DECISION GUIDE
License, build, or consult?
There are three ways to get an enterprise AI platform, and the right answer depends on one question more than any other: do you intend to own this capability internally, or do you intend to use it? Licensing gets you to production fastest and keeps the platform someone else’s problem to maintain. Building in-house is the only path that ends with the capability genuinely yours. Consulting buys a decision, not a system.
Choose by intent
01
Choose licensing when the capability is important to your operation but is not the thing you sell. You want it running, governed, and maintained — not staffed.
02
Choose building when the capability is your competitive product, or when regulation makes external dependency unacceptable. Budget for the team permanently, not for the build.
03
Choose consulting when the real problem is that you do not yet know what to build. Buy the decision, then re-run this comparison with it.
04
The most expensive path is starting one and switching after nine months, which is what happens when the decision is made on initial cost alone.
What each path actually costs after year one
Initial cost is the least informative number in this decision, because the three paths diverge after the first year rather than during it. Licensing has a flat, predictable ongoing cost and no hiring risk; the exposure is vendor dependency and whatever the exit terms say. Building has a low marginal cost per feature once the team exists — and the team is the cost, permanently, including the replacement cost when its key engineer leaves. Consulting has no year two: whatever was recommended either got built or did not.
The comparison below assumes an equivalent scope in all three columns. Where a column is empty, that is the point — it is not a gap in the table.
The questions that actually decide it
Is this capability something you sell, or something you run on? If a customer will ever pay for it directly, build. If it makes your existing product or operation better, license.
Can you hire and retain the team? Not "can you afford the salaries" — can you offer work interesting enough that the people who can build this will stay. A half-staffed platform team produces a system nobody wants to be on call for.
What does your regulator require? Some sectors make external data processing genuinely impossible, which decides the question regardless of economics. Check this first; it is the only input that can eliminate an option outright.
What happens if you want out? Licensing terms should specify what you take with you — data structures, prompts, indexes, governance rules. If a licensor cannot answer that clearly, the dependency is worse than it looks.
Three paths, compared
| Technical licensing | Build in-house | Consulting | |
|---|---|---|---|
| Upfront cost | Low to moderate | High | Moderate |
| Ongoing cost | Predictable, flat | Permanent team cost | None after delivery |
| Time to production | Weeks | 6–18 months | No system delivered |
| Internal capability required | Operator, not builder | Full platform team, permanently | Enough to judge the advice |
| Control over roadmap | Shared with licensor | Total | None |
| Who carries delivery risk | Licensor | You | You |
| Best fit | Capability supports the business | Capability is the business | Direction is still unclear |
Should we license an AI platform or build our own?
License it if the capability supports your business but is not the product you sell; build it if the capability is the product, or if regulation forbids external dependency. The deciding factor is not budget — it is whether you will still be staffing this team in three years.
Where this leads
PVL Nexus and technical licensing
What is licensed, layer by layer, and what a licensee can operate independently.
What is an AI Knowledge Infrastructure Company?
Why handover — not lock-in — is the defining test of this category.
Enterprise RAG that stays maintainable
The maintenance load you are choosing to own or to delegate.
Still between two columns?
Tell us the capability, the regulatory constraint, and whether you can staff a platform team. That is usually enough to eliminate one column in the first conversation.