Most AI advice is written by people who have never shipped an AI system. Keelhouse ships them.
The advice is usually written by consultants who don't build, or sold by tool vendors who don't run a business.
Keelhouse is different for one reason.
Have an AI idea for a real business? The Fast‑Lane will tell you within seven days whether it is worth building. Start it →
A consultancy runs a workshop, produces a slide deck, and hands the build to someone else. The strategy disappears into a sprint backlog no one owns. Six months later, nothing has shipped.
Individual tools get bought, one for sales, one for support, one for reporting. Each one promises to save time. Together they create a mess of disconnected apps, duplicate data, and no single view of the business.
An AI system runs, but nobody knows when to trust it or when to override it. Either the team stops using it, or worse, they stop checking it.
If any of those three sound like your business, the Fast‑Lane names which one and what to do about it. No cost →
Keelhouse spends time inside the business. Every process, every data source, every decision that matters, is mapped. Then each candidate AI project is scored against the four filters to find the ones actually worth doing.
The top projects get properly designed before anyone writes code. What data feeds it, what it connects to, what it decides on its own, and what a human still needs to sign off. This is the step most AI projects skip. It is why they go over budget.
The system gets built. Either directly, or with the operator's team under close supervision. No large‑firm handoff. No offshore team the owner never meets. Everything ships with tests, logs, and a way to roll back if something breaks.
The operator receives the working system, the manual for running it, and the rules for when to change it. Optional: a quarterly review to check it is still doing what it should.
The whole four‑stage engagement starts with one written intake. Free, seven days to a verdict →
These are the four questions asked before Keelhouse recommends building anything. If a project cannot clear all four, it does not get built. Try one.
Anyone can buy an AI tool. What competitors cannot copy is data unique to this business — customer behaviour, historical decisions, operational patterns. If the project does not use or generate that kind of data, it has no long‑term edge.
e.g. A retail chain has years of transaction data no one else has. That AI project is defensible. Public web data isn't.
Every project has to name the specific cost, revenue line, or time saving it will improve. If the impact cannot be measured, it does not get built. This kills a lot of "AI‑powered dashboards".
e.g. "Reduce time‑to‑quote from 3 days to 2 hours" is a valid target. "Improve customer experience" is not.
AI should propose. Rules should execute. A person should be able to see what happened and why, and change the rules when needed. Any system that makes decisions no one can review or reverse is a liability, not an asset.
e.g. AI that ranks leads and lets the sales manager tune the ranking is safe. AI that emails prospects with no visibility to anyone is not.
The best sign an AI system is working is that it is still switched on twelve months later. Most fail this test — built, used for a quarter, then quietly abandoned. Keelhouse designs every system for the second year, not the first launch.
e.g. The measure isn't the demo. It is whether the ops team still opens the system on Monday morning six months in.
Not sure if your idea passes the filters? The Fast‑Lane is where it gets tested properly. Free, seven days →
Keelhouse operates its own group of businesses. The list below describes each by shape, not name.
An AI‑assisted engine that evaluates opportunities against a defined set of rules. Runs on private infrastructure. Every rule change is human‑approved. Every decision is logged. The same pattern works for any business making frequent, rule‑based decisions where a wrong call is expensive.
An AI that watches how users move during a workout and corrects their form in real time. Live in a paying consumer product. Shows that a business can build a proprietary dataset inside a product customers already pay for, without asking them to hand over data separately.
A location‑based discovery product built to strict engineering standards. Every schema change reversible. Every deployment tagged and logged. Shows that a small team with the right discipline can build data infrastructure that behaves like it came from a much larger company.
The screening method Keelhouse uses to decide whether an idea deserves any real investment. Runs on three gates: founder time first, small paid tests second, formation and first contract only after both pass. Same framework offered externally as the Fast‑Lane.
A health telemetry platform designed from day one around a specific commercial buyer and a specific consent model. Shows that a data business can be architected for its exit before it launches, not after.
Venture names, sectors, and operator identity are shared only under mutual NDA. No mailing list. No newsletter. No cold sequences. The portfolio was built the same way any operator engagement starts — with a written intake and a verdict.
Three routes are available. The first is free. The other two are paid, and generally follow the first.
A short written intake. A written verdict. No cost.
The operator describes the venture, its intent, what they are trying to do with AI, and the one assumption the idea depends on. Within seven days, Keelhouse returns a short written diagnostic and the one‑week test that will show whether the assumption is right or wrong.
"A service that will not turn business away isn't a service any operator should trust. If it costs nothing to say no, Keelhouse can say no honestly."Request the Fast‑Lane →
For operators whose Fast‑Lane one‑week test has passed.
A written pilot design for the AI integration the test greenlit. Not a plan to build, a spec to run: who runs the pilot, what gets built in week zero, what gets measured, the numeric criteria for killing it early, what it costs, and the shape of a rollout if the pilot passes. A private document, delivered in two to three weeks, written to be handed to a pilot lead and executed without us.
A recruitment firm passed the Fast‑Lane one‑week test on a proposed AI layer that reranks incoming CVs. Their pilot design named one consultant, one live senior role, four weeks, and a specific place in an existing spreadsheet for the score to land. It set a numeric kill criterion tied directly to the test result, read at week three, week six, and week eight. It built in a sampling protocol so the CVs the AI ranked low still got read cold once a week, catching the candidates a machine would silently drop. It named three assumptions the operator was best placed to push back on before week zero. The document is roughly eight pages. Their ops manager could execute it without another call.
For businesses that have completed a Structural Audit and are ready to build.
Monthly retainer. Direct build supervision under the same four‑stage framework. Quarterly review, with a fresh audit at each quarter. Fee is quoted inside the reply.
All engagements begin under mutual NDA. Keelhouse does not keep a mailing list, does not run automated email sequences, and does not respond to unsolicited outreach outside the forms above. Fees are quoted in the reply, not on this page.
Free. Seven‑day turnaround. All submissions are treated as confidential and are covered by mutual NDA on request. Keelhouse will reply from matthew@keelhouse.co.