Operations
Part of Operating creative advertising software without losing control of the source material
Running creative advertising work from controlled intake to documented withdrawal
Run creative advertising work through a controlled intake, evidence check, build, approval, release, monitoring and documented withdrawal sequence.
A usable workflow, as set out in the operations and delivery guide, moves one identified job through named decisions. It does not begin with a blank canvas or end when a file exports. The route below creates an evidence trail from request to withdrawal.
What to take away
- A controlled job needs a named owner and a publication route before any creative work starts.
- Each objective claim requires exact wording, supporting record, owner, expiry and approved qualification.
- Reviewers must record pass, reject or conditional pass against a version ID to unblock items.
- Release requires a reversible hand-off with a second witness for high-risk activation.
- Measure queue age, blocked decisions, rework and withdrawal attempts by job ID.
1. Open a controlled job
The requester supplies a job ID, objective, audience, placement, deadline, offer, required formats and accountable campaign owner. They also list prohibited data, locked wording and any sector review. Reject an intake that lacks an owner or publication route.
Controlled intake requirements
- Job ID and accountable campaign owner
- Objective, audience and placement
- Deadline, offer and required formats
- Prohibited data and locked wording
- Sector review and publication route
- Source register beside the brief
Create a source register beside the brief. Each objective claim needs the exact wording, supporting record, owner, expiry and approved qualification. Under section 226 of the Digital Markets, Competition and Consumers Act 2024, a misleading action includes an overall presentation likely to deceive the average consumer, so a claim without a record should not leave the building.
Check each claim and mandatory line against the CAP Code and the ASA advice that goes with it before the item goes to approval. Name the sector review in the job record when a regulated product is involved.
2. Prepare inputs before production
The producer checks asset licences, product data, brand rules and destination specifications. Copy source files into a controlled workspace rather than editing the only original.
Name every tool in the chain. A typical chain has four links:
- An asset library, for example Bynder, Frontify or Brandfolder.
- A template or HTML5 ad builder, for example Google Web Designer, Bannerflow or Creatopy.
- A proofing and approval tool, for example Ziflow or Filestage.
- The ad platform that receives the file, for example Google Ads or Meta Ads Manager.
Assign permissions by role, and keep live customer data out unless the processing has been approved.
Pre-production input checks
- Check asset licences and product data
- Apply brand rules and destination specs
- Copy source files to controlled workspace
- Assign permissions by role
- Keep live customer data out
- Record platform, account, template, version
Record the platform, account, template and version that will be used. For a software-as-a-service creative platform, the NCSC's lightweight cloud security approach names four areas to expect from a supplier: data encryption, authentication and access control, security logging and incident management, and governance.
3. Build against a variant matrix
Make a row for every placement, language, audience treatment and offer. Columns should identify copy source, image rights, crop, mandatory line, destination and maker. A row cannot move forward while a required cell is blank.
| Placement | Language | Copy source | Image rights | Mandatory line | Maker |
|---|---|---|---|---|---|
| Square social | English | Approved headline v3 | Licence record | Offer end date | Designer |
| Vertical video | English | Approved script v2 | Licence record | Price qualifier | Editor |
Generate or assemble the first variant with your versioning software, whether that is a dynamic creative platform such as Celtra, Flashtalking or Adform or a template library, then correct the template before producing siblings. This prevents a faulty disclaimer, price or crop from spreading through the set. Save rejected outputs outside the approved library so they cannot be selected later.
4. Separate review decisions
The maker performs a technical self-check but cannot grant final release. Route claim accuracy, rights, brand and technical checks to their owners.
Send personal data questions to the data protection review, which applies the UK GDPR. Confirm the data protection fee status of the agency and the advertiser, since the fee applies to organisations that use personal information unless exempt. Each reviewer records pass, reject or conditional pass against a version ID. Comments without a decision leave the item blocked.
Test a late price change, a missing feed field and an unauthorised edit before launch. Those three failures cover the cases that most often slip past a visual check.
5. Release with a reversible hand-off
The release controller verifies the approved version, destination, schedule and stop route. A second person witnesses high-risk activation. Record who pressed publish or supplied the file, when it happened and which approval set authorised it.
Monitor the first live appearance against the release record. If copy, crop, destination or disclosure differs, pause further delivery and preserve evidence. Do not repair the live item without opening a tracked change.
An exception ticket needs the job and asset IDs, discovery time, affected destinations, immediate containment, decision owner and next checkpoint. Classify the consequence rather than the inconvenience, and rehearse the cases in the risk scenarios article. A wrong crop and an exposed customer identifier may both begin as production faults, but they demand different reviewers and escalation.
6. Close, withdraw and learn
At campaign end, remove access that is no longer needed, archive the released assets and retain evidence according to the organisation's approved schedule. Record withdrawal confirmation for expired offers in a withdrawal log that names the asset ID, destination, date removed and the person who confirmed it.
Hold a short review for failures, exceptions and manual work. Assign every action an owner and due date. Reopen the workflow only after the revised control has been tested on a representative job.
Measure queue age, blocked decisions, rework and withdrawal attempts by job ID, following the measurement and reporting guide. Do not reward faster output if unresolved review moves downstream. The campaign owner should examine recurring exceptions at an agreed cadence and either change the process, add capacity or narrow the service boundary.
Before you act
- Reject intake without an owner or publication route.
- Create a source register beside the brief.
- Check claims and mandatory lines against the CAP Code.
- Copy source files into a controlled workspace.
- Record platform, account, template and version.
- Save rejected outputs outside the approved library.
- Record withdrawal confirmation for expired offers.
Common questions
What must a requester supply when opening a controlled job?
The requester supplies a job ID, objective, audience, placement, deadline, offer and required formats. They also name an accountable campaign owner. They list prohibited data, locked wording and any sector review. An intake that lacks an owner or publication route must be rejected.
How should objective claims be supported before distribution?
Each objective claim needs the exact wording, supporting record, owner, expiry and approved qualification. The record must exist before the advert is distributed, because a misleading presentation can arise from the claim as a whole and not only from one sentence.
What happens when a live item differs from the release record?
If copy, crop, destination or disclosure differs, pause further delivery and preserve evidence. Do not repair the live item without opening a tracked change. An exception ticket needs job and asset IDs, discovery time, affected destinations, immediate containment, decision owner and next checkpoint.
