How an Engagement Runs
Four phases, and you decide at the end of each one. What each phase delivers, which is free and which is paid, and what you keep either way.
This is a route, not a turnstile. Most firms enter where they actually are rather than at the beginning. Some start with a plan. Some already have systems built and want them hardened. Some want the role filled from day one. Each phase ends in a decision, and nothing here assumes the phase after it. The next one has to be earned by the quality of the last.
The whole arc at a glance: four phases, and the decision that ends each one.
The Conversation
A recorded conversation, and it produces something.
You walk me through how the work actually moves and where it gets stuck. I ask questions rather than present. This isn't a session where I show you what other businesses are doing with AI, because the useful direction of travel is the other way. With your permission it's recorded, so the thinking afterwards comes from what you actually said rather than what I managed to write down while talking. If you share your screen, keep client names, records and logins off it, because the recording keeps whatever it sees.
You don't need to know what to build. Your job is to describe the outcome you want. Engineering the path there is mine.
A written proposal for the next step, with its price, that you can accept, decline, or take elsewhere. No charge, and no obligation either way.
The Plan
Paid discovery that ends in a prioritised plan you can act on, not a document that sits on a shelf. It starts where you want to focus and widens only as far as the work justifies.
The real workflow, mapped
How the work moves in practice, which is rarely how the org chart describes it.
Where things stand today, measured
A before-number for what we're fixing, the target you want, and a plain sum of what it's worth against what it costs to build and run.
A prioritised plan you keep
Yours either way. On a single workflow that's a short written plan; on a larger engagement it becomes a roadmap living where your team already works, never a file only I can open. Everything out of scope is listed alongside it, so the good ideas that surface don't quietly get lost.
What to automate, what to train, what to leave alone
Including the things that shouldn't be built. Saying so is part of the job.
The access and governance design
What each system can see, what it does on its own, what stops for a human, what gets logged.
You decide whether to build. The build is earned on the quality of the plan, not assumed by it.
The Build
The plan becomes engineered systems inside your business. Working pieces land as they're ready. The first is a small pilot, checked against the before-number before the rest goes ahead.
The systems themselves
Automations, control planes, ask-able libraries, integrations, engineered for your operation rather than assembled from widgets.
The guardrails around them
Monitoring, health checks, and escalation paths, so a failure surfaces instead of going quiet.
Documentation a non-technical person can run alone
Written so the system lives in your business rather than in one person's head.
Enablement on your team's real work
Not a generic training session. Their actual tasks, with the guardrails written into the tools.
Signed off, then 30 days of fixes
Each piece has written criteria for done and your sign-off. Anything that falls short of them in the 30 days after, I fix at no charge. New work is priced first.
You decide what happens next: run it yourself with the handover guide, keep me on for monthly upkeep (moving to a new AI model when a provider retires one, watching the alerts, small fixes), or the seat.
The Seat
An embedded AI-integrator and operations seat, priced as a seat rather than a scoped project.
For organisations that would rather have the role filled than fill it. Not a helpline and not a check-in call. What the seat actually does:
Listens to you and your team, and turns what's said into organised, sequenced, owned work.
Makes sure requirements are genuinely understood on both sides before anything gets built, rather than discovered broken at delivery.
Drives execution daily: chases action items, surfaces blockers, holds owners to their own estimates.
Owns the operational communication (status, notes, follow-ups) so you don't have to ask where things stand.
Makes decisions inside scope, says so immediately, and owns the outcome.
Spare capacity goes to your own projects rather than being billed away.
How one person holds a seat this size
AI agents do the production work. I architect, direct, review and govern what they produce, which is the part that can't be delegated, and the reason the output holds up.
Not a stage you pass through. The conditions the whole thing runs under.
Governance is the starting point, not a finishing touch
Human-in-the-loop by design. Draft-only and read-only by default, so the normal state of a system is that it prepares work and waits for a person. Anything that runs on its own, like a reminder to your own staff, is named in the plan and switched on only with your yes. Health checks, escalation paths, audit trails, and adversarial review before anything goes live. It's worth knowing what a system won't do before hearing what it will.
You own it, and you could replace me tomorrow
Your systems, your files, your documentation, your accounts. Everything is written down so somebody else could pick it up, and continuity is audited rather than assumed, including my own dependency in the map, treated the same as any other risk. No lock-in, and nothing that only works while I'm reachable.
Records about people wait for a written agreement
If the work would touch health, client or other personal records, nothing touches them until we've agreed in writing how they're handled. I don't give legal or compliance sign-off; your own adviser does.
Usually mixed rather than chosen: different parts of one operation want different answers.
Train your people
Where judgment and relationships matter. They stay the driver, with better tools and clearer guardrails.
Automate it
Where the work is repetitive and high-volume. Steps you've approved, like reminders, run on their own. Anything that leaves your business or changes a record waits for a person's yes.
Run it
Where you'd rather not run it at all. That's the seat.
Phase 0 costs nothing and ends in a priced proposal for the next step. If the proposal isn't right, you've lost a conversation. Tell me the one workflow eating the most time right now and what tools you're using today, and I'll tell you straight whether this is even the right move.
artificialintelligently.com
