When to rehearse AI marketing tools risk scenarios before a campaign

What to take away

Rehearse one scenario per quarter, and rehearse before any campaign with a fixed launch date. Review the rest.
Rehearsal suits scenarios that are new to the team, that touch a tool able to publish or bid without approval, or that sit on a dated launch.
Keep five live scenarios on one pagescenario, owner, trigger, likelihood, blast radius.
Score with the channel owners, then fund mitigations in score order.
Write triggers as events with a named owner, never as dates.

When to rehearse, and when a review will do

Rehearse when a scenario is new to the team, when it touches a tool that can publish, bid or write to a CRM without approval, or when a campaign has a fixed launch date.

Reviewing means reading the register, checking owners and dates. Rehearsing means walking the scenario with the people who run the channels and writing down the actions you would take. Only the second changes behaviour.

Use this test, in order:

Does the campaign have a fixed launch date inside 60 days?
Has the team never rehearsed this scenario?
Can the tool act without a human confirming?
Does campaign spend exceed one month of paid media budget?

Any yes means rehearse. All no means the quarterly read-through is enough.

Agents widen the case for rehearsal. Microsoft Learn's introduction to generative AI and agents shows how a repeated small error compounds once a tool runs a chain of steps without review.

Example: a completed five-row register

The table below is what a mid-sized UK marketing team might hold for a spring launch. Scores are illustrative.

Scenario	Owner	Trigger	Likelihood	Blast radius
Retargeting list serves ads to people who opted out	Paid social lead	Consent audit finds a list older than 12 months	Medium	High
Agent publishes an unapproved price claim	Content lead	Tool gains publish rights with no sign-off step	Low	High
CRM enrichment mismatches B2B records	Marketing ops	Match rate falls below the agreed floor	High	Low
AI image includes a competitor logo	Brand lead	Design review finds third-party marks	Medium	Medium
Model change alters email personalisation tone	Email lead	Vendor notes a model or subprocessor change	Medium	Medium

Each row carries one owner who can pause the campaign. The trigger column is what turns the register into a working document.

If a row touches a claim in an advert, the AI marketing tools advertising rules in England set out what sign-off must cover before launch.

The ASA advertising codes are the rules that sign-off is tested against, so a price or health claim earns its own row.

Scoring likelihood against blast radius

Two columns are enough. Likelihood is how often you expect the scenario in a year. Blast radius covers spend, reputation and the time to correct it.

For illustration, a team spending £15,000 a month on paid social might score a mis-targeted audience as likely but cheap to repair.

A model drafting regulated financial promotions scores the reverse: rare and expensive, with a correction cost that lands on the whole campaign.

Score likelihood against blast radius

Scenario

Mis-targeted audience: Likely
Regulated promotions: Low

Likelihood

Mis-targeted audience: Cheap to fix
Regulated promotions: Heavy cost

Blast radius

Mis-targeted audience
Regulated promotions

Score with the people who run the channels, not only the people who write policy. A paid social manager sees a targeting failure a compliance officer never will.

Keep the session to one hour and date the notes. A dated row is evidence; an undated one is opinion.

The ICO's guidance on AI and data protection covers lawful basis and profiling under the UK GDPR, which is where most retargeting rows begin.

When a score changes a channel's reported numbers, the AI marketing tools attribution methods explain why the same spend can show different results without any tool breaking.

Triggers that force a rehearsal

Write triggers as events, with a name beside each one. Without a name, a trigger sits unread until somebody asks a question.

A regulator publishes guidance that touches a row (ICO, ASA, CMA, FCA).
A vendor changes its model, retention terms or subprocessor list.
Channel mix shifts by an agreed share of spend.
An incident occurs anywhere in the business, even outside marketing.
The owner named on the row leaves the team.

Keep personal data out of the register. Scenario names and owners are enough, which keeps it shareable across teams.

Five owned scenarios replace an annual document nobody reads.

Static to live register

Practice

Register: Rewritten annually
Trigger: When something breaks
Evidence: Tool screenshots
Skills: One person
Sign-off: Verbal in stand-up

Before

Register: Five live scenarios
Trigger: Written event
Evidence: Dated decision log
Skills: Two people
Sign-off: Logged against brief

After

Register
Trigger
Evidence
Skills
Sign-off

The lines in an AI marketing tools budget template show where rehearsal time and mitigation spend sit, which helps when finance asks what the register costs.

Common questions

How often should we rehearse rather than review?

Rehearse one scenario per quarter, plus any scenario attached to a fixed launch date. Review the whole register monthly and after a trigger fires.

Who should own the register and the triggers?

A marketing operations lead who can pause a campaign, working with a data protection officer. Name the person, not the team.

Do trial or pilot tools belong in the register?

Yes. Pilots often touch live customer data before procurement signs, so the same triggers apply from the first test.

What is the smallest useful register?

Five scenarios on one page with owner, trigger, likelihood and blast radius. Add a row only when a trigger names it.