Outlook
Part of A monitoring framework for a marketing attribution outlook
Attribution skills planned from the work, not from imagined headcount
Turn current analytical standards into buyer-specific attribution capability scenarios without inventing England job demand, shortages, salaries or headcount.
This article sets out a task-based capability method for marketing attribution. It maps current analytical guidance to buyer scenarios, not a demand forecast for vacancies, salaries or team size.
The source review for this capability map closed on 6 September 2026. Government frameworks describe their own context and cannot be presented as requirements for every private employer. No interviews, vacancy analysis or workforce survey was conducted.
What to take away
- Capability should be assessed against real attribution tasks rather than imagined headcount or forecast labels.
- Evidence of capability comes from reviewed work, such as event dictionaries, reproducible queries and documented model criticism.
- Training attendance alone does not prove a person can perform an unfamiliar attribution task.
- Retire a capability scenario when the task disappears and reopen it when data, model, regulation or authority changes.
What current guidance can establish
The Government Analysis Function's Career Framework describes core skills including quality assurance, communication, data visualisation and programming. Use it for task language, not commercial hiring demand.
GovS 010: Analysis, last updated on 26 August 2025, sets expectations for analysis supporting government decisions. A business in England can borrow its planning and assurance questions, but the standard does not govern private marketing work.
HM Treasury's Magenta Book, updated on 15 May 2026, covers evaluation scoping, methods, data and management. Match capability to the evaluation question; it does not forecast staffing.
Build capabilities around attribution tasks
Attribution task capability triggers
Platform credit
- Work to document
- Event definitions
- Leading indicator
- Unexplained difference
- Decision consequence
- Assign owner
- Recheck trigger
- Configuration change
Reconciliation
- Work to document
- Provenance rules
- Leading indicator
- Growing unmatched queue
- Decision consequence
- Assess real task
- Recheck trigger
- Repeated correction
Marketing mix modelling
- Work to document
- Statistical specification
- Leading indicator
- Unsupported causal claim
- Decision consequence
- Challenge design
- Recheck trigger
- Decision paper
AI-assisted workflow
- Work to document
- Prompt records
- Leading indicator
- Unreviewed model output
- Decision consequence
- Add human review
- Recheck trigger
- Security change
| Task | Trigger | Evidence | Owner |
|---|---|---|---|
| Platform credit reconciliation | Unexplained difference after a platform or tag change; recheck after any configuration or definition change | Event definitions, eligible populations, platform rule changes, missing observations | Owner who can reproduce and correct the report |
| Data reconciliation | Growing unmatched queue or repeated correction | Reconciliation work may require data provenance, identity rules, query versioning and exception handling | Analyst who can assess against a real task, with required proficiency left blank until scoped |
| Marketing mix modelling | Decision paper making a causal or budget claim that its design cannot support | Statistical specification, diagnostics, uncertainty, model criticism; experiment design, counterfactual and causal interpretation | Modeller with statistical and causal skills, not general dashboard skills |
| AI-assisted workflow | Need to document models, prompts, error analysis, security controls, human review and rollback | The NCSC's secure AI development guidance, published on 27 November 2023, supports documentation of data, models, prompts, limitations and failure modes. It does not establish that a particular role or tool will be needed. | Owner who can apply that guidance to the workflow |
Use an evidence-based capability record
For each unresolved task, record the owner, evidence input, expected output, legal and security boundary, required review, observed error, proficiency needed, learning route and reassessment date. Keep accessibility, privacy and PECR, security, statistical quality and commercial judgement as separate capabilities.
Evidence-based capability record fields
- Record owner
- Evidence input
- Expected output
- Legal and security boundary
- Required review
- Observed error
- Proficiency needed
- Learning route and reassessment date
| Task | Owner | Evidence input | Expected output | Legal and security boundary | Required review | Observed error | Proficiency needed | Learning route | Reassessment date |
|---|---|---|---|---|---|---|---|---|---|
| Platform credit reconciliation | Analytics lead | Tag audit log and event dictionary | Reconciled spend by channel with exceptions listed | PECR consent check and access control | Peer review of query and reconciliation method | Unmatched queue grew after a platform tag change | Write and version SQL query; interpret unmatched queue | Internal query clinic and paired review | Next quarterly review |
Evidence of capability should come from the work itself: a reviewed event dictionary, reproducible query, documented model criticism, accessible report or controlled experiment. Attendance at training can be recorded, but it does not prove someone can perform an unfamiliar attribution task.
The buyer should use supervised work, peer review and correction history to decide whether further support is needed.
Confidence should attach to the observed task, not to a forecast label. If a worker can reproduce a reconciliation but cannot assess causal claims, record that boundary plainly. A supplier demonstration is not evidence of internal competence.
Forecast status: Not established for England. The buyer should retire a capability scenario when the task disappears, automate only after controlled evidence, and reopen the assessment when data, model, regulation or decision authority changes.
Before you act
- Leave required proficiency blank until the work is scoped.
- Recheck after any configuration or definition change.
Common questions
What counts as evidence that someone can perform an attribution task?
Evidence should come from the work itself, such as a reviewed event dictionary, reproducible query, documented model criticism, accessible report or properly controlled experiment. Attendance at training can be recorded but does not prove a person can perform an unfamiliar attribution task.
When should a buyer reopen a capability assessment?
The buyer should reopen the assessment when data, model, regulation or decision authority changes. A capability scenario should be retired when the task disappears, and automation should follow only after controlled evidence.
