Why SAP consultancy bills run into the millions — and where AI changes the maths

Ask a finance director what frightens them about an SAP upgrade and the answer is rarely the technology. It is the invoice.

There is good reason for that. On most large SAP programmes, implementation consultancy is the single largest cost line — routinely a bigger number than the software itself. The work is genuinely hard, the specialists are genuinely scarce, and the day rates reflect both. None of that is a scandal. But a lot of what sits inside that number is not deep technical work at all.

Where the money actually goes

Strip a programme back and the external spend tends to fall into two piles.

The first pile is specialist build and configuration — the work that genuinely requires deep, current SAP expertise. That is money well spent.

The second pile is everything that surrounds the build: readiness assessments, requirements gathering, data profiling and cleansing preparation, test-scenario design, cutover planning, stakeholder communications, training material. This work is essential, but it is not specialist in the same way. It is structured, repeatable, knowledge work — and historically it has been billed at specialist rates simply because the in-house team had no efficient way to do it themselves.

That second pile is where the maths is changing.

What AI actually shifts

The economic argument for AI is no longer speculative. McKinsey's analysis puts the potential value of generative AI across the economy in the trillions of dollars, concentrated precisely in structured knowledge work; their ongoing research shows organisations moving from experiment to adoption, and spending forecasts from Gartner point the same way — investment is rising, not levelling off.

For an SAP programme, the practical effect is narrow but real: the surrounding work in that second pile is exactly the kind of structured knowledge work AI is good at supporting. An in-house team with the right method can now produce a credible readiness assessment, a first-pass data-quality review, or a set of business test scenarios in hours rather than billing weeks of external time for the same output.

This does not replace the specialists. It changes what you need them for — and shrinks the second pile.

The catch: it only works if the prompting is good

There is a real caveat, and it is the whole game. Ad-hoc AI use — typing a vague question and hoping — produces vague, inconsistent, sometimes confidently wrong output. That is worse than useless on a programme where decisions carry real money.

The difference is method. A structured, context-rich approach — one that sets the role, the inputs, the constraints and the required format — produces output you can actually take into a steering meeting. That skill is learnable, but learning it across a team, mid-programme, is its own cost.

That is the gap our toolkits close. They package the method, so your team gets specialist-quality structure on the surrounding work without paying specialist rates to acquire the skill first.

The bill will never go to zero — nor should it. But the share of it that is really just structured knowledge work, billed externally for want of a better option, is no longer a fixed cost of doing business.