Five risk scenarios to test a branding agency on before you hire it

These branding-agency risk scenarios show how an English agency should rehearse its response before an incident escalates, and give a client who is choosing an agency five things to test on the shortlist. They do not forecast whether an event will occur or estimate losses. Each begins only when an observable trigger appears, and each has an owner who can stop and restart work.

What to take away

Each scenario starts only when an observable trigger appears and has an owner who can stop and restart work.
A classification break freezes competitor market-share comparison until the analyst retrieves metadata and publishes a break note.
Disputed AI brand assets stay unpublished until a qualified UK IP lawyer, named in advance, checks the material and terms.
Before you hire, ask each shortlisted agency to walk you through how it would handle these five scenarios on your account.
Generated claims need documentary evidence before publication under CAP Code Section 3 (Misleading advertising).
Run every scenario with synthetic pitch and campaign material and review after a near miss, source change or failed exercise.

A classification break distorts the market view

Source context: ONS UK SIC 2026 guidance says implementation of the April 2026 revision is staged. Current outputs will not all move together. A branding agency tracking competitor market share across SIC-coded segments will find its baseline shifted mid-report.

Triggera new source uses a different SIC version from the baseline.
Indicatorschanged code, unit, coverage, geography or unexplained series movement in a competitor or category comparison.
Owner and actionthe analyst freezes comparison, retrieves metadata and publishes a break note.
Stopno trend or market claim proceeds while correspondence is absent.
Restartthe statistical reviewer approves a documented bridge or separate-series presentation at the next release review.

An AI concept has uncertain rights

Source context: the March 2026 government copyright and AI report records evidence and further policy work on inputs, outputs, licensing and digital replicas. It is not a new blanket permission. A branding agency generating logo concepts, mood boards or campaign imagery with AI tools cannot assume clearance: copyright in the output still turns on the Copyright, Designs and Patents Act 1988, and a proposed mark on the Trade Marks Act 1994.

Triggera proposed client brand asset—logo, campaign visual or trade-mark application—uses generated material, a reference work or a person's likeness.
Indicatorsmissing provenance, unclear licence, close resemblance to an existing mark, unapproved prompt input or contract restriction.
Owner and actionthe creative lead quarantines the asset while a named UK IP lawyer checks the actual material and terms.
Stopdo not show, publish or transfer the disputed output.
Restartuse a cleared asset with a recorded chain of rights, or return to human-created material under the agreed brief.

Client information enters an unsafe AI workflow

Source context: the ICO's Guidance on AI and Data Protection covers accountability, lawfulness, transparency, fairness, security and rights, and is marked under review after the Data (Use and Access) Act.

The NCSC's Guidelines for Secure AI System Development covers secure design, development, deployment, operation and maintenance.

Triggerpersonal data, confidential pitch decks, unreleased brand guidelines or client campaign material is proposed for an AI service.
Indicatorsunknown retention, supplier access, processing location, model reuse, logging gap or untested deletion.
Owner and actionprivacy and security owners block the transfer, preserve evidence and assess the supplier and data flow.
Stopthe workflow stays disabled if purpose, role, access or recovery is unresolved.
Restartnamed reviewers authorise a bounded configuration and a synthetic test passes the documented controls.

A generated claim outruns its evidence

Source context: CAP Code Section 3 (Misleading advertising), rule 3.7, requires advertisers to hold documentary evidence for objective claims before publication. Section 3.1 prohibits marketing communications that mislead the consumer.

Triggergenerated or edited copy introduces a benefit, competitor comparison, testimonial or qualification change in a brand campaign or pitch deck.
Indicatorsno claim-register entry, altered meaning, missing limitation or unverifiable quotation.
Owner and actionthe advertising reviewer withdraws the variant and traces the approved source.
Stoppublication remains blocked while the claim or overall impression lacks support.
Restartauthorised copy and its qualification pass legal, regulatory and client approval; otherwise delete the claim.

An inaccessible asset reaches release review

Source context: the Public Sector Bodies (Websites and Mobile Applications) (No. 2) Accessibility Regulations 2018 require compliance with WCAG 2.1 Level AA, and government accessibility guidance explains the responsibilities retained by an outsourcing public body. Other clients need their own legal scope assessment.

Triggertesting finds a barrier or the promised evaluation has not occurred on a client-facing brand site, campaign microsite or downloadable asset.
Indicatorskeyboard, contrast, structure, alternative-text, document or assistive-technology failure against the defined WCAG 2.1 AA method.
Owner and actionthe accessibility lead records the defect, affected users, source file and correction route.
Stopthe asset does not pass its acceptance gate because a supplier or tool labels it accessible.
Restartan authorised fix passes retesting and the buyer accepts the evidence and any remaining limitation.

What to ask an agency on your shortlist

A client cannot see an agency's incident log, but it can ask how the agency would act. Put one question per scenario to every shortlisted agency and compare the answers with its references and reviews:

Market data:who checks that competitor or category figures use the same classification before they appear in a pitch or report?
AI rights:which generated or reference material may reach your brand assets, and who clears the chain of rights before you see a concept?
Client data:which AI services may receive your briefs or customer data, under what contract terms, and who can switch a workflow off?
Claims:where is the claim register kept, and who withdraws copy that lacks documentary evidence?
Accessibility:which WCAG method and tester sign off a site or asset before it is handed over?

An agency that names an owner, a stop rule and a restart condition for each answer has rehearsed the scenario. One that answers with a general assurance has not.

Exercise and revision

Run each scenario with synthetic pitch and campaign material. Record decisions, time stamps, evidence location, deputies and communications. Review after a real near miss, source change, supplier update or failed exercise. A probability may be added only when a qualified method and relevant data support it.

Before you act

Identify the observable trigger before starting any scenario.
Name an owner who can stop and restart work.
Quarantine disputed assets and record the chain of rights.
Block publication while a claim lacks support.
Record decisions, time stamps, evidence location and deputies.
Review scenarios after a real near miss or source change.

Common questions

When does a risk scenario begin?

Each scenario begins only when an observable trigger appears. It does not predict an incident, assign probabilities or estimate losses. Every scenario also has an owner who can stop and restart work, so the response stays controlled rather than speculative.

What happens when a new source uses a different SIC version?

The analyst freezes comparison, retrieves metadata and publishes a break note. No trend or market claim proceeds while correspondence is absent. The statistical reviewer approves a documented bridge or separate-series presentation at the next release review before work restarts.

Can a supplier label make an inaccessible asset pass?

No. The asset does not pass its acceptance gate because a supplier or tool labels it accessible. The accessibility lead records the defect, affected users, source file and correction route. Restart requires an authorised fix to pass retesting and the buyer to accept the evidence.