The consultant who can build — and what it means for your finance department

When a shift in working life is real, you notice it less in forecasts than in the moment someone counts up inside their own firm and is surprised by the answer. That is exactly what has just happened at Boston Consulting Group, and it is worth a moment from anyone who will one day buy software — even if strategy consulting has never come near them.

What BCG found in its own house

Mel Wolfgang took over as BCG's North America chair in May 2026. One of his first questions was a plain stocktake: how many of the firm's traditional consultants actually hold skills close to those of its technical staff — the people from IT architecture and software development? It was not measured by self-assessment, but through tests and through technical work people had demonstrably done on client cases (Paradis 2026).

««I've actually kind of got double the workforce that I thought I had.» — Mel Wolfgang, BCG»

Nearly as many people from a classic consulting background — in his words, «the English lit major with an MBA» — had acquired skills approximating those of the firm's technical hires. BCG is now hiring more of both profiles and explicitly looking for the overlap: technical proficiency together with the judgment consulting has always prized. In interviews, candidates are now asked whether they have worked with AI tools.

Why the split was always artificial

Dividing the world into «one understands the business» and «one builds the software» was never an insight about people. It followed from how expensive writing a program used to be. Because building cost so much, it had to be done by someone who did nothing else all day — and that person then had no time left to understand why the VAT return reconciles the way it does.

Out of that came the translation layer every software project knows: the specialist describes what they need. Someone writes it into a specification. Someone else builds something from it. At acceptance it turns out three edge cases are missing — obvious in accounting, absent from the document. Every one of those handovers costs time, and in every one a piece of the knowledge that actually constitutes the process goes missing.

A large corporation could absorb that. It can afford both people and put them in the same room. An SME with thirty or a hundred employees never could — which is precisely why the shift BCG is observing in itself is the more interesting news for smaller firms.

What it means for a Swiss finance department

In practice: you no longer have to choose. Not between someone who understands your year-end close and someone who can write a program. That choice is why so much in finance departments still runs by hand — not because nobody wanted to automate it, but because the road from «surely this could work» to something that actually runs led through too many other people's heads.

When the same person can do both, the specification disappears as an intermediate step. You look at the routine together, and what comes out is not an interpretation of your description but a program written for your actual workflow. The edge cases that appear in no document surface early, because someone is asking about them who knows they exist.

The catch: judgment gets dearer, not cheaper

It would be convenient to conclude that everything now gets easier. Wolfgang says the opposite: «The role of judgment has increased enormously, and that's an apprenticed skill» — and judgment is not learned on a course but at work, next to someone who has it. What a tool produces, he notes, often takes several passes and a lot of refinement before it is any good.

There is now clean evidence for this, and it comes from a field experiment with 758 BCG consultants. On tasks inside what the tool does well, participants produced more, faster and better. On tasks beyond that boundary the effect reversed: those working with AI were 19 percentage points less likely to reach the correct solution than the control group without it (Dell'Acqua et al. 2026). And the boundary does not run where you would guess — it is jagged, and from the outside you cannot see which side of it you are standing on.

A second large study shows where the gain goes: across 5,172 customer-support agents, issues resolved per hour rose by 15 per cent on average, with the least experienced benefiting most clearly (Brynjolfsson, Li & Raymond 2025). Capability, in other words, spreads downward. Judgment does not — it stays scarce, and becomes more valuable for it.

For your department that is not an academic nicety but the central design question: where may a program run through on its own, and where does it stop and wait for a person? An internal report can finish without asking. With a VAT return, most people want to look first. That decision is yours, and it belongs at the start of a project — not at the end, once the program is already running.

Where Swiss SMEs currently stand

The Swiss figures show a pace many underestimate. In a survey of 300 SMEs from German- and French-speaking Switzerland, 34 per cent said they deliberately build AI into their work processes — a year earlier it was 22 per cent. The share doing nothing with it at all fell from 45 to 29 per cent (AXA & Sotomo 2025).

The more telling number comes later: of the firms already using AI, only 33 per cent had clear data-protection rules for it; among the smallest, with five to nine employees, it was 23 per cent. Use has outrun order. That is not an argument for slowing down — it is a sign that the question «where does it stop, and what does it record» tends in practice to be asked far too late.

What you can do with this

The conclusion for an SME is not to hire this profile yourself. A fifty-person business will not build a four-tier AI curriculum the way BCG runs one for its own staff, and it should not try. The point is different: you can buy the result of this development without having to live through the development yourself.

Concretely, that means taking a single, clearly scoped workflow — not the whole department — and having it built by someone who understands what it is actually about. You set the points where the program waits for your approval. At the end it belongs to your business: code, documentation, access. No system you could not replace, and no dependency that gets more expensive over time.

To find out whether that holds for you takes no strategy and no preparation. It takes one answer: which routine costs you the most time every month? With that single question in hand, one conversation is enough to establish honestly whether a program is worth it for you — and just as honestly when it is not.