The three alternatives, priced.
A large consultancy, two senior hires, or a contract build team. What each costs over the same weeks, and — because it is usually the more useful answer — the cases where each of them beats us.
The defaults are published market ranges, cited below. They are not our claims about anyone — change them to the quotes actually in front of you. The arithmetic runs in your browser and nothing leaves this page.
How each column is built: the consultancy and the build team are staffed at three people, forty hours a week, so the comparison is a comparable team rather than a favourable one. The build shop figure applies the 1.5× loaded multiplier the rate surveys report for coordination, ramp-up and rework — the quoted rate is not the paid rate. The two hires are loaded at 1.3× base for employer cost, and only the slice falling inside your window is counted, which flatters them: the roles do not end when the project does. Our column is $18,000 for the roadmap plus $75,000 for the build, with embedded co-development added at $32,000 a month past week ten. Model and infrastructure costs are yours in all four columns. If your figures differ, use yours — the shape is the argument, not the decimals.
- Consultancies do not publish rate cards. Big-four AI practices price RFP by RFP, partner-controlled. The $150–500 blended band is from 2026 rate surveys of large implementers; project engagements in those surveys run $100,000–500,000 for a single use case. We publish ours on the pricing page, which is the actual difference this row is measuring.
- Senior ML engineer salary depends heavily on who you ask. 2026 US surveys report base pay from about $122,000 to $216,000 for the same title, and total compensation well above that once equity is counted. $190,000 sits inside that band; move it to whatever your offers actually look like.
- Agency rates are quoted low and land high. 2026 guides put US agencies at $80–200 an hour and nearshore at $40–100, and separately report that the fully loaded cost runs 1.4–1.8× the quoted rate once coordination and rework are counted. We apply 1.5×.
Ranges as published in 2026 by third-party salary and consulting-rate surveys, read on 2026-08-29. They describe categories, not any named firm, and none of them is a quote anyone gave us.
Against a big-four AI practice
They are the safer signature. We are the faster system. The honest test is which of those your problem actually needs.
When they are the right answer
- You need a name on the paperwork that your board already recognises, and clearing procurement is itself a deliverable.
- The work spans several countries and regulators, and the change-management programme is bigger than the build.
- You need forty people on it next month and your organisation can absorb the ramp.
- The problem is operational rather than technical — the systems are fine and the operating model is not.
If two of those are true for you, stop reading this section. We would rather lose the engagement than sell against a fit that is better than ours.
If what you need is the signature, take the signature — that is a real requirement and we are not it. We will still leave you the architecture notes from the briefing, and they are yours to hand to whoever builds it.
Against two senior hires
Hiring is the right answer more often than a firm likes to admit. It is also the slow answer, and it is the one you cannot undo in a week.
When they are the right answer
- The work is permanent and central. If this system is your product, it should be built by people who stay.
- You already have someone senior who can interview for it. You cannot hire what you are not yet able to evaluate.
- You can wait months for the first line of production code, and longer for the second hire to land.
- What you are buying is the capability itself, not a particular system.
If two of those are true for you, stop reading this section. We would rather lose the engagement than sell against a fit that is better than ours.
The best outcome of an engagement is frequently that you hire — and hire better, because you now know exactly what the role has to do. Embedded co-development exists for that case: it is the shape that leaves your team able to do the work without us.
Against a contract build team
Usually called an offshore team, though geography is not the difference that matters. The difference is who carries the specification, and what happens when it is wrong.
When they are the right answer
- The specification already exists, is complete, and someone on your side reviews the work every week.
- The work is volume of a known shape: screens, integrations, migrations, a port.
- Rate is the binding constraint, and a wrong specification would cost you little.
- You have an engineering manager with the time to be the missing half of the arrangement.
If two of those are true for you, stop reading this section. We would rather lose the engagement than sell against a fit that is better than ours.
If your specification is genuinely finished, a build team is cheaper than us and you should use one. Our discovery stage exists because that specification almost never exists yet, and because writing it is the part that decides whether the build works.
Three honest cases where someone else is the better answer, and one calculator that you can turn against us. If the numbers still point here, the briefing is 45 minutes and free, and you leave it with the architecture notes whether or not you hire us.