PlaybookChecklistReferenceAI UGC

Agency–Client AI UGC SLA Cheatsheet

The agency–client SLA for AI UGC: volume by spend, turnaround and revision caps, approval windows, disclosure ownership, kill-rule authority, and the artefacts that make a retainer measurable.

Updated 2026-08-2411 min read

LanguageEnglish

AI UGC retainers die in the inbox: unlimited revisions, unnamed owners, and ‘we’ll know it when we see it’ creative. This SLA sheet locks volume, turnaround, revision caps, approval windows and who owns disclosure so the retainer is a contract, not a mood.

An SLA is a volume-and-time contract, not a vibe about quality

Clients do not buy ‘AI ads’. They buy a stated number of testable 15–30s files per cycle, delivered by a weekday, with a cap on revision rounds, into a naming schema the buyer can launch. If those numbers are missing, the agency will be judged on taste and the client will request a seventh hook on Friday afternoon. Write the unit: a finished MP4 at the agreed ratios, captions burned, disclosure ready, token in the filename. Then write the cycle: weekly or biweekly, not ‘ongoing’. Taste reviews are allowed only after the file clears spec and claim QA. A retainer without a unit and a clock is how both sides decide the other is difficult. The locked tables are the numbers we actually put in statements of work — not adjectives.

Revisions and approvals are where retainers go to die

Unlimited revisions turn a 24-hour render pipeline into a two-week committee. Cap rounds: one written brief, one first-cut batch, one revision pass that is list-based (timestamps + the actual line), then lock. Approvals need a named human and a window — 24 hours on a weekly sprint, 48 on a biweekly — after which silence is approval or the slot slips, depending on what you wrote. Legal/compliance is a separate gate with its own clock; it cannot sit inside ‘the creative round’. If the client’s brand team only meets on Thursdays, the SLA either starts the cycle on Thursday or admits the launch is the following week. Do not pretend a same-day Ads Manager push is possible when brand-review is weekly.

Who owns the click, the claim and the label

The advertiser of record owns policy, claims, AI/synthetic disclosure, and the landing page. The agency owns file spec, volume, naming, UTM schema, and the kill-rule recommendations. Write that split so a Meta rejection is not a vendor fight. The agency can refuse to ship a banned claim; it cannot certify that a health line is lawful in every geo. Kill rules (spend ≥2× CPA with zero purchases, hook under a floor at 1,000 3s views) should be pre-agreed so the buyer does not need a Slack vote to pause a loser. Reporting cadence is part of the SLA: a Friday readout with win/loss codes, not a 20-slide recap of CPMs. If the client wants unlimited concepts, they are buying a second sprint, not a favour inside the first.

How to use the tables in a live retainer

Paste the volume table into the SOW against the client’s daily spend band. Paste turnaround, revision, and approval rows into the kickoff deck with names in the owner column. Run the artefact checklist at the end of week 1 — if the brand kit, claims list, or product URL is missing, the clock does not start. When the SLA calls for 8–20 variants in a cycle, a product-URL pipeline in Klip Kanvas is how the agency hits the clock without booking a creator for every hook; the SLA still names the human who QAs claims and disclosure before spend.

6 SLA tables inside: volume by client spend band, turnaround clocks, revision caps, approval & silence rules, ownership split (claims/disclosure/kill), and the week-1 artefact / scorecard that starts the clock.

Volume SLA by client spend band (per 7-day cycle)

$50–150/day: 4–6 new ads · 1 concept · 2 ratios max · 1 refresh batch if a winner fatigues · do not promise 20
$150–500/day: 8–12 new ads · 2 concepts · 3 hooks × 2 faces on the primary · 1 offer-line variant
$500–2k/day: 15–25 new ads · 3–4 concepts · weekly hook rotation on evergreen · dedicated refresh day
$2k–5k/day: 25–40 new ads · 4–6 concepts · 1 human-hero slot/month allowed on top · AI handles volume
Unit definition: 1 unit = 1 finished MP4 · agreed ratios (usually 9:16 + 4:5) count as 1 concept-cell, not 2 retainers
Ratio pack: Default 9:16 always · 4:5 if Feed >30% of spend · 1:1 optional · 16:9 only if YouTube is in scope
Length pack: Primary 15–30s · 1 cutdown 6–12s per winning cell · cutdowns are included, not extra invoices
Over-volume: Extra cells = extra cycle fee or a written change order · ‘just one more hook’ is the next sprint

Turnaround clocks (business hours, kickoff-complete)

Clock start: Starts when artefact checklist is green · missing claims list or PDP URL = clock not started
First-cut batch: AI UGC: 24–48h from green brief · human-hero overlay: 10–15 business days · do not mix those clocks
Revision pass: 24h for list-based notes on AI UGC · 5–7 business days if a reshoot/human is required
Launch package: Files + names + UTMs + disclosure notes in the folder 4 business hours before the agreed go-live
Emergency refresh: Fatigued winner (freq 2.5–3.5 cold): 3 hook variants in 24h · counted against the cycle volume
Localisation add-on: Winning cell into 1 extra language: +24–48h + native QA · not included in the base cycle unless scoped
Business hours: Define TZ in the SOW · a Friday 4pm brief does not get a Monday 9am batch unless the clock says so
Slip rule: If agency misses first-cut by >24h, client gets that cycle’s extra revision round free · not a refund of the month

Revision caps and what counts as a round

Rounds included: 1 first-cut + 1 revision round per cell · round 3+ is billable or dropped · write it
A valid revision: Timestamped list · line-level · ‘make it pop’ does not start the 24h clock · agency may reject vibe notes
Not a revision (in scope): Spec fix (ratio, loudness, safe zone, captions ≥48px) · agency-caused · does not consume the round
Not a revision (out of scope): New offer, new claim, new SKU, new avatar family, new language · that is a new cell
Brand-kit drift: If client changes hex/fonts mid-cycle, clock resets on affected cells · kit freeze date in the SOW
Legal rewrite: Compliance rewrite after first-cut counts as the revision round unless legal was in the week-1 artefacts
Batch vs one-off: Notes must cover the whole batch in one pass · drip notes over 4 days still count as one round if inside the window
Kill instead of revise: If the cell already fails hook-floor on a prior live twin, do not revise — mark B (rebuild) and spend the slot on a new angle

Approval windows and silence rules

Named approver: 1 primary + 1 backup · Slack group is not an approver · out-of-office backup required
Weekly sprint window: 24h from folder drop · business hours · default: silence = approved so the slot does not slip
Biweekly / brand-review: 48h · if brand only meets weekly, launch is the meeting + 1 day · do not promise same-day go-live
Legal/compliance parallel: Own 48h window · can block go-live · cannot add taste notes inside this window
Silence = approve (default): Write it · after window, agency launches the approved spec · client may still kill within 12h post-live
Silence = hold (optional): Regulated categories (health, finance, kids-adjacent): silence = hold · clock pauses · volume SLA pauses with it
Change after approve: Counts as a new revision round or a new cell · Ads Manager ‘quick edit’ by the client voids the file-spec SLA
Go-live owner: Buyer of record (client or agency) named · UTMs and disclosure toggles are that owner’s click, not the editor’s

Ownership split: claims, disclosure, kill, reporting

Advertiser of record: Owns claims, landing page, geo eligibility, AI/synthetic disclosure, platform toggles, ad account health
Agency: Owns spec, volume, naming, UTM schema, recommended kill rules, Friday readout, file archive
Refusal right: Agency may refuse a banned claim or missing disclosure · that cell is paused, not a missed SLA
Not certified: Green Ads Manager review ≠ legal opinion · agency does not certify US/UK/EU/IN lawfulness
Kill-rule authority: Pre-agreed: spend ≥2× CPA + 0 purchases = pause · hook <18% at 1,000 3s = pause · no Slack vote required
Scale authority: +20–30% only if CPA ≤0.90× baseline and freq <2.5 cold · else written client OK · cooldown 48–72h
Reporting: Friday one-pager: S/I/R/B/P/K codes · spend by code · next 6 cells · 20-slide CPM recap is out of SLA
Disclosure artefacts: On-creative line + platform toggle checklist in the folder · advertiser signs · agency files the screenshot

Week-1 artefacts & monthly scorecard (clock does not start red)

Product truth pack: PDP URL · SKU list · price · shipping window · returns · in-stock confirm <24h old
Claims & banned list: Allowed phrases · banned phrases · required qualifiers · legal contact · dated
Brand kit: Hex · fonts · logo SVG · safe-zone rules · faces/avatars allowed · music: licensed or none
Tracking: Pixel + CAPI live · test purchase <24h · UTM schema signed · destination URL final
Access: Ad account partner access · folder · sprint doc · named approver + backup TZ
Scorecard — on-time: Green: ≥90% cells on clock · Amber: 70–89% · Red: <70% · 2 red months = re-scope volume, not ‘try harder’
Scorecard — usable: Green: ≥70% of shipped cells hit sample floor · Amber: 50–70% · Red: <50% (too many ads for the budget)
Scorecard — learning: Green: ≥1 family sentence per cycle · Amber: only file-level winners · Red: no readout artefact
Red-row rule: Any red artefact = clock not started · agency sends a 5-line block list, not a first-cut batch

Locked

Unlock the full cheatsheet

Enter your email and we'll instantly unlock the complete cheatsheet — plus send you fresh ad playbooks every month.

No spam. Unsubscribe anytime.

Ready to put this into practice?

Create your first AI UGC video ad in minutes — no filming, no actors, no editing.

Try Klip Kanvas free