How to Onboard an Agency Client to AI UGC Creative

In the first meeting, show the working unit: 6 creatives, 3 hooks × 2 avatars, one locked body, 48–72 hours before a decision, kill and iterate rules. Explain that winners are rented — cold fatigue often shows at frequency 2.5–3.5 — so volume is the product. If they wanted one annual brand film, stop and rescope. Do not nod and hope they will like six talking heads later.

Bring one slide that is a matrix, not a reel of pretty work. Label rows as hook types and columns as avatars. Write the metrics you will report: hook rate (30%+ target on cold), hold (12–20%), outbound CTR (1%+ cold), then CPA. Write what you will not report on day one: ROAS leaderboards on 200 impressions. Ask who is allowed to pause ads. If the answer is 'anyone with the ad account password,' that is the first deliverable to change. Access is part of onboarding, and it belongs in the kickoff notes next to the 6-creative grid, not in a later 'we'll figure it out.'

Set a 30-day picture: week 1 intake and first grid, week 2 first readout and iterate, weeks 3–4 a second and third grid plus a graduate in a scaling campaign. Put dates on it. Clients fear AI UGC because it feels like a black box; a dated grid is the opposite. Record the call. You will need the sentence where they agreed not to rebuild the offer mid-test. Kickoff is a contract for behaviour, dressed as a workshop. If they will not accept a 48–72 hour no-touch window, you do not have a testing client yet — you have a preview committee.

Send a single intake form: URLs, offer, landing page, pixel status, banned claims, must-use phrases, competitor examples, avatars they will not accept, and a folder for product B-roll. Do not start generating from a deck. Missing proof and a vague brand voice are the usual delays. Freeze intake at 48 hours with 'not provided' marked rather than waiting forever for the perfect packshot.

The claims list is the most important file. Ask for three substantiated claims and a longer list of phrases legal will reject. Health, income, 'guaranteed,' and competitor naming need explicit yes/no. If legal will not turn around in two days, you write conservative first-person experience lines and they can loosen later — you cannot onboard into a vacuum. For avatars, get age band, setting, and hard nos (no kids, no medical coat, no celebrity lookalikes) rather than a Pinterest board of 'vibe.' Two approved faces is enough to start the 3 × 2. Ten optional muses is how week 1 disappears.

Check the store path yourself: add to cart, pixel or CAPI events, mobile load. Creative onboarding that ignores a broken thank-you page will be blamed on the avatar. Screenshot the live offer. If merchandising plans to change the bundle on Thursday, that date belongs on the creative calendar or the first test is contaminated. Intake is finished when you can fill a brief: product, offer, three hook hypotheses, two avatars, kill rule, destination URL. Anything still in email is not intake, and generating from it is how week one becomes a redo.

Name one client approver, one agency producer, and one media buyer. Define what the approver may reject: policy, wrong claim, off-brand logo, banned face. They may not reject a hook for being 'too casual' without a written brand rule. Set a 12-hour SLA on the first grid. Silent approval after the SLA is a feature, not an insult. Write this in the SOW appendix or a one-pager both sides sign.

Show a QC checklist instead of a taste form: first two seconds readable muted, caption safe zones, claim match, correct URL, correct aspect ratios, file names that map to the matrix. If they want legal on every line, put legal on the approved-claims list once, not on every render. Re-approving six near-identical body scripts is how AI UGC becomes slower than a shoot. Offer a live 20-minute review of the first two seconds of all six, on a phone, then launch. Desktop theatre in a conference room selects for polish, which is the opposite of this format.

Lock Ads Manager roles. The client should see reports; they should not have pause rights on the test campaign during the 48–72 hour window. If that is politically impossible, create a written exception list (policy, broken URL, offensive comment spike) and everything else waits for the readout. Put the no-touch rule in the same one-pager as the SLA. Most 'AI did not work' post-mortems are a founder who paused three cells on night one. Onboarding that does not mention this will pay for it in the first sprint, and you will be arguing about avatars instead of hook rate.

Render 3 hooks × 2 avatars from the locked brief, QC on a phone, launch one test campaign, equal budgets, conversion objective, broad targeting. Send a launch note: six names, no-touch until 48 hours, time of the first allowed check, time of the readout. This note is the rest of onboarding. Without it, Slack will fill with screenshots at hour six and someone will pause a cell.

Klip Kanvas is the production seat for this grid when the intake URL and claims list are in the brief — generate, recut hooks, export 9:16 and 1:1 without a custom shoot for every cell. If the client supplied B-roll, intercut it on the demo beat across all six so you are not testing footage and hooks at once. File names should match the matrix so the readout table is obvious. If a cell is rejected, fix that cell; do not pause the campaign. A missing cell is better than a reset, and a reset on night one is how onboarding turns into another kickoff.

Invite the approver to the launch note, not to Ads Manager live view. Give them the muted first-two-seconds cutdown if they need to feel involved. Then you wait. Use the wait to draft the reporting template and the next three challenger hooks from their reviews. Onboarding that goes quiet between launch and readout makes clients nervous; a scheduled 'delivery OK' ping at 24 hours (no winners) is the professional middle. Do not send a ROAS screenshot with that ping — you will train them to ask for one every morning.

At 48–72 hours, walk the six-row table: hook, hold, CTR, spend, conversions, bucket (kill, iterate, graduate). Recommend actions, do not present a mystery. Book next week's 3 × 2 before the call ends. If you leave without a next brief, the client will fill the vacuum with a redesign of the winner's body. Reporting is how you keep the ladder, not how you perform a weekly show-and-tell.

Teach the diagnosis in public. Weak hook: new first line, same body. Weak hold: the open over-promised. Weak CTR: offer or close. Weak CPA with healthy hook: page, tracking, or auction — not 'the avatar is creepy' as a first theory. If they want more brand, put it in retargeting with a persistent face, and keep a rotating cast on cold. Show frequency on the graduate and the 2.5–3.5 band so a refresh request in two weeks is expected, not a failure. Clients who understand fatigue do not fire you when a winner cools.

Put a monthly creative calendar in the shared folder: batch count, themes, gates, readout dates. That is the retainer's operating system. Include who owns asset refreshes (new B-roll, new reviews) so you are not chasing a Dropbox on week three. If the first test was a kill across the board, say so, and ship a new open immediately. Honesty in week one is cheaper than a polite story that buys you a quiet month and a cancelled quarter — and it teaches them that kills are part of the 6-creative model.

Before you call them onboarded, finish the plumbing: ad account roles, brand kit (fonts, colours, logo, tone, claim guardrails), billing for production tools, and a written rule for pausing. Confirm pixel and CAPI with a test event. Onboarding is not done when the first video is pretty. It is done when the second grid can ship without another kickoff.

Save the two avatars, the approved claims, and the winning hook family as a reusable kit so a new producer can pick up the account. Document the landing page URL and the offer as of launch day. If the client will localise later, note languages and on-screen text rules now. Add a 15-minute monthly 'access review' so former staff lose Ads Manager. The kill switch is also yours: if they keep breaking the no-touch window after two written warnings, you either take pause rights away or you stop guaranteeing test reads. That sentence belongs in onboarding, while everyone is still friendly.

Hand them a one-page 'how we work' they can forward internally. It should mention the 6-creative grid, 48–72 hours, frequency 2.5–3.5, and who to ping for emergencies. Most internal chaos is a CMO who was not in the kickoff. Your job is to make the model portable. When that page exists, the brand kit is in the tool, and grid two is on the calendar, you can mark onboarding complete and start charging the week as a sprint, not as setup. If grid two still needs another workshop, onboarding is not done — the kit and the access were theatre.