The questions we get in every first meeting.

No. There is no seat and nothing for your team to log into. The system sits underneath the tools your people already use, accumulates what happens as cases, forms patterns across them, and puts the result in front of the person who has to decide, inside the system they already work in.

A CRM records activity. It never forms a view. Memory is the layer that reads hundreds of recorded episodes and produces a claim: that quotes to this customer type stall at week three, that this discount shape precedes a loss. That claim carries its evidence, an owner and a confidence that decays. A notes field cannot do any of that, which is why four years of diligent CRM use still cannot tell you why deals die.

No. The method improves and your data never moves. What we carry between engagements is how to build memory well: the resolution logic, the patterns, the evaluation harness. Your cases, your judgment and your numbers stay inside your tenancy, are never pooled, and are never used to train anything.

No, and you cannot. The Factory is ours and never ships. What you receive is built from it and is yours on a perpetual licence, including the memory it has formed. There is no seat count and nothing switches off if you stop paying us.

It always is. That is the job. Half of what we run in the first meeting needs nothing from you at all, and discovery exists to find out how far the rest can be trusted before anything is built. If the answer is not far, we say so and you keep the map.

It is the qualification. A company that will not spend two weeks finding out will not spend six months building. The fee is credited in full against the rollout, so it costs nothing if you go ahead, and it stops both of us wasting a quarter if you do not.

Then it is reported as not clearing it. Both outcomes are in the contract. You paid for two weeks and a pilot, not for a rollout, and you stop knowing something you did not know before. We publish the ones that missed as well as the ones that landed.

Expiry. Every claim carries a half-life set by knowledge type, and confidence collapses when contradicting evidence arrives rather than averaging against it. Stale claims are held out of decisions until re-evidenced. Separately, every release passes evaluation gates, and lowering a gate is a versioned decision with a named owner rather than a setting.

Only after one of your people approves the write, per play, per field and per value threshold, after a dry run, with a full audit trail and a thirty-day rollback window. Our engineers hold a builder role and cannot approve a write.

Whichever the task needs. 408 across 51 providers, routed separately for extraction, reasoning, speech and embedding, and swappable without touching the play. If your policy requires a region, a sovereign provider or open weights inside your VPC, each step routes accordingly. The full catalogue with prices is on the models page.

We do not bill bodies by the month and we do not deliver decks. The output is a working system priced against the number it moves, and what it learns stays in your building. A consultancy's learning leaves with the team.

The system keeps running whether or not you renew anything. What you can renew is assurance: a named engineer inside your team, the quarterly review, definitions re-versioned as your business changes, confidence re-calibrated, models re-routed, and a floor on the metric with remediation included if it falls below.

A small senior team, and the same people from the first meeting to handover. There is no pyramid, no bench being sold to you, and no rotating cast of juniors learning your business at your expense. A deployed engineer sits inside your team, and someone who owns the number attends your review.

One engineer on our team can read your ERP schema, write the connector, build the resolution logic, design the evaluation suite and sit with your sales head to argue about what a stalled quote means. We do not hand work across a data team, a model team and a delivery team, because most of what goes wrong in these projects goes wrong in the handovers.

A consultancy is optimised to staff. We are optimised to finish. Our commercial model pays us for a number moving, not for months elapsed, which changes who we hire: people who have carried a quota or closed a month-end, not only people who have written code. Several of our team have run commercial functions themselves, which is why the first meeting is about your quotes rather than about our architecture.

Yes. For every sector we work in, we bring a subject matter expert who has actually run that function: a former service head for an installed-base build, a former broking operations head for an insurance book, a customs and freight practitioner for a lane margin engagement. They are in discovery, not just on the proposal, and they are the reason the definitions survive contact with your team.

Deep enough to write connectors against undocumented systems, build resolution engines that clear ninety-seven percent on your own labels, and run evaluation harnesses that block our own releases. We are also honest about our edges: we do not do research-grade model training or large-scale data platform engineering, and when a build needs it we bring a partner rather than learn on your budget.

Your team runs it. Handover includes the code, the memory, the definitions, the runbooks and two weeks of paired work with whoever will own it. Assurance is available if you want a named engineer to keep the number true as the business changes, and cancelling it does not switch anything off.

Few, deliberately. The model only works if senior people are actually in your build rather than supervising it, so we would rather turn work away than staff it thin. If we cannot start when you need to, we will tell you that in the first meeting.

Innovation labs and proofs of concept with no number. General purpose chatbots bought on productivity and defended on sentiment. Research nobody will sign for. People on a rate card. Anything where a model's output affects a person without a human in the path.