Our honest position
We build custom AI systems for a living, and most organisations asking us to should buy something instead. The build case is real but narrower than the market suggests — and knowing which side you are on saves months.
This is a decision framework rather than a pitch. It includes the conditions under which we would tell you not to hire us.
Three options, not two
| Buy a product | Configure a platform | Build custom | |
|---|---|---|---|
| Time to value | Days | Weeks | Months |
| Upfront cost | Low | Moderate | High |
| Ongoing cost | Per seat, forever | Platform plus config | Hosting plus maintenance |
| Fit to your workflow | You adapt | Mostly fits | Exact |
| Who maintains it | Vendor | Vendor plus you | You |
| Data control | Their terms | Their terms | Yours |
| Differentiation | None — competitors have it too | Some | Potentially real |
The middle column is the one most teams skip. Configuring a platform with custom actions and integrations delivers a large share of custom-build value while the vendor keeps maintaining the hard parts. Rule it out deliberately before committing to a build.
Buy when…
- The task is standard — transcription, summarisation, generic writing assistance, coding help.
- Your workflow is not unusual for your industry.
- Team size keeps per-seat pricing reasonable.
- You need it working this month.
- The capability is not a competitive differentiator for you.
- You have no one to maintain a custom system.
Build when…
- Your process is genuinely different, and products force you to work their way.
- The AI must sit inside a product you sell — most vendors prohibit embedding.
- Data residency, confidentiality or regulatory constraints rule out hosted products.
- Per-seat costs have become a significant annual line item.
- You need deep integration with internal systems no product supports.
- The capability itself is part of your competitive position.
If a product exists that does 80% of what you want, building the remaining 20% from scratch is rarely the efficient path. Buying the 80% and integrating the 20% usually is.
The arithmetic, done honestly
Compare over five years, including the parts people omit:
| Buying | Building |
|---|---|
| Per-seat licence × users × 60 months | Build cost |
| Annual price increases | Maintenance at 15–20% of build, annually |
| Integration work to connect it | Model API usage |
| Configuration and admin time | Hosting and infrastructure |
| — | Ongoing evaluation and prompt maintenance |
| — | Internal owner's time |
The two rows with no counterpart are where build cases go wrong. Evaluation maintenance and an internal owner are not optional for production AI — models change, data drifts, quality degrades quietly. Comparisons that omit them make building look cheaper than it is.
The questions that decide it
- Have you actually trialled three products with your real data? Not demos — a genuine trial. Many build decisions are made before this.
- What specifically does no product do? If you cannot name it precisely, you are not ready to build.
- Is that gap worth the cost difference? Sometimes adapting your workflow is cheaper than building around it.
- Who owns it in year two? If the answer is nobody, buy.
- Would this differentiate you? If competitors could subscribe to the same thing, building is unlikely to pay.
The hybrid most organisations land on
- Buy for commodity capability — transcription, coding assistance, generic drafting.
- Configure a platform where your process is nearly standard.
- Build only the one or two things genuinely specific to your business.
That allocation puts your engineering budget where it produces differentiation rather than where it recreates something you could rent.
When we would tell you not to hire us
Stated plainly, because a partner who cannot name these is selling:
- Your requirement is standard and a good product already exists.
- You have not trialled the obvious alternatives with real data.
- Nobody internally will own the system after handover.
- Your documentation and data are poor — fix that first; it may resolve the problem entirely.
- The volume does not justify the build, however well it goes.
Weighing this decision? Tell us what you have already tried — if a product would serve you better, we will say so. See our AI agent service, partner selection questions, and the general build-vs-buy guide.