A focused first engagement covering one or two processes often follows a six-week shape: week one maps the work and agrees how success will be measured, weeks two and three design and build, week four tests on real examples, week five goes live with a safety net, and week six covers handover and a first review against your starting numbers.
Treat that as a shape, not a promise. The same stages can fit into two weeks for a single simple automation, or stretch over three months across several departments. What stretches most engagements happens on the client side: waiting for access, sample data or a decision. Ask any consultant for their own week-by-week plan in writing, and compare it with the one below.
The six weeks on one page
| Week | The consultant | You | What you should hold by Friday |
|---|---|---|---|
| 1 | Walks through the process, counts volumes, times tasks, lists exceptions | Gives access and sample data; makes the process owner available | A process map, baseline numbers, agreed success measures |
| 2 | Designs the workflow and drafts prompts | Answers rule questions; approves the design | A one-page design and a rules sheet |
| 3 | Builds a first working version in your accounts | Reviews early outputs | A version that runs on test data |
| 4 | Runs it against real past cases and in shadow mode | Process owner checks results | A test sheet with a pass rate against the agreed bar |
| 5 | Switches it on for a subset, with approval steps and alerts | Approves outputs; reports oddities | A live system with a safety net |
| 6 | Writes runbooks, runs the handover and break test, compares numbers | Staff run a fix themselves; owner signs off | Handover pack, early results, a decision on what's next |
Each stage below uses the same illustrative business: a barber shop with four chairs, about 480 appointments a month through a booking app, and two problems the owner wants fixed. No-shows run at 8%, which is about 38 empty slots a month; at an average of $28 a cut, that's roughly $1,060 of booked time. And only about a third of clients rebook within five weeks, which the owner believes is low for regulars.
Before week one: the proposal, the scope and a kick-off pack
Most of what decides the six weeks is settled before they start. The proposal should name the processes in scope, what will exist at the end, how you'll both judge whether it works, and what the consultant needs from you. If scope is still unclear, a short paid discovery stage comes first; running a paid discovery phase explains how that works, and scoping an AI project covers deliverables and acceptance criteria.
For the barber shop, the scope might read: "Reduce no-shows with an extra reminder and a same-day confirmation; send personalised rebooking nudges to clients who haven't rebooked within four weeks; build in the shop's own accounts; hand over runbooks and prompt cards." Just as important is what's out: Instagram DMs, the rota and stock ordering. Writing the exclusions down now saves an argument in week four.
Week one: mapping the work and fixing the baseline
The consultant spends most of week one learning how the work really happens, which often differs from how it's described. For the barber shop that means a walkthrough with the front-of-house manager, a look at the booking app's reminder and export settings, and a count of the numbers above from real records rather than memory.
The exceptions matter more than the normal cases. The walkthrough turns up three: some regulars book by text message directly with their barber, so the booking app never sees them; walk-ins aren't in the system at all; and one barber works only Thursdays to Saturdays, so "rebook in four weeks" means something different for his clients. None of these would appear in a tidy description of the process, and each changes the design.
By Friday the shop should hold a one-page process map, the baseline (8% no-shows, about 35% rebooking within five weeks) and agreed targets, for example no-shows under 5% and rebooking at 45% within three months. Targets need a time frame and a measurement method, or nobody can say later whether they were met; setting success criteria that hold up goes into how.
Weeks two and three: design, then a first working version
Week two produces a design on paper before anything is built: which trigger starts each step, what the AI writes and what's fixed text, who approves what, and where alerts go. The owner is asked for decisions only they can make. Should the rebooking nudge suggest a specific barber? (Yes, the client's usual one.) Should it offer a discount? (No.) What happens if a client replies "stop"? (They're removed from nudges immediately.)
Then the prompts are drafted and tried on real examples. A first draft for the rebooking nudge might be:
Write a short, friendly text (under 300 characters) inviting a client
to rebook. Use their first name and their usual barber's name.
Suggest booking via the link. No discounts, no emojis, no pressure.
Client: [first name], usual barber: [barber], last visit: [date]
Barber's working days: [days]
Shop opening days: Tuesday to Saturday
Booking link: [link]An illustrative output from an early test:
Hi [first name], it's been four weeks since your last cut with [barber]. Fancy a fresh one? He has space this Monday or Tuesday: book here [link]
It reads well and it's wrong twice: the shop is closed on Mondays, and the barber doesn't work Tuesdays. The AI invented availability because the draft asked it to be helpful and nothing stopped it naming days. The fix is to stop it naming days at all and let the booking link show real availability: "Don't mention specific days or times; the link shows availability." Week three builds the corrected version in the shop's own accounts, so nothing has to be moved later.
The weekly check-in, and the status note worth asking for
Between the stages, a short weekly check-in keeps everyone honest. Thirty minutes is usually plenty, and it works best with a written note sent beforehand, so the call is spent on decisions rather than updates. The barber shop's note at the end of week three might read:
| Heading | End of week three |
|---|---|
| Done | Rebooking nudge built in the shop's own Zapier account; reminder flow connected to the booking app |
| In progress | Same-day confirmation text, waiting on the booking app's message templates |
| Blocked | Nothing |
| Decisions needed from the owner | Should clients who book by text directly with their barber be added to the booking app? (About 1 in 10 regulars) |
| Risks | Phone cancellations are logged late, which may trigger reminders to cancelled clients |
| Next week | Test against 40 past clients; start shadow mode on Thursday |
Two lines deserve attention every week. "Decisions needed" is where engagements stall, so answer those before the call ends. And "Risks" should change from week to week; a risks line that's always empty means nobody is looking.
Signs an engagement is drifting by week three
- No working version by the end of week three. Something should run on test data, even if it's rough.
- Week-one questions still being asked. If the consultant is still asking how bookings work, the mapping didn't happen properly.
- Everything built in the consultant's own account "for now". Moving it later costs time; ask for it in yours from the start.
- No test sheet planned for week four. Without one, week four becomes a demo.
- Your decisions piling up. Sometimes the drift is on your side. Three unanswered questions in the status note is a week lost.
Any of these is worth raising at the next check-in. Early in an engagement a course correction costs a conversation; after go-live it can cost a rebuild.
Week four: testing on real work
Testing uses the shop's own history. The consultant takes 40 past clients due a nudge, generates the messages without sending them, and the manager marks each one: would you have sent this? The agreed bar might be 38 of 40 acceptable, with no message naming a day, a price or an offer.
Then comes shadow mode: the automation runs on live bookings for a few days but sends nothing, while staff compare its choices with their own. Shadow mode catches the problems test data can't. At the barber shop it confirms the risk flagged in the week-three note: reminder texts queued for two clients who had cancelled by phone that morning, because phone cancellations were entered in the booking app hours later. The fix is procedural rather than technical: cancellations go into the app at the time of the call, and the reminder step checks the appointment is still active before sending.
This is the week that's most tempting to shorten, usually because the demo looked good. It's also the week that decides whether customers ever see a mistake.
Week five: going live with a safety net
Go-live is gradual. The nudges start with one barber's clients for a few days, then everyone's. For the first week a person approves each rebooking message before it goes out, and every error alert goes to the owner as well as the consultant.
Two platform details often surprise owners here. Automation tools check for new data on a schedule, and on Zapier that interval depends on the plan: every 15 minutes on Free, every 2 minutes on Professional and every minute on Team and Enterprise. A same-day confirmation that seems "late" may simply be waiting for the next check. And if the approval step uses Zapier's human-in-the-loop feature on the Professional plan, approvals can only go to the account holder, so the barbers can't approve their own clients' messages unless the shop moves to Team or approves in a different way. Details like these belong in the design, but week five is where they show up if they were missed.
Week six: handover and the first honest look at the numbers
The final week turns a working system into one the shop can run. The consultant writes a one-page runbook for each automation, hands over the prompt cards and test sets, moves any remaining access into the owner's control and runs a break test: the manager pauses an automation, re-runs a failed step and re-authorises a connection, without help. The full list is in the AI consultant handover checklist.
Then the numbers, with a caveat. After two weeks live, the barber shop's no-shows might be down from 8% to 6%, and rebooking up a few points. Two weeks is too short to be sure of either, because a single quiet week moves the percentages. The honest week-six report says what moved, what's too early to judge, and when to look again, usually a month or two later.
Where the weeks slip, and how to stop it
| Cause | How it shows up | Prevention |
|---|---|---|
| Access not ready | Week one spent waiting for a login or an admin approval | Set up accounts and permissions before kick-off |
| Sample data late or unusable | Photos of paper records; exports missing key columns | Export and check the files in advance |
| Decisions pending | The build stops on questions only the owner can answer | One decision-maker who replies within a working day |
| Process owner unavailable | Testing squeezed into gaps between clients | Book testing slots outside peak hours |
| Scope growing mid-build | "While you're at it, could it also answer DMs?" | A written change process; new items quoted separately |
| A vendor change | A renamed field or retired feature mid-project | A few days of slack in the plan |
Scope growth is the one owners cause most cheerfully, because the ideas are usually good. Park them on a list for after week six rather than folding them in; handling change requests in AI projects covers how to do that without souring things.
Costs that start before the engagement ends
The consultant's fee isn't the only money that moves during these weeks. Tool subscriptions usually start in week two or three, when the build moves into your accounts, and usage charges start in week five, when real messages flow. For the barber shop that means a Zapier plan, extra text-message credits in the booking app for the additional reminders, and whatever the AI step uses. A sensible rule is to stay on monthly billing until testing has passed: Zapier's Professional plan with 750 tasks is $29.99 a month billed monthly against $19.99 billed annually, and the annual saving isn't worth much if week four shows you need a different design or a bigger tier. Switch to annual billing in week six, once you know what you actually use.
Watch usage during shadow mode too. If shadow mode runs the AI step and logs its output to a sheet, those steps count towards your plan's tasks or credits just as live runs do, so a busy shadow week can use more of the month's allowance than you'd expect.
How the shape stretches or shrinks
The six stages stay the same; the time each takes changes with the job.
- A two-week version. A personal trainer connecting a booking app to an AI-drafted weekly check-in has one process, no staff and simple data. Mapping takes a morning, build and testing a few days, and handover is a single runbook. Every stage still happens, just smaller.
- A three-month version. A garden centre automating enquiries, click-and-collect and supplier orders across three teams repeats weeks two to five for each process, often overlapping, with a shared week one and a combined handover.
- A strategy-first version. If the owners disagree about priorities, a short strategy stage comes before week one, and the build starts only when it's settled.
For a broader view of how long different kinds of project take, see how long AI implementation takes for a small business.
Misconceptions about the middle weeks
- "The consultant disappears to build." A well-run engagement has a short check-in every week, and the build happens in your accounts, where you can see it.
- "If the demo worked, testing is a formality." Demos use tidy examples. Testing on real history and in shadow mode is where the cancelled-by-phone clients and the invented Mondays turn up.
- "Go-live is the finish line." It's the start of the only data that counts. The first weeks live are when approvals, alerts and early numbers matter most.
- "More weeks means a better result." Beyond what the scope needs, extra weeks usually mean waiting. A plan with clear Friday outputs keeps the engagement moving whatever its length.
So judge any consultant's plan by its Friday outputs. If you can't say what you'll hold at the end of each week, ask them to rewrite the plan until you can.
Questions about the weeks of an AI engagement
How much of my own time does a six-week engagement need?
Plan for about an hour a week as the decision-maker, plus a longer kick-off and a final review. The person who does the work today needs more: a walkthrough of a couple of hours in week one, then 30 to 45 minutes a week reviewing designs and testing. Put those sessions in calendars before week one, because testing squeezed into a busy shift is where engagements lose most time.
Can the engagement be done faster?
Yes, mainly by preparing before week one: access set up, sample data exported, rules written down and one person with the authority to decide. A smaller scope helps too; one automation can often fit into two weeks. What shouldn't be cut is testing on real examples. A week saved there tends to be repaid with interest after go-live, in front of customers.
What happens after the final week?
The automation runs on its own, and you watch the numbers against the baseline for another month or two, because early results are noisy. Decide whether you need ongoing support based on what actually breaks, not on what might. Many owners book a single review a month or so after handover to check the numbers and adjust anything the first weeks revealed.
Further reads
- What Happens in a 1:1 AI Implementation Consultation? — The single call that often comes before an engagement like this.
- AI Proof of Concept vs Pilot vs Production: What Changes? — How a proof of concept, a pilot and production differ in practice.
- AI Implementation Roadmap for Small Businesses: 5 Phases in 90 Days — The longer roadmap when one engagement becomes several.
- How to Get Staff Buy-In When You Introduce AI — Keep the people testing in week four on side from week one.
- What an AI Consultant Needs From You: Access, Data, and Time — The access, data and time to have ready before week one.
- Is an AI Consulting Retainer Worth It After Your First Project? — Whether to keep the consultant on once the six weeks are up.
- What Does an AI Implementation Consultant Actually Do? — The six stages of an AI implementation consultant's work, the evidence each stage should leave behind, and how the role differs from strategy or IT help.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Zapier help and pricing pages (polling intervals by plan, Free plan limits, approval steps on Professional); booking-app and automation vendor documentation for the tools named in the examples.