Skip to content
Fritz Gerald ZephirinMenu

Social

The Service Company That Becomes Software

4 min read

Andy Grove had a name for the thing that is currently happening to every services firm on earth: a strategic inflection point — a change in the environment so fundamental that the old way of operating becomes a liability, felt at the edges of the company long before the center believes it. I run customer-experience and employment operations, which puts me at the edge in question. Here is what it looks like from there.

For sixty years, service businesses lived under a ceiling that the economist William Baumol described in 1967: in work where productivity can’t grow, costs only rise. A support conversation took roughly as many human minutes in 2019 as in 1989, so service margins stayed pinned at 20–35 percent while software enjoyed 70-plus. The entire structure of the industry — offshore labor arbitrage, seat-based pricing, the BPO model itself — is an adaptation to that ceiling.

AI just removed it. Not rhetorically — measurably. The best field study we have showed AI assistance lifting support-agent productivity 14 percent overall and 34 percent for novices. In production, well-run AI handles the majority of tier-one volume. For the first time since Baumol wrote, the cost of delivering a unit of service is falling while the value delivered holds. That gap — falling cost, stable value — has to go somewhere. The entire strategic question of the services industry this decade is: to whom?

There are exactly three answers. The savings go to the client, as discounts. They go to a software competitor who eats you. Or they go to you, as margin. The first is the default, and the default is a trap.

Watch how it happens. A service firm prices by the hour or the seat. AI cuts the hours. Under hourly pricing, the firm has just engineered its own revenue decline — every efficiency gain passes automatically to the client, and the firm keeps its old margin percentage on a shrinking base. This is why the pricing migration, not the technology, is the actual battle: the firms that survive this transition will be the ones that moved from selling effort to selling outcomes — the resolved ticket, the completed payroll run, the closed book — before the effort became visibly cheap. Price the outcome and the efficiency gain lands in your margin. Price the hour and you’ve volunteered to shrink.

History offers a clean pair of cases. ADP started in 1949 as a payroll service bureau — humans processing paychecks — and converted, over decades, into a software company with service wrapped around it; it has now outlived essentially every company that ever competed with it. Typesetting shops faced the same kind of transition when desktop publishing arrived, discounted to defend volume, and vanished. Same inflection, opposite pricing instincts, opposite outcomes.

The honest counterargument is Klarna. In 2024 it announced its AI assistant did the work of 700 agents; in 2025 its CEO conceded the cost-first version produced “lower quality” and began hiring humans back. Read correctly, that’s not evidence against the conversion — it’s evidence about sequence. Klarna cut labor first and discovered quality later. The conversion that works runs the other way: hold quality fixed as a constraint, let AI absorb volume underneath it, keep humans on the judgment-dense residue — escalations, exceptions, the moments where trust is won — and reprice as outcomes while quality is provably intact. Cost falls out as a consequence. Companies that chase it as the objective end up issuing public corrections.

There’s a second failure mode worth naming: clients may simply insource. AI makes service work more legible — if a model plus a knowledge base can do it, why pay a vendor? The defense is the part of the work that isn’t legible: the accumulated exception-handling, the escalation trust, the compliance scar tissue. Which is to say, the workflow, not the model — and it has to be deliberately deepened, now, while the window is open.

Because the window is the point. Margins like this get repriced once per technological generation. Service founders are holding contracts, relationships, and domain knowledge that software companies would kill for, at the exact moment the margin ceiling lifted. The ones who treat AI as a cost program will hand the difference to procurement. The ones who treat it as a repricing event get to become what their industry always envied: a software business that knows how to serve.

Sources & further reading