Where the cost of software stopped being zero

Most of our 7,400 hospitals sign one-year subscriptions, but the corporate groups that now own a growing share of US veterinary practices negotiate agreements of three to five years, with ramped prices, implementation credits and, often, termination rights. Under ASC 606, the US revenue standard, the length of a contract for accounting purposes depends on what each party can enforce, not on the term printed on the cover page, so a five-year agreement that the customer can end after two years with notice is in substance much shorter. I review every agreement above $1 million of annual value with our revenue team before signature. Our disclosure of remaining performance obligations covers the enforceable portion only, sales commission is paid on annual contract value rather than total contract value, and the board sees a schedule of cancellable revenue by customer. Our contracted backlog looks smaller than some peers' as a result. It is also the backlog we expect to collect.

When I arrived, stock compensation cost 19 per cent of revenue, and the share count grew 4.6 per cent a year after shares withheld to cover employees' tax. Grants were set in dollars, so when the share price fell from $88 in November 2021 to $31 a year later, the same dollar grants would have needed almost three times as many shares. In 2022 I proposed, and the Compensation Committee adopted, an annual share budget: gross grants capped at 3 per cent of shares outstanding in 2023, 2.5 per cent in 2024 and 2.25 per cent in 2025, with refresh grants set in share bands by level and a ceiling of 1.5 per cent on net dilution. We moved part of junior engineers' pay into cash, which costs about $14 million a year. Each quarter I report gross burn, net dilution and stock compensation per employee to the committee beside the expense itself.

Our AI notes assistant turns an exam-room conversation, recorded with the pet owner's consent, into a draft medical record that the veterinarian edits and signs. Each note incurs transcription and model-inference charges from two outside providers under usage-based contracts. During the beta most of that bill sat in research and development; from the third quarter of 2024 I moved it into cost of revenue, where it belongs once paying customers depend on a feature. The move made the problem visible. By early 2025 a note cost $0.31 and subscription gross margin had fallen from 81.2 to 77.4 per cent. I now keep a unit-cost ledger for every AI feature, covering cost per use, uses per clinician and the allowance in each price tier, and engineering reviews it with my team each month. Routing simpler consults to a smaller model, trimming audio and caching common passages have taken the cost to $0.11 a note.

About a third of our revenue comes from processing card payments at hospital front desks and through online invoices, under arrangements with a sponsor bank and a payment processor. We report that revenue net of interchange and card-network fees, so the figure shows the margin we keep rather than the $22.8 billion of volume that passed through in 2025. Payments behave differently from subscriptions. Card mix, the share of payments taken online, chargebacks over disputed treatment bills and the occasional fraudulent clinic account all move the result. My team reconciles processed volume, net take rate and losses each month by hospital cohort. In 2023 I opposed a proposal to win larger groups with a flat processing price below our blended cost for premium rewards cards; we offered tiered pricing instead and lost one 60-hospital group. Our net take rate has held at about 0.86 per cent of volume since 2022.

By early 2025 we held $610 million of cash and investments and owed $287 million on convertible notes maturing in March 2026. I set out every use of that cash in one paper with one return test: the acquisitions corporate development had found, a repurchase programme, early repurchase of the notes, and simply holding cash. A buyback is an investment in our own business at the market's price, so I asked it to clear the same hurdle as a deal, using our own estimate of value rather than a hope that the shares were cheap. The board approved a $64 million purchase of Cobbleway, a pharmacy-fulfilment software company, a $120 million buyback over two years and repayment of the notes in cash at maturity. We declined a $240 million acquisition that our chief executive had initially favoured, and I describe that disagreement among my impact cases.

I have built public-company finance twice: once from a team of twelve at Veyrholt Security before its 2018 listing, and once by tightening the controls of an already listed Brindlewood. The work is less visible than a roadshow and decides more. It means documenting every control over revenue, stock compensation and the close, testing them before an auditor does, restricting who may approve non-standard contract terms and closing the books in days rather than weeks. Newly listed companies have a transition period before their first management assessment of internal control over financial reporting and, depending on size and filing status, longer before an auditor must attest to it. I advise boards to treat that period as preparation rather than relief. At Veyrholt our first required assessment found no material weakness. At Brindlewood I brought the quarterly close down from eleven working days to six.