AI Consultant Handover Checklist: What You Need Before They Leave

Before the consultant leaves, you should hold: owner-level control of every account and connection, fresh API keys, exported copies of each workflow, the final prompts with the tests they passed, a one-page runbook per automation, a list of running costs and renewal dates, failure alerts that reach you, and staff who have run a fix once.

One test sums it up: could someone in your business fix the most likely failure on a Monday morning without calling the consultant? If not, the handover isn't finished, however good the build is. Agree this list at the start of the engagement and tie part of the final payment to it, so nobody is negotiating it on the last day.

Follow me on Instagram@sagnikteaches

Accounts, keys and billing

Start with control, because everything else depends on it. The ownership questions behind this group are covered in who owns the AI workflows a consultant builds; here is what to check on the day.

Connect on LinkedInSagnik Bhattacharya

The same routine you'd use when a member of staff leaves applies here; the offboarding checklist for AI tools and shared accounts covers the steps tool by tool.

Subscribe on YouTube@codingliquids

The workflows themselves

For the illustrative tattoo studio used later in this tutorial, the finished inventory is four rows. It became four only after the final check caught a missing one, as the scored example at the end shows:

Each row points to a runbook and, where there's an AI step, a prompt card. If you inherited automations from before this engagement too, fold them into the same inventory. Finding the Zaps and scenarios nobody owns shows how to sweep for them.

Prompts and the evidence they work

The prompts hold weeks of decisions about tone, rules and exceptions. They need handing over as carefully as the automations, together with the proof that they behave.

• The final version of every prompt, as plain text, dated and numbered. Not just "it's inside the Zap". If the live copy and your copy ever differ, you need to know which is current.
• The test set and its last results. The real inputs used for testing, the expected output for each, and the pass rate at handover. Without these, the next person to edit a prompt can't tell if they broke it.
• The model and settings each prompt runs on. These affect cost as well as behaviour: an "AI by Zapier" step uses 1, 3 or 5 tasks per run depending on the model tier chosen.
• Known failure cases and the workaround for each. Every AI step has inputs it handles badly; the consultant will know them.
• Nothing depending on a custom GPT. Custom GPTs, for instance, stop working on 11 December 2026 when OpenAI retires them. Shared assistants belong in a Project or equivalent in your own workspace.

A prompt card makes this concrete. For an illustrative two-artist tattoo studio whose enquiries are sorted and drafted by AI before a person approves them:

PROMPT CARD: enquiry triage and draft reply        v1.4, [date]
Runs in: Zapier (studio account), AI step, [model tier]
Triggered by: new website form enquiry or email to bookings@
Purpose: sort into BOOKING / AFTERCARE / PRICE / COVER-UP / OTHER
         and draft a reply for a person to approve. Never sends.

Rules it follows (summary):
- Never quotes a price or promises a date
- Health questions (skin conditions, allergies, pregnancy): no draft,
  flag for the artist
- Cover-up enquiries: ask for a photo of the existing tattoo

Test set: 60 past enquiries in /handover/tests/triage_tests.csv
Last result: 57 of 60 correct category; 3 OTHER that should have been
COVER-UP (wording like "rework" or "fix an old piece").
Known workaround: staff re-label; rule for "rework" added in v1.5 draft
Owner at the studio: [first name]

A runbook your staff can actually follow

A runbook is a one-page instruction sheet per automation, written for the person who'll be on shift when it misbehaves. It should use your staff's words, not the platform's, and it should be tested by one of them before sign-off. The studio's runbook for the triage automation might read:

RUNBOOK: enquiry triage (drafts replies, never sends)

What normal looks like
- 30-50 drafts a week in the Gmail "AI drafts" label
- Each draft has a category tag in the subject line

How to pause it
- Zapier > Zaps > "Enquiry triage" > switch Off. Enquiries still
  arrive in the inbox as usual; nothing is lost while it's off.

If no drafts appear for a working day
1. Check Zapier > Zap history for red errors
2. Most likely cause: the Gmail connection needs re-authorising
   (Zapier > Apps > Gmail > Reconnect, sign in as bookings@)
3. Turn the Zap back on; check the next enquiry produces a draft

If drafts look wrong (wrong category, odd tone)
- Don't edit the prompt in Zapier. Note three examples in the
  "triage issues" sheet and tell [first name].

Monthly check (first Monday): task usage in Zapier billing,
test 5 recent enquiries against the prompt card.

Notice what it leaves out: no explanation of how AI works, no platform jargon beyond the menu names staff will see. If your business has no obvious person to hold these documents, settle that first; who should own AI in a small business helps choose.

Alerts that reach someone who'll act

Automations rarely fail loudly. They pause, skip or quietly send errors to an inbox nobody reads. Before sign-off, find out exactly how each platform you use behaves when things go wrong, and route every alert to a person who'll see it.

• Zapier automatically pauses a Zap when 95% of its runs error over seven days, and it sends no error emails when an error handler has run. So if the consultant built error handlers, each one needs its own notification step, or failures stay invisible.
• Power Automate switches flows off after 14 days of continuous failure, and after 90 days without being triggered unless a Premium licence covers them. A flow for a seasonal job can switch itself off before the season comes round.
• Make renamed its error handlers in September 2026 (Ignore became Skip, Break became Retry; Resume, Commit and Rollback kept their names). Make sure the runbook uses the names your staff will see.
• A weekly heartbeat. A short automatic summary (runs this week, errors, drafts created) sent to the owner turns silence into something you notice. No summary on Monday means something is wrong.

For AI that replies to customers directly, monitoring needs more than failure alerts: monitoring AI that talks to customers covers hand-offs and error sampling.

Gaps that only show up weeks later

Some handover problems are invisible on the day and obvious a month later. Realistic examples, and the check that would have caught each:

• Customer emails sent from the consultant's address. An automation's email step was still connected to the consultant's test Gmail, so customers' replies went to them. Check: send a test through every customer-facing step and look at the sender.
• A knowledge file in the consultant's drive. The price list the AI reads was a shared link to a file in the consultant's own drive. It worked until they tidied their files. Check: open every file the automations read and confirm your business owns it.
• A model tier changed after handover. Someone switched an AI step to a more capable model to fix one awkward reply; at 3 or 5 tasks per run instead of 1, the monthly count jumped past the plan's allowance. Check: record the tier on the prompt card and look at task usage monthly.
• Polling where an instant trigger was intended. A Make scenario set to check for new data every 15 minutes uses about 2,880 credits a month before it does any work, well above the free plan's 1,000. Check: note each trigger's schedule in the inventory and ask why it was chosen.

Costs, renewals and dates that will catch you out

The consultant knows exactly what everything costs to run, and you may never have seen the figures together. Ask for a one-page cost sheet. The studio's might read (illustrative, USD list prices):

Then ask for the dated changes that affect what was built. Some examples from the tools small businesses commonly use: custom GPTs stop running on 11 December 2026; WhatsApp Business Platform service replies become chargeable from 1 October 2026 after the first 1,000 a month per number (the free WhatsApp Business app is separate); and Power Automate's AI Builder credits are being replaced by Copilot Credits, with seeded credits ending for new or renewed licences from 1 November 2026. The costs that arrive after go-live are covered in more depth in what you pay after an automation goes live.

The handover session and the break test

Documents are necessary but not sufficient. Book a 60 to 90-minute session, record it, and have the person who'll look after the automations do four things themselves while the consultant watches:

1. Pause an automation and switch it back on.
2. Find a failed run in the history and re-run it.
3. Re-authorise an app connection.
4. Change a fact the AI relies on, such as a price in the knowledge file, and confirm the next output uses the new one.

If any of the four needs the consultant's hands on the keyboard, the runbook needs another pass. It's much cheaper to find that out now than the first time something breaks.

Two small agreements cover what documents can't. First, a question window: a fixed period after sign-off, such as 30 days, during which your staff can ask short questions, either included in the fee or at an agreed rate. Second, a glossary of the names used during the build ("the triage Zap", "the aftercare flow", "v1.4"), so staff and consultant mean the same thing when a question does come up.

Writing the handover into the scope from day one

Handover goes best when it's a deliverable with its own acceptance line rather than a courtesy at the end. In the proposal or contract, list the handover items by group, name who will take part in the break test, and state that the final payment follows sign-off. One sentence does most of the work: "Handover is accepted when the client's nominated staff member has completed the break test using the runbooks and every item on the attached list is marked done." Consultants who work this way tend to build with handover in mind from the first week: connections authorised through your accounts, prompts kept in files rather than only inside tools, and runbooks written as they go instead of on the final afternoon.

How the list changes with the kind of engagement

• Done for you. The consultant built everything, so every group applies in full, and the break test matters most because your staff have never touched the automations.
• Done with you. Your staff built parts alongside the consultant. Accounts and connections are probably already yours; concentrate on the prompt tests, the alerts and a review of anything your staff built alone.
• Continuing on a retainer. It's tempting to skip the handover because the consultant is staying. Don't: retainers end, and this list is what lets you end one calmly. The consultant can keep a user login for the retainer while ownership, exports and runbooks stay with you.
• One small automation. Scale the list down, not out. One runbook, one prompt card, one connection check and a five-minute break test still apply.

The tattoo studio's final check, scored

Here's how the illustrative studio's last day might go, with the handover list reduced to 16 lines and a fifth of the final invoice held until it's complete:

Twelve of sixteen sounds good, but look at which four are missing. The API key is the item most likely to stop everything without warning; the silent error handler is the item most likely to hide it when it does. Both are quick for the consultant to fix and hard for the studio to discover later. With the holdback in place, the four were closed within three days, the studio's front-desk manager ran the break test again, and the final payment went out. Without it, those four lines are exactly the kind that drift for months.

If you're reading this after the consultant has already gone and the list is short, don't panic: start with the accounts and keys, then the inventory, then the alerts. Everything else can be rebuilt from a working system; very little can be rebuilt from a system you can't log into.

Further reads

• How to Judge Whether Your AI Consultant Delivered Value — Once handover is done, judge whether the engagement delivered.
• How to Rescue a Stalled AI Project or Exit It Cleanly — What to do if the consultant has already gone and the list is short.
• Is an AI Consulting Retainer Worth It After Your First Project? — Whether to keep the consultant on after handover, and on what terms.
• How to Scope an AI Project: Deliverables and Acceptance Criteria — Put the handover list into the scope before work starts.
• AI Consulting Contracts: 9 Clauses to Check Before You Sign — Contract clauses that make the handover an obligation.
• What If Your AI Vendor Shuts Down? Checks Before You Commit — Exported copies are also your plan B if a platform closes.
• 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.
• What an AI Consultant Can't Do for You, and What You Must Own — Seven responsibilities that stay with the business when you hire AI help, an ownership card for every automation, and promises no consultant should make.
• 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.
• When Should a Trades Business Bring In an AI Consultant? — The signals that a trades firm needs outside AI help, a ten-minute scoring sheet, what to gather first and what a good engagement leaves behind.
• Questions to Ask Before You Pay Anyone to Set Up AI Marketing — 30 questions in six groups, with what a good answer sounds like, plus a filled-in comparison of two real-looking quotes for a food truck.
• Should You Hire an Automation Consultant or Build It Yourself? — A five-question test, what DIY really costs in hours, one build that worked and one that didn't, and the hybrid route most small firms should take.
• How to Choose an AI Consultant: 20 Questions to Ask First — Twenty questions to put to any AI consultant, what strong and weak answers sound like, and a scoring sheet filled in for a farm shop.
• AI Consultant vs Agency vs Freelancer: Which Should You Hire? — Who diagnoses, who builds and who supports: a decision table for consultants, agencies and freelancers, with a coffee roaster's three jobs sorted.
• 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.
• Done-for-You vs Done-With-You AI Implementation: Which Suits You? — What done-for-you and done-with-you AI implementation leave you holding, how upkeep decides it, and a four-shop butcher that uses both.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI API documentation on spend limits; Zapier help (auto-paused Zaps, error handlers, copying assets between accounts, changing owners) and pricing page; Make help on scenario blueprints and error handlers; n8n documentation on exporting workflows; Microsoft Learn on Power Automate flow suspension and orphaned flows; OpenAI's custom GPT retirement FAQ; Meta's WhatsApp Business Platform pricing notice.