AI Consulting: Advice You Can Use, or Work That Gets Done?

What AI consulting covers

AI consulting is paid advice on which parts of a business could use artificial intelligence, what that would involve and what could go wrong.

The term is broad, and AI consulting services come in very different shapes. At one end sits strategy work: workshops, interviews with managers, a map of processes and a written roadmap. In the middle sits tool selection: which software to buy, how to set it up and how to train the team. At the far end sits implementation, where someone actually builds the automation inside your own systems and looks after it once it is live.

Most offers mix these in some proportion, so it helps to name each part before you compare quotes. The table below sets out the usual kinds of work an AI consultant may sell, and what you hold in your hands when each one is finished.

Advice-only consulting and where it stops

Advice-only AI consulting ends with a document, and the work of turning that document into something running falls back on you.

That is not a flaw in itself. A large organization with its own technical team may want exactly that: an outside view, a ranked list of opportunities and a set of risks to manage, which its own people then carry out. Advice is also useful when the real question is a policy one, such as which uses of AI the company is comfortable with at all.

For a smaller business, though, the gap between the report and the result is usually where things stall. Nobody inside has the time to build what the report recommends, the tools it suggests need someone to configure and watch them, and six months later the same invoices are still typed by hand.

Before you pay for advice alone, it is worth checking whether you are in a position to act on it. Advice-only work tends to stop at these points:

• It names the jobs that could be automated but does not automate them.
• It recommends tools but leaves someone inside to set them up, connect them and fix them when they break.
• It describes controls such as approvals and audit records but does not put them in place.
• It ends on a fixed date, while the work it describes carries on every week.
• It rarely measures anything in your own hours, because nothing was built to measure.

Consulting that ends in a running system

The other kind of AI consulting ends in a system that does one job inside your own accounts, every time it is triggered, with a person signing each result.

This is what AI implementation means in practice. Instead of a roadmap that lists supplier bills as an opportunity, someone reads how your bills actually arrive, writes down the rules, builds an automation that enters them as drafts and keeps it working as suppliers and layouts change. The advice is still there, but it is folded into the build: you learn what fits by seeing it run on your own work.

For AI consulting for small business, this is often the more useful shape. A small team rarely needs a long strategy document; it needs a few hours back each week on the jobs that repeat, and the confidence that nothing leaves the company without a person's say.

The route from first conversation to a running job usually looks like this.

1. Pick one job that repeats often, follows rules and ends in a document, an entry or an email.
2. Read the real examples: the emails, files and records that job produces today, with read-only access.
3. Write the job down as a brief: where the work arrives, what is done with it, where the result goes and who signs it.
4. Agree the exceptions: what the system must pass to a person instead of guessing.
5. Build it inside your own mailbox, folders, sheets and accounting system, producing drafts only.
6. Switch it on only with your yes, check the first drafts closely, then let it run while a person keeps signing.

Questions to ask an AI consultant

The most useful questions for an AI consultant are about what will exist when the engagement ends, and who looks after it.

Titles and slide decks say little about whether AI consulting will help you. What matters is how the consultant treats your data, your approvals and your hours. These questions bring that out quickly, whatever the size of the firm you are talking to.

What will I have at the end: a report, a trial or a job that runs every time it is triggered? Will the system work inside my own mailbox, files and accounting system, or move my work into a new platform? Does it produce drafts that a person approves, or does it send and post on its own? What happens to a case it cannot handle exactly: is it guessed, or passed to a person? Is every action written to an audit record I can read? How is it stopped if something looks wrong? Who keeps it running when a supplier changes a layout or a rule changes? And if we stop working together, what stays with me?

A good answer to each of these is short and specific. A vague answer to the last one deserves particular care, because it tells you who will own your workflow later.

What it costs and how it is priced

AI consulting is priced in a few common ways, and the right comparison is what each fee leaves you with, not the fee alone.

Advice is usually sold by the day or as a fixed project. Implementation is often a build fee followed by a support contract. A third model is rental: the automation is built, run and looked after by the provider for one monthly figure, and stops when the rental stops. Each model suits a different situation, and none of them is cheap or expensive in itself.

We do not publish a price list, and nobody honest can give you a figure before reading the work. What can be said is what drives the price in any model: the hours the job takes from your routine today, and how complex the automation is. A simple job with clear rules is built quickly; a job that needs several automations working together takes longer and costs more to run.

Whatever the model, check with your adviser how an AI consulting fee should be treated in your own accounts.

How we work

We are not a strategy consultancy and do not sell AI consulting by the day: we read the work, then build and run AI employees that do it, rented monthly.

An AI employee is an automation with one written job description. Ours run on our servers against the client's own accounts; today that means Google Workspace mailboxes, Google Drive, Google Sheets and Xero. The robot only produces drafts, a person signs anything that leaves the company, and nothing is sent by a robot. Every action is written to an audit record, anything it cannot do exactly goes to a person, a kill switch stops any robot in one command, and nothing is switched on without the client's yes.

Running today for a client: sales invoicing, supplier bills, purchase orders, invoice filing, a price history register, a daily mail digest, a mailbox watch, online contract signing and handover notes when a key person leaves. Offered and built to order on the same engine: drafted customer replies, enquiry follow-up, expense receipts, client and supplier onboarding, bank matching suggestions and payment reminders.

The price is one monthly figure per workflow, stated in the quote, based on the hours the job takes from your routine and on the complexity of the automation. If the rental stops, the workflow stops, and every email, entry, file and register stays in your own accounts. Our how it works page sets out the full route, and the checklist on this page is a quick way to see which kind of help you need.

Questions