AI-Assisted cost optimisation and strategic sourcing

Two things get sold together under this heading and should be separated, because they work differently and one of them is frequently oversold.

Strategic sourcing changes the shape of your cost base. AI-assisted optimisation finds waste faster than a person would. The first is structural, the second is a tool — and confusing them produces a business case that does not survive contact with year two.

Why IT cost is under scrutiny now

Three things converged for Swiss and DACH mid-market companies.

Technology costs rose in absolute terms — cloud, licences and engineering salaries all moved upward, and the total became large enough to appear as its own line rather than inside overhead. Boards learned to ask, after a decade of digital programmes with disappointing returns. And cost comparison became easier, because cloud billing makes consumption visible in a way that capital purchases never did.

The result is that IT cost is now managed rather than approved. That is reasonable, and it changes what a credible proposal looks like.

What strategic sourcing actually means

It is not the rate card. Rate is not cost — cost is rate multiplied by time taken, plus the coordination overhead you absorb, plus the cost of work that has to be redone.

What makes sourcing strategic rather than transactional is that it changes the structure of your capacity:

• Permanent, internal for work where continuity is the value — business knowledge, architecture ownership, anything customer-facing
• Flexible, external for work that is specialised or time-bounded — a platform build, a migration, an evaluation methodology

The financial effect is that a share of engineering cost becomes variable. In an uncertain market, variable cost is worth a premium per hour, and that is the trade being made. It should be made explicitly rather than arrived at by hiring failure.

The counterpart risk is real and worth naming: a sourcing arrangement that never transfers knowledge converts a capacity problem into a dependency. The criteria that prevent that are in choosing a software development partner in Switzerland.

Where AI genuinely reduces cost

Three places, with honest expectations.

Cloud spend analysis. Tools that identify over-provisioned resources, orphaned storage and idle environments do work that a person could do, faster and continuously. The saving is real. It is also available without AI — the tooling just makes it cheaper to sustain. The prerequisite is cost attribution; without tagging, any tool shows you an unattributable bill in a nicer interface.

Support and ticket triage. Classifying and routing incoming issues removes genuine manual effort, and the output is reviewed by a person, which keeps it in a low-risk tier.

Anomaly detection on spend and usage. Catching a cost spike in hours rather than at month end is a measurable saving, and this is a problem shape models are actually good at.

Where it is oversold: predicting future spend with confidence intervals, and automated remediation without review. The first inherits your historical patterns including the anomalies; the second is how a cost-optimisation tool takes down production.

Telling a structural saving from a temporary one

This is the distinction that decides whether a cost programme holds.

Structural — the cost no longer exists. A licence not renewed, a data centre contract ended, a system retired, a capacity model changed from fixed to flexible. Verifiable in the ledger, and it does not come back.

Temporary — the cost was deferred or compressed. A hiring freeze, a postponed upgrade, maintenance cut to fund new build. These show up as savings this year and as a larger number later.

The third category is the one to watch: savings that were never real. Hours “saved” counted as money when nobody left and no capacity was redeployed. This is the most common line in a weak business case, and a finance reviewer will find it.

The honest framing for AI-assisted work: the first review produces a step change, subsequent ones produce maintenance. If you are promised a recurring large percentage every year, that is the first year repeated in a spreadsheet.

Measuring so it holds up

• Capture the baseline before starting. It cannot be reconstructed afterwards, and without it every later claim is arguable.
• Report unit cost, not total. A growing business should spend more; cost per customer or per transaction tells you whether it is spending well.
• Name an owner. One person reporting variance monthly, not a committee. Without this, a cost programme is a project that ends.
• Report the misses. A leader who reports what did not deliver buys credibility for what did.

What not to cut

Two things get cut in cost programmes and cost more than they save.

Logging and monitoring. It looks efficient and removes the evidence you need during an incident — and for regulated workloads, part of what you are obliged to retain under revDSG/nDSG, GDPR or FINMA expectations.

Maintenance, to fund new build. The most common move in a squeezed year. It converts a visible cost into an invisible liability, and the bill arrives two years later as an outage or a forced migration.

Related reading

• A CEO’s guide to cloud cost optimisation
• Maximising engineering ROI: the CFO’s guide
• Dedicated engineering teams

If you want a structural read on your cost base rather than a list of quick wins, book a call. No obligation.