An AI consultant can't decide which problems matter most to your business, know your customers and exceptions better than your team, make staff adopt a new tool, guarantee the AI never gets things wrong, or carry your responsibility for what it does. You must own the choice of problem, the process knowledge, your data, a named owner for each automation, and the results.
None of that is a criticism of consultants. It's the line between help you can buy and work only the business can do. Knowing where it sits before you start prevents the most common disappointment: a well-built automation that fades out within three months because nobody inside the business owned it.
The split at a glance
| A good consultant can | Only you can |
|---|---|
| Show you where AI could save time, with estimates | Decide which of those problems is worth your money and attention |
| Interview your team and map the process | Supply the unwritten rules and the awkward cases |
| Tell you what data the automation needs | Own the accounts, fix the records and decide who sees what |
| Build, test and document the automation | Make it the way the work is done, and retire the old way |
| Explain what to check and how often | Actually check it, every week, after they've gone |
| Describe the risks and options | Choose how much risk the business accepts |
| Point out legal questions to raise | Answer to customers, staff and regulators for the result |
Seven responsibilities that stay with the business
To make these concrete, picture an illustrative dog-grooming salon: the owner and three groomers, bookings through a booking system, and a consultant brought in to set up AI-drafted replies to booking requests and aftercare messages after each groom.
1. Choosing the problem worth solving
A consultant can rank opportunities by hours saved, but only you know what matters to the business right now. Maybe the salon's real problem isn't reply time but no-shows, or staff burning out on Saturdays. If you hand the choice of problem to someone outside, you'll get an efficient answer to a question you didn't most need answering.
The minimum you must do: arrive with your top two problems ranked, and say why each one matters in money, time or stress.
It needn't be long. An illustrative family-run hardware shop might bring this:
1. Trade customers' emailed orders (about 30 a week) get
retyped into the till system. Mistakes cost us roughly
2 wrong deliveries a month, each a wasted van trip and
an annoyed builder. MONEY and REPUTATION.
2. Stock questions by phone ("have you got 8mm masonry
bits?") interrupt the counter 20+ times a day on
Saturdays. STRESS, and queues walk out.
Not on the list: social media. It's fine as it is.The third line is as useful as the first two. It tells the consultant not to spend your hour on the thing they might otherwise have suggested first.
2. Knowing the exceptions
Every business runs on rules nobody wrote down. At the salon: a doodle with a matted coat takes twice as long; puppies' first grooms are short and gentle; a particular client's dog must never be booked after another dog that stresses it. A consultant can ask good questions, but they can't know what they aren't told, and the AI will only follow rules that have been written into its instructions.
The minimum you must do: have the person who does the job each day, not only the owner, spend an hour listing the cases they handle differently, and review the AI's instructions for anything missing.
Here is how a missing exception shows up. In the first week, a drafted reply offers a 4pm Friday slot to the owner of a 14-year-old terrier. The senior groomer spots it only because she happens to be approving that day: elderly dogs always get a morning slot, when they're calmest and the salon is quiet. It was never written down because she has done it for years without thinking. One line went into the instructions ("dogs aged 11 or over: offer morning slots only, before 11am"), and the team then spent twenty minutes asking each other which other rules they follow on autopilot. They found four more.
3. Getting the data and access in order
The automation is only as good as the records it reads. If half the client records in the booking system have no breed or size, the drafted replies will guess. A consultant can tell you what's missing; somebody in the business has to fill it in, and the business must own every account involved. What an AI consultant needs from you lists the access and data to prepare.
The minimum you must do: hold the admin login for every system, clean the fields the automation depends on, and decide who in the team can see client details.
A quick export tells you how big that job is before anyone quotes for it. An illustrative window-cleaning round wants AI-drafted "we're in your street on Thursday" messages, so the owner exports the customer list to a spreadsheet and counts: 640 customers, 210 with no email or mobile number, 95 with the access notes ("side gate code 4471", "dog in garden") typed into the address field, and 38 who haven't been cleaned in over a year. The decision about what to fix is the owner's: archive the 38, move the 95 access notes into their own column (an evening's work, since the messages must never print a gate code), and collect missing numbers on the round over the next month. None of that is consultant work, and all of it decides whether the messages come out right.
4. Making the team actually use it
A consultant can show the groomers how the new drafts work. They can't make it how the salon runs. If the owner keeps answering messages by hand "just this once", the team will too, and within a month there are two systems and trust in neither. Adoption is a management job, and the manager is you. Getting staff buy-in when you introduce AI covers the conversations that help.
The minimum you must do: pick a date after which the old way stops, and stick to it.
Two systems running side by side usually show up as duplicates. At an illustrative four-partner accountancy practice, AI drafts replies to routine client queries for the practice manager to approve. One partner, used to answering his own clients, keeps replying from his phone first. Within a fortnight, several clients receive two answers to the same question, one of them giving a different deadline for sending in their paperwork. The consultant can't fix that; the partners have to agree that routine queries go through the shared inbox, and the partner in question has to stop. It took one partners' meeting and a rule that client emails to personal addresses get forwarded, not answered.
5. Owning each automation after handover
Automations drift. A price list changes and nobody updates the instructions; a connected app changes its settings; a new service appears that the AI doesn't know. If no named person checks it weekly, the first sign of trouble is a customer complaint.
The minimum you must do: name one owner and one backup for each automation, give them 15 minutes a week to check it, and write it into their role. The ownership card below is a starting point. For the bigger question of who should own AI across the business, see roles and responsibilities for AI in a small business.
6. Deciding how much risk you'll accept
Which messages go to customers without a person reading them first? A consultant can lay out the options; the choice is yours because the consequences are. The salon might decide that reminders can go out automatically, booking replies need approval, and anything about an injury, a skin reaction or a complaint is never drafted by AI at all.
The minimum you must do: write down, in one line each, what the AI may send alone, what needs approval and what it must never touch.
The same three lines look different in every business, which is why only you can write them. For an illustrative holiday-cottage letting business they might read:
SENDS ALONE: check-in instructions 48 hours before
arrival; Wi-Fi, parking and bin-day details.
NEEDS APPROVAL: availability and price replies; pets;
early check-in or late check-out requests.
NEVER TOUCHES: damage deposits, refunds, complaints, and
anything about a guest's health or
accessibility needs.A consultant could have proposed those lines. Deciding that a wrong answer about a deposit is worse than a slow one is the owner's call, because the owner is the one who'll be refunding it.
7. Answering for the results
If an AI-drafted message promises a customer something wrong, the promise came from your business. In a 2024 tribunal decision, an airline was ordered to compensate a passenger after its website chatbot described a refund rule the airline didn't actually offer; the argument that the chatbot was responsible for its own words didn't succeed. A consultant's contract may cover their workmanship, but it won't move your responsibility to customers onto them. Who is liable when your AI chatbot gets it wrong goes further, and for anything specific to your situation, ask a solicitor.
The minimum you must do: check your insurance, keep a record of what the automation sends, and have a way to correct mistakes quickly.
Checking insurance is one short email to your broker, sent before go-live rather than after a complaint. Adapt this:
We're starting to use AI to draft replies to customer enquiries.
A member of staff approves each draft before it's sent, except
appointment reminders, which go out automatically.
1. Does our current cover apply if a reply contains a wrong price,
date or promise and a customer claims a loss?
2. Does it matter whether a person approved the message?
3. Is there anything you need us to tell the insurer, or any
condition we should meet (for example, keeping records)?The answer is often "yes, covered as normal", but question 3 is the one that catches people: some policies expect you to disclose a material change in how you work, and the broker is the person who knows whether this counts.
An ownership card for every automation
Fill one in for each automation before the consultant leaves. Keep it where the team can find it, and review it every quarter.
AUTOMATION OWNERSHIP CARD
Name of automation: e.g. Booking reply drafts
What it does: Drafts replies to booking requests for approval
Owner: [name] Backup: [name]
Weekly check (15 min): Read 10 recent outputs; confirm prices and
slots match the price list and diary
What it may send alone: Nothing (all replies approved by a person)
Must never handle: Injuries, complaints, refunds, medical questions
Depends on: Booking system, shared inbox, price list v3
Monthly cost: $[amount], paid from [account]
How to pause it: [exact steps, written for a non-technical person]
If it goes wrong: Pause it, tell [owner], log it in the error log
Next review date: [date]
If you can't fill in the owner line, the automation isn't ready to go live, however well it works.
Signs ownership has already slipped
If an automation has been running for a while, check it against this list. Two or more of these and nobody really owns it, whatever the handover document says.
- Nobody can say what it cost last month without logging in to three accounts.
- No one has read a sample of its output in the past four weeks.
- Prices, opening hours or services have changed since launch, and its instructions haven't.
- Staff have quietly gone back to doing the task by hand "because it's quicker".
- The cases it passes back to a person are piling up unanswered.
- The only person who knows how to pause it has left, or is the consultant.
At the salon, slippage might look like this four months after handover. The medium-dog full groom goes up from $65 to $72 in the booking system, but the price list the AI reads is still version 3. The owner of the automation is on leave, the weekly check lapses, and a groomer who joined last month approves drafts without knowing the old figure. By the time a client queries her bill at the till, 23 replies have quoted $65. The salon honours them, at $7 each, and the fix takes five minutes: update the price list, add "prices match the booking system" to the weekly check, and make price changes a trigger for updating the instructions the same day.
The fix is rarely technical. Name the owner again, give them the time, and run the weekly check together once so the habit restarts. Budget for this honestly: for a single customer-facing automation at a small salon, the owner's share is usually around an hour or two a month, plus a longer look each quarter when prices or services change.
Promises no consultant should make
These all sound reassuring. Each describes something outside a consultant's control, so treat them as warning signs rather than selling points.
- "It'll be 99% accurate." Accuracy depends on your data, your cases and the checks you keep. It can be measured after testing, not promised before.
- "You'll save 20 hours a week." Estimates are fine, labelled as estimates. A guarantee made before seeing your process is a sales line.
- "This makes you compliant." A consultant can flag legal questions. Unless they're qualified to advise on the law that applies to you, sign-off belongs with a solicitor or your data-protection adviser.
- "Set it and forget it." Nothing that talks to customers should be forgotten.
- "The AI will learn your business as it goes." Most business automations don't learn from use unless someone updates their instructions and reference documents. That someone is your owner.
The honest version of each promise is a measurement with a date on it. Instead of "It'll be 99% accurate", a consultant who has done the testing can say: "On 40 of your past enquiries, 34 drafts needed no changes, five needed a small edit and one got a price wrong, which is now fixed. We'll check another 40 after the first month live." Instead of "You'll save 20 hours a week": "Your team spends about six hours a week on these replies now; if drafts cut each one from ten minutes to three, that's roughly four hours back, and we'll time it in week three." Both versions are specific enough to be wrong, which is exactly what makes them useful.
Where the line blurs
Some jobs are shared, and it helps to agree the split in writing.
- Your AI policy. A consultant can draft one. You adopt it, explain it to staff and enforce it.
- Choosing and buying tools. They can recommend and compare. You sign the contract and own the account.
- Fixing the process. They can point out that a step makes no sense. Changing how your team works is your call, and automating a muddled process first rarely ends well, as automating a broken process explains.
- Staff worries. A consultant can explain what the tool does. Conversations about roles, hours and jobs are between you and your team.
A handover week that actually moves ownership
The handover document matters less than whether the owner can run things without help. In the last week of an engagement, ask for five small tests, one a day:
- The owner runs the week's check while the consultant watches and says nothing.
- The owner handles an awkward case the automation passes back, alone.
- The owner pauses the automation and restarts it from the written instructions.
- The owner finds this month's running cost without asking where to look.
- The owner explains the automation to the backup person in under ten minutes.
Test 5 is the one that most often fails, and usefully. At an illustrative garden-maintenance firm, the owner of an automation that drafts quote follow-ups took 25 minutes to explain it to her backup and never mentioned what it must not handle: quotes involving tree work, which need a site visit and a separate insurance check. She knew it; she just didn't think to say it. The fix was to move the "must never handle" line to the top of the ownership card, and to have the backup explain the automation back to her the following week.
If any test fails, fix that before the consultant leaves. It's far cheaper than calling them back in three months to rediscover how their own work functions.
Further reads
- AI Consultant Handover Checklist: What You Need Before They Leave — Everything you should receive before the consultant leaves.
- AI Governance for a Small Business: Who Decides, Approves, Checks — Who decides, approves and checks AI use in a small firm.
- Done-for-You vs Done-With-You AI Implementation: Which Suits You? — Choose how much of the work you want to keep in-house.
- What Does an AI Implementation Consultant Actually Do? — The other side of the line: what the consultant does.
- AI Error Log: Track Mistakes and Stop Them Happening Again — A simple log for the owner of each automation to keep.
- Does Your Business Insurance Cover AI Mistakes? — Check whether your cover extends to AI mistakes.
- Do I Need an AI Consultant? 8 Signs It's Time to Get Help — Eight signs a small business needs outside AI help, each with a test you can run this week, plus the signs that it doesn't need a consultant yet.
- What an AI Consultant Does for a Law Firm, and What It Costs — The five phases of a law-firm AI engagement, what each should hand over, how fees are built from consultant days, and proposal red flags.
- AI Consultant or DIY: Which Suits a Small Professional Firm? — When a small professional firm can set up AI itself, when outside help earns its fee, and the split approach many firms settle on.
- Do Small Medical Practices Need an AI Consultant? — When a small medical practice can handle AI itself, the signs outside help will pay for itself, what a consultant shouldn't do, and a scoped brief to adapt.
- Do Small Charities Need Outside Help to Adopt AI? — What a small charity can do with AI on its own, the five signs it needs outside help, and how to brief that help so the work stays yours.
- AI Consultant vs In-House Hire: Which Costs a Small Team Less? — The fully loaded cost of an AI hire, the same year's work priced from a consultant, and the break-even hours a family butcher would need.
- What Is a Fractional Chief AI Officer and Do You Need One? — What a fractional chief AI officer does each month, what providers publish as prices, and cheaper set-ups that suit most small businesses.
- How to Judge Whether Your AI Consultant Delivered Value — A weighted scorecard for judging an AI consultant: scope delivered, results against baseline, adoption at 90 days, team independence and advice quality.
- AI Consultant Red Flags: 12 Warning Signs to Walk Away From — Twelve warning signs when hiring an AI consultant, each with an example, the question to ask, its innocent version and a scored two-proposal comparison.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.