An AI executive assistant for every person in your business
Develop an assistant around each person's role, using your business knowledge and your way of working. You guide it, review its work and refine it over time.
Your first AI executive assistant
A free introductory webinar series for business owners and their teams. Once ten businesses register interest, Peter will schedule the first session. We'll email you the details. Dates will follow.
Three proposed topics:
- Your business brain
- Your individual assistant
- One process first
Having trouble with the form above? Open registration form.
Everyone has AI now. Almost nobody has a method.
Most businesses are in the same place. You pay for AI. People open a chat window, paste something in and hope. Some of it is useful. None of it is reliable, and none of it is yours.
That is not a failure of the tool. A chat window is very good at answering a question. It just doesn't know your prices, your tone, your standards, who signs off what, or how your business actually does the thing you're asking about. It has no way to know any of that, and no memory of it tomorrow.
So the answer isn't a cleverer prompt, and it isn't another subscription. It's a structure.
And the drift has a cost that doesn't show up on an invoice. It shows up like this:
- —The same task comes back five different ways, because five people worked out five ways of asking, and none of them wrote it down.
- —Nothing compounds. A good result on Tuesday teaches the business nothing, because there is no place a good result goes.
- —Your best people get good privately — and that skill leaves with them.
- —Checking takes as long as doing. Nobody can tell at a glance whether a draft was built from your actual rules or from a plausible guess, so everything gets read end to end anyway.
- —You cannot answer a simple governance question. If someone asks what your business allows AI to touch, the honest answer is that nobody decided.
Meanwhile the licences renew.
Why buying an agent doesn't solve it
The market's answer right now is to sell you an "AI agent" as a shortcut. An off-the-shelf agent is built around an average company's process — and the way your business actually works is usually the whole reason you're competitive.
Here are five things an off-the-shelf agent cannot give you without a method.
| What you need | Why a subscription can't give it |
|---|---|
| It knows your business | Your facts, rules, tone and standards aren't in it, and there's nowhere to put them |
| It comes out the same every time | Nothing has defined what "done" looks like for your job |
| Someone is accountable | There's no record of what it used, or who approved the result |
| You can change your mind | Your process ends up locked inside somebody else's product |
| Anyone can be taught it | There's no method to teach — only a login |
Buying an agent is a purchase. Controlling AI is a capability. They cost about the same to start and they end up nowhere near each other.
The method: five steps, four rules
This is the minimum structure required to make AI consistent, checkable and teachable inside a small business. Less than this and you are hoping. More than this and nobody follows it.
This is how LOOKUP runs its own business, and how we set it up in clients' businesses. It fits on one page on purpose — a method nobody can be taught is not a method.
The five steps
- 1Write the business down, once. The facts, the rules, your tone of voice, your standards. Written once, in plain documents, in one place. Every fact has exactly one home, so nothing goes stale in three copies.
- 2Every job gets a brief. Before anything runs: what "done" looks like, what the assistant is allowed to use, and who signs it off.
- 3The assistant works inside those rules. It sees only what that job needs — not everything you own. Small, focused and checkable beats big and impressive.
- 4A person checks every stage. The assistant prepares. People decide, send and sign. Every time, not just on the risky ones.
- 5Everything is saved in plain documents. Files you can open, read and edit without us. You can always see how a result was reached.
The four rules underneath
- One source of truth. Every fact is written down in exactly one place. Copies go stale; links don't.
- A person at every stage. The assistant never sends, signs or commits to anything.
- Prove it once. A process only gets scaled after one real example works, measured against the way you do it today.
- The right tool first. Routine, predictable steps go to ordinary software. AI is kept for the judgement.
Because the method lives in your documents rather than inside a product, you are never locked into one AI vendor. Change the model, keep the method.
The method is the asset, not the assistant
Five steps and four rules. That is the whole thing, and the size is the point.
A method that only works when one clever person runs it isn't a method, it's a dependency — and it walks out the door when they do. This one can be taught to anybody in your business, in an afternoon, and it holds whether the person applying it is your most experienced hand or someone in their first month.
That is also where the real return sits. Not in replacing anyone, but in the fact that a good person with an assistant that already knows your standards produces more, and produces it consistently, than the same person starting from a blank page every morning.
What it looks like on one real process
Here is the method run end to end on something ordinary — our own process for turning a first meeting into a proposal.
Step 1, written down once. What we sell, how we price it, the tone we write in, what a proposal must contain, and who is allowed to approve one. Written once, not re-explained per job.
Step 2, the brief for this job. Produce a proposal for this prospect. Done looks like a deck and a document that agree with each other, no claim in either that has not been confirmed, and a named person's approval before anything leaves the building.
Step 3, working inside the rules. The assistant gets the meeting transcript and the client brief. Not the whole business, not other clients, not last year's pricing — just what this job needs to be done properly.
Step 4, a person at every stage. The client brief is confirmed by a person before the meeting. The summary is corrected by a person afterwards. The proposal is approved by a person before it goes. Three checkpoints, none of them optional.
Step 5, saved as plain documents. The transcript, the brief, the proposal source. All openable, all editable, none of it locked in a tool.
What that buys. The deck and the document are generated from one source file, so the two cannot contradict each other — which is the failure that quietly costs you deals. And the whole run, from final answers to a reviewable proposal, took an afternoon.
Nothing in those five steps is clever. That is rather the point: it is repeatable, and it can be handed to someone else.
What you actually get, and how it is checked
A method you cannot inspect is a philosophy. So here is the machinery, in plain terms.
The documents you end up owning
These are the artefacts, not the sales process. They live in your systems, in formats your own people can open and edit.
| Document | What it does |
|---|---|
| The business record | Your facts, rules, tone and standards, written once. The thing everything else reads from |
| A job brief, per process | What "done" means for that job, what the assistant may use, and who signs it off |
| The allow-and-disallow register | What AI may touch and what it may never touch. The answer to the governance question, in writing |
| The approval map | Which named role approves which output. No "someone should check it" |
| The process record | How the work is done today — observed, written down, and read back to you to confirm before anything changes |
| Output templates | The proposal, report or summary built to your rules, so the format is not re-decided each time |
| The change log | What changed, when, and who approved it |
| The review schedule | When each of the above is next checked for drift |
| The handover guide | How to run it without us |
The checkpoint, defined
"A person checks it" is not a control unless you can say what is checked and against what. So:
- What is checked: the output against its job brief — does it meet the definition of done, and did it use only what it was allowed to use.
- Against what standard: the brief and the business record. Every claim traces to a source. Anything not confirmed is marked unconfirmed rather than filled in.
- By whom: the named role on the approval map. Not whoever is free.
- Where it is logged: in the process record, with the version that was approved.
Some checkpoints are mechanical rather than human, and those are the strongest ones. Our own website pages, for example, must score 100 on a fixed technical audit before they can publish. It is pass or fail, nobody's opinion is involved, and it has never been waived — including for the page you are reading.
Who owns it on day 30
One named person on your side owns the business record. They approve changes to it, and changes go through the change log. If nobody owns it, it goes stale within a quarter — which is the most common way this fails, and why we ask for the name before we start rather than after.
Three ways this goes wrong, and what we do about each
We would rather tell you these than have you discover them.
| What goes wrong | Why | The countermeasure |
|---|---|---|
| The written record goes stale. Prices change, a process moves, nobody updates it — and the assistant is now confidently working from last quarter | Updating documentation is nobody's favourite job | Every fact has exactly one home, so there is one place to change. Each document carries a review date. The owner on day 30 is the person who holds it |
| People stop following the brief when they get busy | If the method is extra work, it loses to the deadline every time | The brief has to be the shortest path to a finished job, not a form to fill in first. If people are working around it, the brief is wrong — we fix the brief, not the people |
| Shadow use. Someone pastes something sensitive into a chat window outside the process, because it was quicker | The official route was slower or did not cover their job | The allow-and-disallow register makes the boundary explicit, and the fix is to give that person a better default rather than a policy reminder |
None of those three are hypothetical, and none of them are fully solved by software. They are solved by the record, the owner and the checkpoint — which is why the method has all three.
Where this sits: stages 06 to 08
This isn't a separate idea bolted onto the side of your business. It is the AI extension of the LOOKUP Business Modernisation Framework — the eight-stage method we use to sequence any modernisation work.
The AI Executive Assistant Method is stages 06, 07 and 08. The first five stages come first, and there is a reason — two of the Framework's own rules are governance before AI and standardisation before automation.
When you're ready, and when you're not
Most of the industry will tell you the answer is yes, today. It usually isn't, and we would rather say so before you spend anything.
Wait if any of these are true:
| Signal | Why it blocks |
|---|---|
| The process only exists in one person's head | There's nothing to write down in step 1 |
| Three people do the same job three different ways | You'd be automating an argument |
| Nobody owns the exceptions | There's no one to sign step 2 |
| Your data is somewhere nobody can reach | There's nothing safe for step 3 to read |
| No one has decided what AI is allowed to touch | Governance before AI |
None of those take long to fix. They just have to be fixed first, and fixing them is worth doing whether or not you ever add AI.
And fixing them is work we do — it is stage 06, Prepare. Start with the governance check and we will tell you which of these blockers you actually have, before anyone talks about scope. Deciding not to go ahead is a legitimate outcome.
What each person actually gets
An assistant per person, built around that person's job. Nobody is replaced — the point is to give the work back to them, not away from them.
| Role | What their assistant does | What it never does |
|---|---|---|
| Owner | Prepares the decision — the numbers, the options, the risks, in one place | Makes the decision |
| Operations | Stops the rebuild of the same report every Monday | Sends anything |
| Sales | Turns a meeting into a brief, a proposal and a follow-up, all to one standard | Commits to a price |
| Finance and admin | Works through the repetitive checking and flags only the exceptions | Approves a payment |
| Whoever holds the know-how | Writes their knowledge down once, so it belongs to the business too | Replaces them |
What that looks like on a Tuesday
Your operations manager. The weekly report that takes two hours to rebuild is drafted before they sit down, from the same sources they would have opened, in the format the business already agreed. They spend their two hours on the three numbers that look wrong instead of on assembling the other forty that don't.
Your salesperson. The meeting they had at ten is a written summary by eleven, with the follow-up actions pulled out and dated. The proposal is drafted against your real pricing rules rather than their memory of them. They still decide what to offer, they still make the call, and they still sign it — but they arrive at that point with an afternoon back.
Neither of them was replaced. Both of them stopped doing the part of the job nobody ever wanted.
Two examples
A Sydney services business: answers at lunch, proposal by five
The client answered our last open questions at 12:36. The proposal deck and the proposal document were ready for review at 17:46 the same day.
A person approved the exact version the client saw. Anything we hadn't confirmed said "to be confirmed" — nothing was guessed to to fill a gap.
A Sydney logistics company: finding the work AI should do
They were checking 30 to 35 contractor invoices a week, by hand, against about 50 printed reports and more than 2,000 container numbers.
We didn't start with AI. We watched the real process, read it back to them to confirm we had it right, designed a right-sized fix, reviewed the risks on both sides, and then they decided.
The proposal estimates about 24 hours a week handed back to their team. That is a proposal estimate, and it stays an estimate until it is running.
Six situations you'll probably recognise
These are the six places the work usually is. They are written as situations, not case studies — they are here so you can find yours, not as claims about what we've delivered for someone else.
1. The meeting that never turns into anything
The situation. You have a good conversation on Tuesday. The notes are in someone's head or half a page of shorthand. By Friday the three things you agreed have become one thing, and nobody is sure who owns it.
What the assistant does. Turns the conversation into a written summary the same day, pulls out every commitment with an owner and a date, and drafts the follow-up in your business's voice.
What stays yours. You decide what was actually agreed, and you send it.
2. Proposals and documents that come out differently every time
The situation. Three people write proposals. They use three templates, two out-of-date price lists and whatever the last one said. The client sees the inconsistency before you do.
What the assistant does. Builds the document against the rules you wrote down once — your current pricing, your standard terms, your tone — and generates the deck and the document from a single source so the two cannot disagree.
What stays yours. The price, the commitment, and the approval before it leaves.
3. The checking nobody wants to do
The situation. Somebody spends a day a week comparing one list against another — invoices against dockets, orders against deliveries, a report against a system. It is not hard. It is just long, and being tired at 4pm is when the mistakes happen.
What the assistant does. Works through the routine comparison and surfaces only the exceptions, with its reasoning attached so the exception can be checked rather than taken on trust.
What stays yours. Every exception, every judgement call, and every approval to pay.
4. The process that lives in one person's head
The situation. One person knows how the month-end actually works, or how that one client likes things done. They are very good. They are also the single point of failure, and they cannot take leave without something slipping.
What the assistant does. Step 1 of the method is the fix — their knowledge gets written down once, as the rules the assistant works to. Now the business holds it too.
What stays yours. They still run it. They just stop being the only copy.
5. Marketing that stops and starts
The situation. Marketing happens when someone remembers — usually after a quiet week or a sales dip. The voice changes depending on who wrote it. The website is a year out of date because updating it is nobody's actual job.
What the assistant does. Writes to a rulebook — your voice, your claims, your standards — so output is consistent regardless of who initiated it, and the routine measurement happens on a schedule instead of when someone remembers.
What stays yours. What gets said about your business, and when it goes live.
6. The decision you want pressure-tested
The situation. A real call — a hire, a price rise, a new location, a big client's demand. You want it argued with before you commit, and the people around you are too close to it or work for you.
What the assistant does. Builds the case against your decision using your own numbers, constraints, policies and past outcomes, so the weaknesses surface before the money does.
What stays yours. All of it. This is the clearest case of the rule — it prepares the thinking, you make the decision.
Which one of those six is costing you the most hours right now? That's the question we'd start with, and most people can answer it in about ten seconds.
How we start
Nothing about AI or savings gets promised until the evidence is in.
- 1The free 60-second AI Governance Check. Instant result. You don't need to give us any details to see your score.
- 2A strategy session. Where the hours are actually going.
- 3Discovery on one real process. We watch the real thing and read it back to you.
- 4We prove it on one example, measured against the way you do it today.
- 5You decide what gets built.
What this is not
Being clear about the edges saves everyone a wasted meeting.
It is not a headcount reduction plan. Every example on this page ends with the same person still doing the job, with more of their week available for the part that needs them. If the goal is fewer people, this is the wrong method and we would say so.
It is not an autonomous agent running your business. The assistant prepares. It does not send, sign, approve, pay or commit. That is a rule, not a current limitation we expect to lift.
It is not a guarantee of hours or savings. Where we quote a number on this page it is labelled as an estimate, and it stays an estimate until the work is running and measured.
It is not a product we are selling you. The method lives in your own documents. If you change AI vendor next year, or decide to run this without us, it comes with you.
Common questions
No. The method is built the other way around: the assistant prepares the work and a person decides, sends and signs. That's a rule of the method, not a reassurance. What changes is how much of someone's week goes on preparation rather than judgement.
Yes, for preparation work — drafting, summarising, checking, pulling information together, getting a document to a consistent standard. It is not reliable as a decision-maker, and we don't set it up as one.
Write your business down once, give every job a brief that says what "done" looks like and who signs it off, let the assistant work only inside those rules with only what that job needs, have a person check every stage, and keep everything in plain documents you can open. That is the whole method, and it is the five steps above.
The best one is the one built to your rules. The method works across the major AI products, which is deliberate — your process shouldn't be locked inside a product you might want to leave.
No. It is generally easier in a smaller business, because there are fewer versions of each process to reconcile first.
No, it makes them worth what you pay. The method sits above whichever AI you already have — it supplies the rules, the briefs and the checkpoints those products have no way to know on their own. You keep the licences you have.
It is five steps and four rules, and it is designed to be handed over. The written record lives in your documents, in your systems, in a form your own people can read and change. If you decide to run it yourself, you can.
Nothing much, and that is the whole argument for a method. Because the rules and the briefs are written down rather than held in someone's head, the next person inherits a working system instead of starting again.
The governance check takes 60 seconds. A discovery on one real process is a matter of weeks, not months, and it is deliberately scoped to one process so there is something to judge at the end of it. We prove it on a single example before anyone talks about scaling.
Step 3 of the method exists for this. The assistant is given only what a specific job needs, rather than access to everything, and every output is approved by a person before it goes anywhere. What AI is allowed to touch is a decision your business makes and writes down — that is the "governance before AI" rule, and it comes before any of this.
Related services
AI Implementation
Ready to put AI to work across your business, not just one process?
AI & Workflow Automation
Automating the routine work the method identifies as rule-based.
AI Consulting
Strategic advice on where AI fits and what has to come first.