Fixed price vs T&M

The choice between the two foundational consulting contract models — fixed price and time-and-materials (T&M) — is not subjective. It maps to a 2×2 matrix: scope clarity on one axis, risk tolerance on the other. Here's the matrix, plus the two edge cases we never take on.

The 2×2 matrix

                  High scope clarity      Low scope clarity
Low risk tol.     FIXED PRICE             CAP + SHARE OVERAGE

High risk tol.    FIXED PRICE (fast)      T&M

Top left: high clarity + low risk tolerance = FIXED PRICE

The classic case. Customer knows exactly what they want, and doesn't want surprises. We can size the work precisely. We commit at a fixed price. Example: "plug this payment provider into the existing checkout", "write a CSV export endpoint with these 8 fields". Risk sits on our side. If we under-estimated, we eat it.

Bottom left: high clarity + high risk tolerance = FIXED PRICE (fast)

Same as above, but the customer is explicitly optimising for speed. We charge a 15-20% speed premium on top of the fixed price; in exchange we ship in 10 days instead of the normal 14.

Top right: low clarity + low risk tolerance = CAP + SHARE OVERAGE

The hybrid. Scope isn't 100% clear (often AI evals, research-flavoured work, R&D), but the customer has a maximum budget. We build a CAP (e.g. 50 dev-days) and commit to that. If we overrun, the overage is split 50/50: customer pays half the normal day rate, we eat the other half. This aligns incentives — we measure tightly, the customer doesn't over-specify.

Bottom right: low clarity + high risk tolerance = T&M

Classic consulting model. Customer accepts surprises will come, and commits to hourly/daily billing. We make ourselves available, we track hours daily, we report weekly. Example: "build me this new product line, nobody has done this in the market yet". Risk sits with the customer, we stay flexible.

The two edge cases WE never take on

Fixed price for AI evals — never

AI evaluation and R&D work has incredibly high variance. An LLM fine-tune can be 2 days or 6 weeks depending on how "friendly" the data is. A RAG system hitting an accuracy target (e.g. 92%) can be a 3-day sprint or a 4-month research project. Anyone fixing a price on this is either lying to themselves (and will slip), or to the customer (and will ship junk).

If someone says "I'll build you a 95%-accurate document classifier for a fixed 18,000 €" — run. Either they swallow a 50,000 € loss, or they ship at 70% accuracy and tell you "it depends on what we define as 95%".

We only take AI evals on T&M or CAP+SHARE-OVERAGE. Always.

One-off vs retainer

Some customers want to use us once. One project, fixed scope, done. Fine. But the best value (and from our side the highest quality) goes to 12+ month retainer partners. The retainer model:

• Monthly 5-10 committed days, fixed monthly fee.
• Quarterly review where we re-evaluate priorities.
• Customer can request ad-hoc projects anytime, day rate is 15% lower than the new-customer price.
• Knowledge accumulates in the team — by month 6, we don't need to re-onboard at every engagement.

Retainer customers work 40% more efficiently with us by month 6, because the domain knowledge sits in the team. Long-term it's BOTH cheaper for the customer AND higher quality.

The selector aid

On every opening call we ask:

1. How precise is your specification? (1-5)
2. How important is it that the final cost is known up-front? (1-5)
3. Is this a one-off project or a long-term collaboration target?

The answers automatically route to the matching contract type. We don't decide for the customer — the questions do.

The takeaway

Contract type isn't a stylistic choice. It's a data question. Anyone who offers the same type for every engagement (always fixed, or always T&M) is either over-promising or under-delivering. We rotate through 4 templates based on the customer's situation, and that's what produces price-honest, quality-honest collaboration.