Company
What you are actually choosing between.
There are four ways to get AI into a mid-market company. Buy a platform and staff it. Hire a consultancy. Hire an integrator. Or build memory against a number somebody owns. They differ less in technology than in what remains afterwards.
| Agentic AI platform | Consultancy | Systems integrator | Radexus | |
|---|---|---|---|---|
| What you get | A framework to configure and staff | A deck, then a team on a rate card | A migration onto their preferred stack | A system that accumulates your commercial judgment, tied to one number |
| What accumulates | Usage logs | Their experience, not yours | Licences | Cases, patterns and confidence, inside your tenancy |
| Who owns it after | Them. Stop paying, it stops. | Ambiguous. Usually the IP clause favours them. | You own the licences they sold you | You. Perpetual licence to your instance and the memory in it. |
| What moves the price | Seats and usage | Headcount and months | Scope and change requests | The value measured in discovery and proved in the pilot |
| Who names the number | Nobody. Value is described. | Sometimes, after the engagement | Rarely. Success is go-live. | Your CFO, in writing, before the build |
| Your stack | Adopts theirs | Untouched, and so is the plumbing | Replaced | Read in place, never migrated |
| Write-back to your systems | Configurable, usually unguarded | Out of scope | Yes, through the new platform | Gated by your named approver, per field, with rollback |
| Stale knowledge | Retrieved as if current | Leaves with the team | Not in scope | Expires on a half-life and is held out of decisions |
| Year two | Renewal, or it switches off | Another statement of work | Support contract on the platform | Assurance, held against the number, cancellable |
| Time to production | Months of configuration | Depends on the next phase | Twelve to eighteen months | Ten to fourteen weeks after discovery |
This table is our view of the market and is necessarily general. A good firm in any of the first three columns will differ from it. Ask them the questions in the right-hand column and see how they answer.
Twenty people each twenty percent faster is still an organisation that has learned nothing.
That is the test to apply to every proposal in front of you, including ours. At the end of the year, what does the company know that it did not know before, and where is that knowledge held?
If the honest answer is that some individuals got quicker, you have bought productivity. Useful, and it does not compound.
Factory