What an AI automation consultant actually does.
Not “drives AI transformation”. Here is the week-by-week shape of a real engagement, the questions asked during discovery, and exactly what is in the handover package at the end - the part that determines whether you own the automation or merely rent it.
- Week-by-week map
- Real discovery questions
- Handover contents
- You own it, no lock-in
A third discovery, a third building, a third handover.
The instinct is to think the building is the job. It is not. Most automation projects that disappoint were built correctly against a process nobody understood properly, or were built well and then handed over so poorly that the business could not change a single line of copy without calling someone.
A competent consultant spends as much effort on working out what should be automated, and on making sure you can maintain what they leave behind, as on the build itself. What follows is that split, week by week.
Week by week, for one workflow.
Timings assume a single well-scoped workflow for an owner-operated business. Multi-system pipelines stretch the build and testing phases, usually because of access and edge cases rather than build difficulty.
- Week 0
Audit
A 30-minute call plus a written opportunity report within 48 hours. The consultant maps how the work happens today, ranks the automatable workflows by payback, and names the ones not worth automating. Nothing is built and nothing is accessed.
You get: A written report and a prioritised roadmap you keep either way. - Week 1
Scope and roadmap
One workflow is chosen off the roadmap - normally the shortest payback rather than the most annoying task. It gets written up in your own words: the trigger, every step, the edge cases, where a human approves, and what is explicitly out of scope. A fixed price is agreed in writing.
You get: A scope document and a fixed quote. This is the last point at which walking away costs you nothing. - Weeks 2-3
Build
Access is set up in your accounts, the workflow is built against your real tools, and the edge cases from the scope document are handled one by one. Data cleanup, if the audit flagged it, happens before the logic that depends on it. You see progress mid-build, not at the end.
You get: A working workflow in a test mode, and questions about the decisions only you can make. - Week 3-4
Test and parallel run
The automation runs alongside the manual process. Every output is checked before it goes anywhere near a customer. Approval steps stay switched on. This is where the exceptions nobody remembered during scoping surface, and they always do.
You get: Confidence based on watching it behave, rather than on a demo. - Week 4
Go live
The manual process stops, logging and alerting are verified, and the fallback to manual is tested deliberately rather than discovered during an outage. Approvals stay in place until you choose to remove them.
You get: A live automation and a documented way to turn it off. - Week 4-5
Handover
The source, credentials, documentation, and a recorded walkthrough are transferred. The consultant demonstrates how to make the two or three changes you are most likely to want, so the first small edit does not require hiring anyone.
You get: The complete handover package below. You own it. - After
Optional support, then the next workflow
A support arrangement, if you want one, is separate and optional - the automation does not depend on it. Most businesses then wait a few weeks, confirm the first one holds up, and pick the second item off the roadmap.
You get: No obligation to continue, and nothing that stops working if you do not.
Week 0 is the free automation audit. It exists as a separate step precisely so you can stop after it, with a written roadmap in hand, without having spent anything.
The questions actually asked on the call.
These are the questions that determine whether an estimate is real. If nobody asks about your exception rate or your data consistency, you are being quoted from a template.
How the work happens today
- Walk me through the last time you did this task, start to finish. Not how it is supposed to work - how it actually went.
- Who touches it, and what does each person do?
- How many times a week does it happen, and how long does one run take?
- Where does it usually stall or get forgotten?
Rules and exceptions
- What decides which path a case takes?
- Out of the last twenty, how many were straightforward and how many were special?
- What does a special case look like, and who decides what to do with it?
- Is there a rule you follow that has never been written down anywhere?
Systems and data
- Which tools does this pass through, and who owns those accounts?
- Where does the record of truth live, and is there a second place people also update?
- Show me five real records. Do they look consistent to you?
- What happens today when two records are the same customer?
Risk
- If this went out wrong to a customer, what would it cost you?
- Which steps must a human approve, permanently, even once you trust it?
- What should happen if the automation cannot handle a case - stop, escalate, or fall back to manual?
- Who needs to know when something fails?
Value
- What is the loaded hourly cost of whoever does this now?
- Does the delay lose you revenue, or just time?
- If this were handled, what would that person do instead?
- What would make you say in six months that this was worth doing?
The one that changes projects most
What you receive, item by item.
This is the section to compare across consultants. Everything else in an engagement is temporary; the handover package is what you still have in a year.
The workflow source
The actual automation: exported workflow files, configuration, prompts, and any scripts. Not a screenshot of it, not read-only access to it. Files you hold, that another builder could import and maintain.
Credentials, in your accounts
Every tool subscription and API key in your own name, billed to you at list price. The consultant is a collaborator you can remove in one click, not the account owner.
Written documentation
What triggers it, what it does at each step, what the branches are, what happens on failure, and - most importantly - how to change the things you are most likely to want to change: the message wording, the timing, the escalation rule.
A walkthrough video
A recording of someone opening the workflow and explaining it out loud. This is the artifact that matters most in twelve months, when nobody remembers the build and the documentation alone reads like a manual for someone else's business.
The failure runbook
How to tell it has stopped, where the logs are, how to switch it off, and how to run the process manually until it is fixed. One page. Tested, not theoretical.
The running cost sheet
Which subscriptions the automation depends on, what each costs, and what usage would change that. So the monthly number is never a surprise on a card statement.
You own it. No lock-in.
What the job is not.
- Selling you AI. If the payback maths does not work, the right output of an engagement is a recommendation not to build.
- Replacing your staff. In small businesses, automation almost always moves people onto work they did not have time for, because there was never slack in the first place.
- Rebuilding your tech stack. A consultant who wants to move you onto new software before automating anything is solving a different problem.
- Strategy decks. The deliverable is a working workflow and a handover package, not a slide about your AI maturity.
- Owning your process forever. If you cannot change the message wording without calling them, the handover failed.
Where the hours actually go.
| Phase | Rough share of effort | What consumes it |
|---|---|---|
| Discovery and scoping | About a quarter | Getting the real process out of people's heads, finding the undocumented rules, and pricing the exceptions. |
| Data preparation | Varies from nothing to a third | Deduplication, consistent statuses, formats, and reconciling records that live in two places. Entirely dependent on your starting point. |
| Building | About a third | Integrations, logic, prompts, approval steps, error handling. Grows with the number of systems, not with ambition. |
| Testing and parallel run | About a fifth | Running against real cases, catching the exceptions nobody mentioned, tuning the outputs people will actually send. |
| Documentation and handover | About a tenth | Writing the docs, recording the walkthrough, transferring credentials, and testing the failure runbook. |
The reason quotes vary so much for apparently identical workflows is the second row. Two businesses asking for the same invoice-chasing automation can differ by a factor of three purely on how consistent their records are. The cost breakdown covers that line item in detail.
What the job looks like on an ordinary Tuesday.
Strip away the job title and the work is unglamorous in a specific way. A consultant in the middle of a build spends most of the day in three activities: talking to the person who actually does the task, wiring systems together, and testing against records that are messier than anybody described.
Talking to the person who does the work
Not the owner, and not the manager. The person who processes the thing. Owners describe the process they designed; the person doing it describes the process that exists, including the spreadsheet they keep on the side because the official system cannot represent a case that comes up twice a month. Every serious surprise in an automation project is already known to that person, and the job is largely getting it out of their head before it becomes a defect.
Wiring systems together
The building itself is integration work: triggers, field mapping, branching, retries and the failure path. It is the part clients imagine is the whole job and it is usually the fastest part, because the hard decisions were made during scoping. What consumes time here is authentication, permissions, rate limits, and systems whose documentation describes an older version of themselves. Anyone who has connected two business tools before knows the shape of this.
Testing against real records
Not sample data. Real customers, real invoices, real appointments, with the duplicates and the blank fields and the entries someone typed into the wrong column in 2019. The parallel run is where the automation earns trust, and it is also where the honest consultant quietly fixes the three assumptions they got wrong in week one.
And the part nobody advertises
A meaningful share of the week is writing: the scope document, the decisions log, the failure runbook, the change instructions. It is not overhead. It is the difference between owning an automation and renting one, and it is why handover is a term worth understanding before you sign anything.
What exists at the end of each phase.
A useful way to judge a proposal is to ask what artifact each stage produces. Stages that produce nothing you can hold are stages you are paying for on trust.
| Stage | The artifact | Who it is for | Still useful if you stop here |
|---|---|---|---|
| Audit | Written opportunity report and a ranked roadmap | The owner, and whoever signs off spend | Yes. It is a list of what to automate and what to leave alone, in your own language. |
| Scope | Scope document plus a fixed quote | Both sides, as the contract of record | Yes. Any other builder can quote against it, which is exactly why it should be yours. |
| Build | Working workflow in test mode, plus a decisions log | The builder and the person who does the task | Partly. Half a workflow is worth little, which is why the scope stage is the honest exit point. |
| Parallel run | Test results against real cases and a defect list | The team that has to trust it | Yes. The defect list is a map of your process, not just of the software. |
| Go live | Live automation, logging, alerting and a tested off switch | Operations | Yes, and it keeps running without the consultant. |
| Handover | Source, credentials, documentation, walkthrough recording, failure runbook, cost sheet | You, permanently | This is the deliverable. Everything above is how it got made. |
Compare that list against any proposal in front of you. Our proposal template has the same fields and is free to hand to a supplier who is not us, and the workflow brief is the document that turns a vague request into something quotable in the first place.
How consultants actually charge, and what each model does to you.
There are four common shapes, and the shape matters more than the rate. It decides who carries the risk when the work turns out to be harder than either side expected.
| Model | How it works | Who carries the risk | When it is the right choice |
|---|---|---|---|
| Fixed price per workflow | One scope, one number, agreed in writing before work starts. | The builder. If it takes longer, that is their problem. | Almost always, for a defined workflow. It forces proper scoping, which is the part that protects you. |
| Hourly or day rate | You pay for time spent, usually with an estimate rather than a cap. | You. Discovery, rework and integration surprises all land on your invoice. | Open-ended investigation, or ongoing changes to something already built. |
| Monthly retainer | A recurring fee for maintenance, changes and support. | Shared, and it depends entirely on what is in scope. | After handover, when you genuinely want someone on call. It should never be the thing keeping the automation alive. |
| Outcome or revenue share | Payment tied to a measured result. | Nominally the builder, in practice whoever controls the measurement. | Rare and hard to do honestly, because attribution in a small business is usually not clean enough to settle an argument. |
The number itself is the wrong first question. A quote is only meaningful against a scope, which is why what AI automation costs breaks the price into build effort, tool subscriptions and monthly running cost rather than quoting one figure, and why the build cost page separates the one-off from the recurring. For a single well-known build in a single trade, the AI receptionist cost breakdown is the most specific worked example on this site. If you are on the other side of the table and pricing this work yourself, the agency pricing calculator is the tool for it.
The two questions that price a project
What you should be doing while they work.
Clients who get more out of an engagement are not the technical ones. They are the ones who did these four things.
- Give access to the person who does the task, not just to the owner. An hour with them in week one is worth more than any document. Guard it in the calendar.
- Decide who signs off, before the parallel run. The most common cause of a stalled project is not a technical failure, it is a decision nobody is authorised to make.
- Look at your own data honestly. Duplicate customers, three spellings of the same status, dates stored as text. Nobody automates their way past that for free, and you will pay for the cleanup whichever route you take.
- Say out loud which steps must always be approved by a human. That list is a design input, not a limitation. The human in the loop pattern is what makes the first automation safe enough to trust, and guardrails are what keep it that way.
- Ask what happens when it breaks. Not whether it will. The failure path, the retry behaviour and the alerting should be described in the scope document before the build starts.
If you want to understand the material before the first call, the what is an AI agent explainer and the implementation roadmap cover the vocabulary, and the agent archetype library shows what each kind of automation is actually for. The support agent and invoice processing agent pages are the two archetypes that come up most often in a first engagement.
What a consultant actually builds on.
There is no secret toolchain. Most small-business automation is built on a workflow platform, one or two model providers, and the business systems you already pay for. The platform choice matters more than people expect, because it sets a recurring bill for as long as the automation lives and it decides how hard it is to move later.
Published cloud pricing gives you an order of magnitude. As of August 2026, n8n lists its Starter cloud plan at 20 euro a month billed annually with 2,500 workflow executions, and Pro at 50 euro with 10,000 executions. Make.com lists a free tier at 1,000 credits a month, Core at $9 a month and Pro at $16 a month at 10,000 credits, billed monthly. Zapier sits in the same category with a different pricing unit. The comparison that matters is not the headline price, and n8n versus Make works through the one that does, which is whether you can take the workflow with you. The full agent platform directory covers the rest, including the self-hosted options.
The second decision is whose model runs the AI steps. That determines a per-use cost that scales with volume rather than a flat subscription, and it determines where your data goes. The model provider pages carry current published token pricing per provider, including Anthropic Claude and OpenAI, and the cheapest LLM API comparison is the one to read if the workload is high volume and simple. A consultant should be able to tell you what the monthly bill will look like at your volume before you commit, and the monthly running cost breakdown shows how that number is built up.
Everything else is your existing systems. A competent consultant integrates with what you run rather than proposing you replace it, and treats vendor lock-in as a risk to be minimised rather than a business model. If you want to see the full build process on a real workflow before hiring anyone, the free missed call text-back guide, appointment reminders guide and invoice chasing guide are complete builds, and the full guide library has the rest.
How this differs from an agency, a freelancer and an in-house hire.
The role only makes sense next to what else you could do with the money. Each of these is genuinely better than a consultant in some situation, and any page that says otherwise is selling.
| Route | What you are really buying | Its honest weakness |
|---|---|---|
| Solo consultant | Direct access to the person building it, and most of the budget going into the build rather than into coordination. | One person builds one thing at a time. Key-person risk is real and does not go away by ignoring it. |
| Agency | Parallel delivery, cover when someone is unavailable, and a formal vendor relationship procurement can accept. | Overhead. Below a certain project size you are paying for account management, and the builder assigned to you may be their most junior. |
| Marketplace freelancer | The lowest paid price for a narrow, clearly specified task. | Handover and continuity. Unless documentation is in the brief you will not get it, and the same person may not be findable in six months. |
| In-house hire | Permanent capacity and institutional knowledge that stays in the business. | Months to hire, a salary regardless of workload, and undocumented workflows that leave when they do. |
| Doing it yourself | Total ownership and no fee at all. | Reliability. First versions work on the happy path and fail silently everywhere else, and your time is the real cost. |
Consultant versus agency versus DIY takes each of these apart properly, including the budget bands where the answer flips and the cases where we are the wrong choice. If the alternative on your table is a person rather than a supplier, the hire versus automate comparison and the build versus buy calculator put your own numbers against it, and the ROI calculator gives you the payback period the decision actually turns on.
The questions worth asking, and the answers to listen for.
- Who owns the workflow when we are done? The answer should be you, with the source files, credentials in your own accounts, and documentation. Anything vaguer than that is a warning.
- What happens if I stop paying you? Nothing should stop working. If support is what keeps the automation running, you are renting.
- Whose accounts hold the subscriptions? Yours, billed to you at list price, with the consultant added as a collaborator you can remove in one click.
- What is out of scope? A scope document with no exclusions section has not been thought about. The exclusions are where the disputes live.
- What is the monthly cost after handover? An itemised sheet, not a shrug. Subscriptions plus expected model usage at your volume.
- What have you told a client not to automate? Anyone who cannot answer this has either never scoped honestly or is not going to start with you.
The full hiring checklist has fifteen of these with our own answers printed beside them, which is the fairest way we know to be judged. If you would rather test the process than read about it, the free automation audit is week zero on its own, and the automation service page sets out what we build and what handover includes. Our own pricing page covers the other half of the business, which is the courses for people who would rather build it themselves, and the operators course teaches the same method described on this page.
The same role, shaped by the business it lands in.
The method does not change between industries. What changes is which workflow is worth doing first, what the compliance constraints are, and what a saved hour is actually worth. A dental practice starts with recall and no-shows because an empty chair is a fixed cost. HVAC and plumbing firms start with missed calls and unsold estimates, because the job goes to whoever answers first. Law firms start with intake and deadlines. Accounting firms start with chasing documents. Salons and veterinary clinics look like the dental case with different vocabulary.
A consultant who cannot name the first workflow for your kind of business inside ten minutes has not worked in it. That is not a disqualification on its own, but it means you are paying for their learning curve during discovery. The industry pages are written in owner language for exactly this reason, and reading the one closest to your business is the quickest way to judge whether somebody understands the work before you spend a call finding out.
If you build this kind of automation for other businesses yourself, the work with us page is the relevant one, and about explains who is behind this site and how the practice is set up.
The consultant's role: FAQ
What does an AI automation consultant actually do?
They map how a repetitive business process works today, identify which parts can be automated safely, build the automation against the tools you already use, test it alongside the manual process, and hand over the workflow, credentials, and documentation so you own it. The work is roughly a third discovery, a third building, and a third testing and handover.
How long does a typical engagement take?
For a single well-scoped workflow, commonly four to five weeks from first call to handover: a week to scope, one to two weeks to build, a week or so running in parallel with the manual process, then handover. Multi-system pipelines take longer, mostly because of access and edge cases rather than build time.
What questions will a consultant ask me?
The useful ones are about how the work really happens rather than how it is supposed to: walk me through the last time you did this, how many of the last twenty cases were special, which steps must a human always approve, and what would make you say in six months that this was worth doing. If nobody asks about your exception rate or your data consistency, they are not scoping properly.
What do I get at the end?
The workflow source and configuration, all credentials in your own accounts, written documentation covering how it works and how to change it, a recorded walkthrough, a one-page failure runbook, and an itemised running cost sheet. If any of those are missing, you have bought access to a process rather than ownership of one.
Does the consultant need access to my systems?
For the build, yes, and only to what the specific workflow needs. Access should be granted in your own accounts so you can revoke it at any time, and no access at all is needed for the initial audit and scoping conversation.
What happens after handover?
Support should be optional and separate. The automation runs in your accounts on your subscriptions, so it does not stop if you stop paying the consultant. Most businesses let the first workflow settle for a few weeks, then pick the next item off the roadmap.
How is this different from hiring a developer?
A developer builds what you specify. An automation consultant's main value is in the specification: working out which processes are worth automating, which should stay manual, and where a human must stay in the loop. The building is often the easier half.
How do AI automation consultants charge?
Four shapes are common: a fixed price per scoped workflow, an hourly or day rate, a monthly retainer for support after handover, and occasionally an outcome-based arrangement. Fixed price per workflow is usually the safest for the buyer, because it forces proper scoping and moves the risk of a longer build onto the builder. Hourly puts discovery, rework and integration surprises on your invoice. Rates themselves vary widely by market and are not published consistently enough to quote one honest range.
Will an automation consultant tell me not to automate something?
A good one will, regularly. Work that runs twice a month, processes with no stable rules, and workflows where special cases outnumber standard ones usually cost more to automate and maintain than they save. Every honest engagement produces a list of things to leave alone.
Keep reading
- Free AI automation auditWeek 0, on its own, at no cost: a call, a written report, and a roadmap you keep.
- AI automation servicesThe workflows we build and how the engagement runs.
- How to hire an AI automation consultantFifteen vetting questions with our answers, plus the red flags to walk away from.
- Consultant vs agency vs DIYWhether you need a consultant at all, compared honestly.
Start at week zero.
A 30-minute call, a written opportunity report within 48 hours, and a roadmap you keep. Nothing built, nothing accessed, nothing owed.
