakness and its remediation, and our first required management assessment of internal control found no material weakness.
I would start control documentation two years before a listing, not fifteen months. And I would have introduced the deal-desk rules with a quarter's notice instead of adjusting commissions retrospectively, a choice that cost us two good people.