The Six A’s of AI™
The six stages every business climbs with AI, from asking a question to running on an edge competitors cannot match — and why the order you climb in matters more than the tool you buy.
A gadget, or a discipline?
What I see when I look at actual companies is not a shortage of tools. It is two postures.
As a gadget. You wait it out, or you poke at a chatbot and get a draft. You get activity: questions, a burst of curiosity, a few hours saved here and there. Nothing in how the business runs has actually changed. Activity feels like progress. It is not adoption.
As a discipline. You stop asking which model to buy and start asking where you are on the climb. AI has a place in the work. It has a next rung you refuse to skip. That is adoption.
At the centre of that discipline is a simple map: six stages, each starting with an A. Two things rise together as you climb. The first is scope, how much of your business AI touches. The second is trust, how much you let it run without you. Every stage raises both.
You cannot teleport to the top. Skip a rung and you do not move faster. You fall.
Six rungs. One climb.
Each stage raises how much of your business AI touches, and how much it does without you watching.
01. Ask
You ask questions and get answers. Useful, and where nearly everyone meets AI — but it is activity, not adoption.
02. Apply
Drafts, budgets, plans. The same tool gets dramatically better when you give it a role, your context, the outcome, and permission to push back.
03. Align
You write down how the work actually runs. The least exciting rung, and the one everything above it stands on.
04. Augment
A person still feeds it. A simple agent turns raw input into finished work that matches the process you wrote.
05. Automate
A trigger starts the job, the agent pulls what it needs from your systems, and the work completes without someone pressing go.
06. Advantage
Switch AI off and your customers feel the promise slip. That is the tell: not efficiency, but an edge nobody can buy on Friday.
01. Ask — AI as a smarter search
Saying “we use AI” almost always means this rung, and this rung is mostly activity. You open a chatbot, type a question, and get an answer back. It is a fine place to start and a bad place to mistake for a strategy.
One thing to understand before you lean on any of it: AI makes things up. The industry calls it a hallucination, which is a soft word for a real problem. It does not know when it is wrong, and a fabricated answer arrives in the same confident tone as a correct one. Treat what comes back as a smart first draft, make it show you a source you can check, and keep it away from legal, financial, and medical questions without a professional behind it.
Your move: pick one question you would normally Google, ask AI instead, then verify it against a real source.
02. Apply — AI does real work
Apply is where AI starts saving real hours: drafting, summarizing, cleaning things up. The gap between a mediocre result and a useful one comes down almost entirely to how you ask. Type “time for agenda for beginning of Q2” and you get the average of everything. Spend a minute setting it up and you get something you would actually use.
The recipe has four parts, and you run all four every time:
- Give it a role. You are not asking it to pretend. You are narrowing the expertise it reaches for. “Help me with my budget” leaves it standing in a library; “you’re a CFO with twenty years in small business” walks it to the right shelf.
- Give it your context. Who you are, what the business is, and the decision actually in front of you. The more of your real world it has, the less it has to invent.
- Give it the outcome. Name the finish line before you start running: the format, the audience, what “done” looks like.
- Make it push back. Left to its defaults, AI wants to be agreeable and will validate a shaky assumption in beautiful prose. The most valuable role you can hand it is a skeptic paid to find the hole in your thinking.
Then teach it your voice — not with a personality description, but with evidence: real writing, real emails, corrections made out loud and kept. The move that matters for a business is the next one: stop teaching it your voice and start teaching it the company’s, so a proposal from a new hire can sound like the firm. Be deliberate about what you feed it and where. Your voice is worth teaching; your secrets are not, on a platform that was not built to keep them.
03. Align — standards before scale
This is the rung people want to skip, and the one the rest of the climb stands on. Align is not an AI trick. It is the point where you decide how the work actually runs, write it down, and stop depending on whoever happens to remember. If the work still lives in one head, you have nothing honest to hand a new person, let alone an agent.
It will not feel like progress while you are doing it. Do it anyway, and do not confuse writing things down with writing good things down: bad processes scale too. AI just helps you execute a sloppy quote faster.
The unlock is to start from the mess. Do not sit down to “write an SOP” — you never will. Talk the job out the way it lives in your head, out of order, with the exceptions, and let AI turn that rambling into a first draft you can react to. AI sparks. It does not steer. Work through four questions out loud: what problem are we solving, who owns this, what does a good outcome look like with one real example, and what is unique about how we do this. Skip that last one and you will produce a process any competitor could copy.
Then read the draft with the person who actually does the work. When they say “that is not how we handle it when things go sideways,” the process is not done. That argument is the work.
A checklist is not an insult to expertise. Pilots do not use one because they are new; they use one because the work is complex and experience is not the same thing as recall. The list does not fly the plane. It keeps the work complete when the day is loud.
What the process has to contain
If it cannot survive a new hire and a bad week, it is not a process yet.
- Purpose: what this job is for, and what it is not for
- Trigger: what starts it
- Steps: the order of work, including the ugly parts
- Judgment calls: where a person decides, and on what basis
- Definition of good: one real example beats a paragraph of values
- Definition of done: what “finished” means before it leaves the room
- Exceptions: the cases that do not follow the happy path
- What AI is not allowed to decide: price exceptions, promises, legal language
- Owner: whose name is on it when it leaves the room
- Review: when you will look at it again, because the work will drift
A process that never gets revisited becomes folklore with nicer formatting.
04. Augment — bolt AI onto the work
Once the process exists, you can put AI on the job. Augment is not “let it run the company.” It is a simple agent bolted onto work you already know how to do. You hand it raw material — a data export, rough notes, last week’s numbers — and it hands back finished work that matches the process you wrote at Align. A person still feeds it. The busywork between input and output is what dies.
The best work to bolt on is high-volume and low-judgment. Plenty of businesses build a handful of these and stop, and that is allowed: agents can live at this rung for years and still pay for the time you spent on Align.
In our own shop: license audits
Every month we reconcile a customer’s Microsoft 365 licenses against who actually works there. Those two lists never match cleanly, and by hand the job used to eat days. We wrote the process first — what sources we trust, what counts as an exception, how we talk about savings without overselling them — then built an agent on it. We still hand it the files. It reconciles, flags exceptions, and produces the report in our wording. What took days now takes minutes. We still read it, and we still own the recommendation that leaves the building.
05. Automate — AI runs it on its own
This is the rung people picture when they imagine AI doing real work. Something triggers the job — an email lands, a form is submitted, a date arrives — and the agent pulls what it needs from the systems you already run, does the work against your process, and produces the result. Nobody fed it and nobody pressed go.
That is when “auto” earns the name. It is also when a sloppy process stops being an inconvenience and becomes an event, because speed multiplies whatever you point it at, and here you are pointing it at your live business.
In our own shop: triage
Triage was our low-hanging fruit and the place it would hurt most to get wrong. An agent now reads every ticket the moment it arrives: it classifies type and subtype, checks the right customer is attached, and judges urgency. When someone leaves a voicemail instead, the system transcribes it and opens the right ticket, so work that used to wait on a person starts the second the customer speaks.
We deliberately left assignment out. Which engineer gets a ticket looks like a small decision and is not: load, expertise, who already knows that customer. Until the process could say why this engineer and not that one, a person kept the call. Once the recommendations got boringly good, we let it assign. Not because we wanted to be fully automated, but because the process had caught up and the agent had earned the last step.
The coordinator did not get replaced. Triage used to be that job. We took the sorting, labelling and voicemail-listening and gave it to a process; what we did not give away is the part only a person does well. That person now spends the time talking with customers about how we could serve them better. Automate the grind you already wrote down, and keep the judgment and the care attached to a name.
06. Advantage — AI amplifies your edge
Advantage is where AI stops being a tool you use and becomes part of how your business wins, because it is now executing on the one thing only you do: the promise a customer chooses you for, kept more often, for more people, without falling back on one hero.
Everyone can buy the same tools. So the edge cannot be the subscription. The edge is how far you have climbed around it: the process you wrote, the judgment you refused to hand over too early, the data that is yours.
Two tests tell you whether you are here:
- The switch-it-off test. Turn AI off tomorrow, and who feels it? At Automate, mostly your team does, in extra manual work. At Advantage, your customer feels the promise itself get worse. The customer feeling it is the tell.
- The copy-it-by-Friday test. A competitor can buy the same model this week. They cannot buy five rungs of discipline, your processes, or your history of decisions. A model is a commodity. The climb under it is not.
A cabinet shop that turns a phone photo and a few measurements into a rendered design and a price on the first visit, while competitors take two weeks, has made speed part of the promise. A construction company estimating from job photos bids more jobs, and bids them better. In both, a customer feels it. AI did not invent the edge. It made the edge deliverable.
Your move: if you cannot name the promise AI is helping you keep, you are not at Advantage yet. You are somewhere lower on the climb, which is fine. Stay there until the work is real.
The security line
At Ask and Apply, the blast radius is whatever you typed. At Augment you start feeding the business into the machine. At Automate the tool can go get the business without asking.
Protect your information from the first prompt, and add stricter controls as AI gains access and authority. At Augment, decide what an agent may see, where those inputs are allowed to live, and who is allowed to run it. Keep a person between the output and the outside world: the agent drafts, a person releases. That is how you find out what it gets wrong while the mistake is still cheap.
Assume every inbound email and every ticket can lie to the process. An agent will treat text as instructions if you let it, so design it so a hijacked goal cannot complete a high-impact action. It can draft. It cannot send. It can flag. It cannot close the account.
Before you connect a live system
- Who owns this agent. A named person, not “the team.”
- What it may read. Name the sources, not the whole tenant.
- What it may write, send, or change. Draft is not the same as send.
- What it may never do: price exceptions, legal language, moving money, creating accounts.
- How you will know what it did. If you cannot reconstruct a Tuesday, you cannot trust a Tuesday.
- How you shut it off in five minutes. If nobody knows the switch, you do not have a switch.
Where is your business today?
Find the highest row that is genuinely, consistently true. Not “we tried it once” — this is how we operate.
| Stage | What it looks like |
|---|---|
| 1. Ask | You ask AI questions and get answers, like a smarter search engine. |
| 2. Apply | You use AI to do real work (drafts, a budget, a plan) guided by you. |
| 3. Align | You have written processes, so AI and your people know how the business works. |
| 4. Augment | Simple agents take raw input by hand and hand back finished work. |
| 5. Automate | Agents connect to your systems and run the work on their own. |
| 6. Advantage | Switch AI off and you would lose a real edge your customers feel. |
If a higher row is true but a lower one is not, you have not climbed higher. You have built on a gap, and that is the first thing worth fixing. The free Six A’s assessment walks through it in 3–4 minutes and gives you the gaps and a project to try, without asking for an email.
The prompt recipe: role, context, outcome, push-back
You do not need a hundred tricks. You need one recipe, applied every time.
What most people type
I need a budget worksheet for next year.
Not a bad prompt. It has an outcome and a point of view. It hands the AI no role, no numbers, and no format, so it answers from the average of everything.
The same request with all four parts
Role: you’re the CFO I don’t have, twenty years building budgets for small service businesses. Context: I own a managed IT firm, strong on operations, not finance, so walk me through your thinking. Outcome: build next year’s budget from zero — must-spend first, then tiers, shown monthly, annually and as a percent of revenue. Push-back: ask me the questions a good CFO would ask first, then challenge every number I give you.
The role focuses it. The context makes it yours. The outcome tells it what done looks like. The push-back turns it from a yes-man into the skeptic the work actually needs.
The full guide includes the complete prompts, plus a starter library: find your first five AI tasks, get a process out of your head, build your company voice, and run a pre-mortem before you launch.
Written by Sean Fullerton
I run NSN Management, a Tulsa-based managed IT and cybersecurity firm, and that shop is the lab this climb was tested in. The map is meant to work whether you ever call us or not.
The full field guide runs 26 pages: every stage in depth, the security line, the worked examples from our own floor, and the complete prompt appendix. No email required — it is a straight download.