When AI shortens the conversion, stop selling the hours

Most public discussion of artificial intelligence and software asks whether AI will replace the products themselves. In the vertical markets where Sorrelgate operates, I think that is the slower and smaller effect. The faster one is on services, and above all on the work of moving a customer from its old system to ours. My position is that vertical software businesses should stop billing that work by the hour, publish fixed prices by scope, pass a visible part of the saving to customers and accept a fall in services revenue. The alternative, quoting historical hours for work that now takes half as long, will be discovered, and in public-sector markets it will be discovered by the people who write the next tender.

Where the hours went

Consider a mid-sized municipality replacing its permitting and licensing system. It has thirty years of records spread over two or three earlier systems, some fields used for purposes nobody documented, scanned plans attached to the wrong parcels and a building inspector who knows which entries to ignore. Converting that history has traditionally been the longest and costliest part of an implementation. In our Civic group, conversion took about 40 per cent of implementation hours, and implementation services made up about a fifth of the group's revenue.

In 2024 six Civic businesses trialled AI-assisted tools for the mapping and cleansing steps: suggesting how old fields correspond to new ones, flagging records that look inconsistent and drafting transformation rules that a conversion specialist then checks. The specialists remained essential. The hours fell by roughly two-fifths to a half, depending on how old and how untidy the source data were, and the defects customers found at acceptance fell as well, because the tools never tire of the inconsistencies people skip late in the day.

Under contracts that priced conversion on estimated hours at a published rate, that meant less revenue: roughly C$20 million a year across the Civic group once the tools were in general use. The president of the group, Bram Oosterhuis, proposed that we keep quoting on historical estimates for two years, arguing that the value to the customer had not changed and that competitors would do the same. Our chief financial officer was concerned about a visible fall in services revenue in a year when investors were already asking what AI meant for software companies.

Bram's argument deserved a hearing, and it got one in front of our chief executive. Pricing on value is legitimate, and a customer rarely cares how many hours a supplier spends. But our contracts did not say value. They said hours, at a rate, with an estimate. A county auditor or provincial procurement officer who later compared our estimates with the effort recorded would have found a gap we had chosen not to disclose. In markets where reputation travels through purchasing associations and neighbouring municipalities, that is a cost no spreadsheet captures.

Price the scope the customer can see

Since 2025 our Civic businesses have priced conversion by scope bands a customer can check for itself: how many source systems, how many records and documents, and how many years of history are to come across. The prices are given to customers in advance and sit on average about a quarter below what the same conversion used to cost. Where the work proves larger than its band, we absorb the difference within limits; where it proves smaller, the customer still pays the band. Customers told us they preferred certainty to a lower but unpredictable figure, which surprised me less than it surprised our sales leaders.

One complication taught me something. Several public bodies have purchasing rules that require an hourly rate card in the contract whatever the pricing model. Bram insisted we keep one as an appendix, used only when scope changes, and it has mattered more than either of us expected. It lets a procurement officer evaluate our proposal on the terms their rules demand while the customer still receives a fixed price.

The first full year was mixed in the way I expected. Services revenue in the group fell by about C$13 million. Services gross margin rose by thirteen points, because we were no longer staffing conversions against old estimates. Implementations got shorter by about a third, so recurring revenue started earlier. We won more bids against incumbents quoting by the hour. Forty-six conversion specialists retrained for testing and configuration work, and nine chose to leave.

What the maintenance fee will have to carry

The harder question comes next. If AI shortens implementations, it will also change the cost of supporting and improving a mature product, and public-sector customers who pay annual maintenance of eighteen to twenty-two per cent of their licence value will ask what they are paying for. I do not think the answer is to cut maintenance prices pre-emptively. Maintenance funds continuing development, regulatory updates, security work and support, and several of those costs have risen. But I do think we owe customers a clearer account of what the fee buys, and that the businesses which give that account will keep their customers longest.

There is also an argument about moats that I hear from investors and want to answer. If AI makes conversion cheap, the reasoning goes, customers can leave more easily, so the value of an installed base falls. Some of that is true. But the hardest part of leaving a vertical system was never the data; it was retraining staff, rewriting procedures and persuading a council or a court to accept the risk of a change. Cheaper conversion lowers one barrier among several. It hurts most the suppliers whose customers stayed only because leaving was expensive, and those customers were never as loyal as their renewal rates suggested.

I would offer three tests to any vertical software board. First, compare the last twelve months of services estimates with the hours actually recorded, and ask whether the gap is widening. Second, ask your general managers what they would charge if they had to publish the price before winning the work. Third, ask what your customers' procurement officers would conclude if they saw both numbers. If the answers are uncomfortable, the hours have already gone, and the only remaining choice is who tells the customer.