The owner or managing director should own AI decisions: what it's used for, what it costs and which risks are acceptable. One named person, often an operations or office manager rather than the most technical member of staff, should run it day to day. Every AI workflow also needs its own owner, usually whoever's work it changes.
Ownership is split because the jobs are different. Deciding needs authority. Running needs time and steady attention. Each workflow needs someone close enough to the work to notice when the output is wrong. In a five-person firm, that might mean the owner signs off tools and spending, the office manager looks after accounts and settings, and whoever handles bookings owns the AI that drafts booking replies.
A common mistake is to treat "who owns AI?" as a technology question and give it to whoever seems most comfortable with computers. That person may be a good builder, but ownership is mostly about decisions and follow-through: saying no to a tool, cancelling a subscription, telling a customer an answer was wrong.
Here's how that mistake tends to play out. An illustrative nine-person architecture practice handed "AI" to its most tech-minded junior architect. She built good things: a drafting assistant for planning statements and an automation that filed site photos. Eighteen months later the partners discovered three AI subscriptions on her personal card, reclaimed through expenses, one of which had moved to annual billing without anyone approving it. The planning-statement instructions lived in her own chat history. When she was moved onto a demanding site project, nobody else could run the drafting assistant, and nobody had the authority, or the knowledge, to decide whether the subscriptions were still needed. The fix wasn't to take the work off her. It was to make a partner the decision owner, the office manager the AI lead, and her the workflow owner for the two things she'd built.
Four roles every AI setup needs, whatever your size
In a very small firm one person may hold three of these. That's fine, as long as you know which hat they're wearing when a decision comes up.
| Role | Owns | Typical person | Time a month |
|---|---|---|---|
| Decision owner | Which jobs AI does, the budget cap, what AI must never do, sign-off on new tools, serious incidents | Owner or managing director | 1-2 hours |
| AI lead | Accounts and seats, the list of workflows, the prompt library, the usage policy, monthly cost and quality checks, starters and leavers | Operations, office or practice manager | 3-8 hours |
| Workflow owner | One workflow's instructions, reference files, checking step and numbers; first to hear when it goes wrong | The person or team lead whose work it changes | 1-2 hours per workflow |
| Reviewer | Checking output before it reaches a customer, or sampling it afterwards | Staff who do the work | Built into the job |
Outside help, whether an IT provider or a consultant, can set up accounts, security settings and automations. It shouldn't own any of the four roles, because none of those people will be in the building when a customer receives a wrong answer. What an AI consultant can't do for you explains where that boundary sits, and if someone external builds workflows for you, check who owns what they build before work starts.
How the split changes with size
Three illustrative businesses show the same four roles at different scales.
A four-chair barber shop
The owner is both decision owner and AI lead. One barber, who already runs the shop's social media, owns the caption-drafting workflow; the owner owns booking-message replies. Everyone checks their own output. The main risk is that everything lives in the owner's head and personal phone, so two habits matter: every AI and automation account uses the shop's email address, and a one-page list shows what each tool does and what it costs.
A 12-person optician practice
The practice director is decision owner. The practice manager is AI lead, with about four hours a month set aside. The lead dispensing optician owns the workflow that drafts replies to frame and lens enquiries. One optometrist is a named reviewer with a veto over anything patient-facing that touches clinical content, and the rule that clinical questions never go through AI drafting is the director's decision to change, nobody else's.
A 35-person garden centre
The owner holds a 20-minute AI slot in the monthly management meeting and decides from there. The operations manager is AI lead, at around eight hours a month. Department heads for plants, the café and the web shop own their own workflows. An outside IT provider manages accounts and security settings, but takes instructions from the AI lead and doesn't approve anything. At this size, the formal side of approvals starts to matter; AI governance for a small business covers who decides, approves and checks in a bit more depth.
Two partners and nobody else
The smallest case is the hardest, because every role lands on the same one or two people. Take an illustrative husband-and-wife bakery café where one partner runs the counter and the other does the accounts, ordering and website. The second partner is naturally the AI lead and owns both workflows (supplier-email summaries and menu descriptions), but the decision owner role should still be explicit: both of them agree the monthly cap and the "never" list, written on one line each, so neither can quietly add a $30 tool the other finds on the bank statement. The reviewer role is the real gap when there's nobody else. Their answer was a simple swap: whoever didn't write it reads any AI-drafted allergen or menu text before it goes up, every time.
Whatever your size, four triggers tell you it's time to name the roles formally rather than leaving them implied: a third AI workflow goes live; AI output starts reaching customers; a workflow handles health, financial or children's data; or headcount passes about ten, when the owner can no longer see everything that happens day to day.
Where the AI lead's hours actually go
Owners often underestimate the running role, then wonder why it drifts. An illustrative month for an AI lead looking after four workflows might look like this:
| Task | Time |
|---|---|
| Check the bills against the tool list; spot unused seats | 45 minutes |
| Sample ten outputs per workflow with each workflow owner | 1.5 hours |
| Update the prompt library and reference files after changes | 1 hour |
| Set up starters, remove leavers | 30 minutes |
| Answer staff questions and fix small problems | 1.5 hours |
| Assess one new tool request for the decision owner | 1 hour |
| Quarterly review preparation (averaged per month) | 30 minutes |
That's about seven hours. In a month with an incident or a new workflow launching, it can double. If the person named can't find that time, either reduce the number of workflows or move some of their other duties elsewhere; don't simply hope.
Price those hours too, because they often cost more than the tools. If an office manager's time costs the business about $28 an hour, seven hours a month is roughly $196, against perhaps $100 a month for four business AI seats. An owner who only watches the subscription line is looking at a third of the real running cost. That isn't a reason to skimp on the role; it's the reason the monthly check has to prove each workflow earns its keep.
The "assess one new tool request" line deserves an example, because it's where the AI lead and decision owner meet. At the garden centre above, the café manager asked for an AI rota-planning tool. The AI lead's note to the owner ran to five lines:
- Request: AI rota planner for the café, about $40 a month for up to 15 staff.
- Problem it solves: rota takes the café manager about 3 hours a week, mostly juggling part-timers' availability.
- Data it needs: staff names, availability and contracted hours. No customer data.
- Overlap: the payroll software has a basic rota module nobody has tried.
- Recommendation: test the payroll module for two weeks first; if it can't handle availability, trial the new tool for one month.
The owner approved the recommendation in two minutes, because the lead had done the thinking. Without the lead, the same request would have been a yes or no made on the café manager's enthusiasm.
A responsibilities chart you can copy
This uses a common format: R is responsible (does the task), A is accountable (answers for it and signs it off), C is consulted beforehand, and I is informed afterwards. Each row should have exactly one A.
| Task | Decision owner | AI lead | Workflow owner | Staff |
|---|---|---|---|---|
| Approve a new AI tool or paid seat | A | R | C | I |
| Set data rules (what goes into which tool) | A | R | C | I |
| Change a workflow's instructions or files | I | C | A, R | I |
| Check output before it reaches customers | A | R | ||
| Monthly cost and usage check | I | A, R | ||
| Respond to an AI mistake a customer saw | A | R | R | I |
| Remove a leaver's access to AI tools | I | A, R | ||
| Quarterly review: keep, fix or stop each workflow | A | R | C | I |
The row about customer-facing mistakes is worth rehearsing before it happens. Whoever spots the problem tells the workflow owner straight away; the workflow owner pauses the workflow if needed; the AI lead handles the fix; the decision owner decides what to tell the customer. If you'd like that written down properly, there's an AI incident response plan template for small firms.
Walk it through with a real-looking case. The garden centre's web shop publishes AI-drafted plant descriptions, and one tells customers a shrub is "fully hardy to hard frosts" when it isn't. A customer emails on a Monday morning with a photo of a dead plant. The member of staff who reads the email tells the web shop head (workflow owner), who unpublishes the description and pauses new drafts within the hour. The operations manager (AI lead) finds the cause that afternoon: the supplier sheet had a blank hardiness column and the AI filled it in, so the instructions gain a rule to leave blanks blank and flag them. The owner (decision owner) decides on a replacement plant and a short, plain apology. Each person did one thing, and nobody had to ask whose job it was.
Who shouldn't own AI
- The keenest junior member of staff, on their own. Enthusiasm is valuable, but without authority they end up responsible for things they can't control, like cancelling a tool the owner likes. Make them a champion instead; choosing and supporting an AI champion explains how that role differs from the lead.
- The outside IT provider alone. They know the systems, not the work, and can't judge whether an answer to a customer is right.
- The software vendor. Its account manager is helpful, but paid to increase your usage.
- "Everyone". Shared ownership means shared tasks fall through the gaps, especially dull ones such as removing leavers' access.
- The most technical person, by default. Skill with prompts and automations doesn't come with authority over budgets or customer relationships.
Role wording you can put in writing
Adding a few lines to a job description or staff handbook makes the role real. Adapt these:
AI LEAD (about [4] hours a month)
- Keeps the list of AI tools and workflows, with owners and costs
- Manages AI accounts and seats; removes access when people leave
- Keeps the usage policy and the shared prompt library up to date
- Runs a monthly check of costs, usage and a sample of outputs
- First point of contact for AI questions and problems
- Brings new tool requests and any incidents to [owner] for decision
WORKFLOW OWNER ([workflow name])
- Owns the instructions, reference files and checking step
- Keeps reference files current (prices, hours, policies)
- Tracks the workflow's agreed measures monthly
- Pauses the workflow and tells the AI lead if it goes wrong
- Brings a keep, fix or stop recommendation to each quarterly reviewPut the names on your one-page AI strategy too. If you haven't written one, the one-page AI strategy has an owner box for exactly this.
Handing over when the AI lead leaves or changes role
The riskiest moment for AI ownership is a resignation. Much of what the AI lead knows is invisible until it's gone. Give yourself a two-week handover and work through this list with the outgoing and incoming person together:
- Transfer admin rights on every AI and automation account to the new lead, then remove the old lead's admin role.
- Check which accounts, if any, are still registered to the leaver's email address and move them to a business address.
- Walk through the tool list and every workflow's playbook, noting anything undocumented.
- Export or copy the shared prompt library and confirm the new lead can edit it.
- Check which payment card each subscription uses, and update it if the card was in the leaver's name.
- Tell staff who to contact from the handover date.
The second item is the one that bites. In an illustrative dental practice, the departing practice manager's email account was closed on her last day, as normal. The automation that sent review requests after each appointment was registered to that address, so it stopped. No error reached anyone, because the error emails went to the closed inbox too. The practice noticed five weeks later, when new online reviews dried up. Moving the account to the practice's shared admin inbox before the leaving date would have taken ten minutes.
Do the same, in a shorter form, when any workflow owner leaves. A workflow with no owner should be paused until someone takes it on, not left running unattended.
Signs ownership has slipped
Ownership tends to erode quietly. Check for these every quarter:
- A subscription renews and nobody knew it was on the company card.
- An automation that sends customer messages runs from an account tied to someone who left.
- A workflow's instructions exist only in one person's chat history.
- When a draft is wrong, staff aren't sure who to tell, so they just fix it and move on.
- Reference files, such as the price list the AI drafts from, are months out of date.
Each one points to a role that exists on paper but not in practice. The fix is usually time rather than a new person: give the named owner a fixed hour in the diary to do the job, and check at the next review that it happened.
Ownership questions that come up
Should the person who runs AI day to day be paid more for it?
That's a pay decision for you, but at minimum the time must be real. If the role needs four to eight hours a month, take that time out of something else, or it will be done in evenings and then not at all. Writing it into the person's job description and reviewing it in their appraisal signals that it's proper work.
Can two people share the AI lead role?
They can split tasks, for example one handling accounts and costs while the other looks after instructions and quality, but one of them should have the final say. Shared roles without a tiebreaker tend to leave the awkward jobs, such as cancelling an unused tool or removing a leaver's access, undone because each assumes the other did it.
What if the owner doesn't want to be involved at all?
The owner can delegate the running entirely, but not the accountability. At minimum they should approve the budget cap, the list of things AI must never do, and any tool that touches customer or financial data, and they should attend a short review each quarter. Those decisions carry business risk that only the owner can accept.
Further reads
- AI Tool Approval Process: How Staff Request a New AI Tool — The request-and-approve route your AI lead will run.
- Automation Audit: Find the Zaps and Scenarios Nobody Owns — Find automations whose owner has already left.
- How to Write an AI Usage Policy for Your Small Business — The rules document the AI lead keeps up to date.
- Reduce Owner Dependency: Use AI to Capture What Only You Know — Useful if everything currently sits in the owner's head.
- How to Redesign Job Roles Once AI Handles Routine Tasks — How roles change once AI handles routine tasks.
- How to Scale AI From One Workflow to the Whole Business — When ownership needs to be formalised as workflows grow.
- AI Adoption Stages: Where Is Your Business Now, and What's Next? — Place your business on a five-stage AI adoption scale with a ten-question check, then see the one move that gets you to the next stage.
- Why AI Projects Fail in Small Businesses (It's Rarely the Tech) — The seven organisational reasons small-business AI projects fail, what each looks like by week three, and an eight-question check to run before you start.
- How to Stop Zapier and Make Automations Breaking Silently — Four layers of protection against quiet failures: notifications, error handlers that alert someone, heartbeat and volume checks, and validation of AI outputs.
- What an AI Consultant Needs From You: Access, Data, and Time — The access, sample data, staff time and decisions an AI consultant needs from you, grouped as a checklist with why each matters and how to check it's ready.
- AI Consultant Handover Checklist: What You Need Before They Leave — Everything an AI consultant should hand over before they leave, grouped into a checklist with how to verify each item, a filled runbook and a scored example.
- Signs Your Business Isn't Ready for AI Yet and What to Fix First — Eleven signs a business isn't ready for AI yet, how each shows up day to day, and the cheap fix to make before paying for any tool or project.
- How to Implement AI in a Small Business With No Tech Team — Running AI with no IT staff: who covers the technical jobs, which tools need no code, how to secure logins, and what to do when something breaks.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.