Build vs Buy Advisory for AI
A structured comparison of building, buying or combining, costed over three years with switching risk priced in.
The build versus buy question is rarely decided on merit. It gets decided by whoever presented last, or by an engineering team's preference, or by a vendor demo that showed the happy path. The decision deserves better, because it is expensive to reverse and it usually gets reversed at the worst possible moment.
Three options, not two
Framing this as a binary is the first mistake. In practice there are usually four routes worth costing, and the middle two win more often than either extreme.
- Buy a finished product and adapt your process to it.
- Buy a platform and build your specific logic on top of it.
- Build on open foundations with managed infrastructure underneath.
- Build the whole path including model hosting and orchestration.
Each option gets the same treatment: build cost, three year run cost, time to first value, and the cost of leaving.
The factors that actually decide it
Is this differentiating work?
If the capability is part of what customers choose you for, building usually wins over a three year horizon even when buying is cheaper in year one. If it is a commodity function that every company in your sector performs identically, buying almost always wins and the engineering argument for building is usually about interest rather than value.
How specific is your data and process?
Vendor products embed assumptions about how work flows. If yours matches, you get value quickly. If it does not, you spend the first year in configuration and the second year discovering the parts that cannot be configured. We test this by walking your real edge cases through the vendor's model during evaluation rather than accepting the demo path.
What is the exit cost?
This is the line most comparisons omit. We price it: data export format and completeness, whether prompts and evaluation sets are portable, contractual notice, and the migration effort to a replacement. A product that is cheaper per year but holds your data in a proprietary structure can be the more expensive choice.
Who maintains it in year two?
Built systems need an owner. If the honest answer is that no one has capacity to maintain this after launch, that is an argument for buying regardless of the cost model, and we will make it plainly.
How we evaluate vendors during the comparison
Where buying is a serious candidate, we run a short structured evaluation rather than relying on marketing material. That means a requirements matrix built from your process, a scripted test using your own edge cases, direct questions about data handling and model provenance, and reference calls with customers at similar scale. Vendor selection can then continue as separate work if you decide to buy.
What you receive
- A costed comparison of all viable routes over three years, on consistent assumptions.
- A weighted decision matrix with the weights agreed by you before scoring, not after.
- An exit cost estimate for every bought option.
- A written recommendation with the reasoning, including the strongest argument against it.
- A record of any internal disagreement, so the decision is documented rather than relitigated in six months.
We record the case against our own recommendation
Every recommendation ships with the best argument on the other side. Decisions made without a documented counter-case get reopened the first time something goes wrong, and nobody can remember what was considered.
Output
Options comparison, three year cost model, recommendation with dissent recorded. Delivered in editable formats. Full IP transfers to you on final payment.
FAQ
Are you independent, given that you also build?
We do build, so the conflict is real and worth stating. Two things reduce it: the advisory work is priced and delivered as a standalone engagement with no build commitment attached, and we recommend buying in a meaningful share of cases. If you want stronger independence, engage us for advisory only and take the build elsewhere. That is a normal arrangement.
How long do vendor evaluations take?
Two to four weeks inside the advisory engagement for a shortlist of three to five vendors. Longer if procurement requires formal security reviews, which are usually the constraint rather than the technical evaluation.
What if the answer is a mix?
It often is, and that is a legitimate outcome. Buying the platform layer and building the differentiating logic on top is the most common recommendation we make. The model prices that route alongside the pure options.
Does this cover open source options?
Yes. Open models and open source tooling are costed as build routes with their real operating overhead included, which is where enthusiastic open source cases usually understate things.
Often paired with
AI Vendor and Model Selection
Structured evaluation of vendors and foundation models, benchmarked against your data rather than public leaderboards.
Read moreAI ROI Analysis and Business Case
A defensible financial model for an AI initiative, built on your numbers, with the assumptions exposed rather than hidden.
Read moreAI Strategy and Roadmap
A sequenced plan for what to build, in what order, at what cost, with the dependencies made explicit.
Read moreIs this the right engagement?
Tell us what you are trying to decide. If a different service fits better, or if you do not need us at all, we will say so.