Separate bad results from bad management
A weak month is not proof of agency failure. Demand, stock, pricing, competitors, website releases and measurement can move performance. The agency should, however, be able to show what changed, what evidence supports its diagnosis and what it will do next.
Judge four areas separately: commercial results, measurement integrity, operating discipline and governance. A business can retain an agency through a market downturn if the diagnosis and controls are strong. It should not tolerate inaccessible accounts or fabricated reporting because ROAS happens to look good.
Build a review pack before giving notice
- Contract, notice period, customer ID and current access list.
- Google Ads change history and active experiments.
- Goals, primary conversion actions and CRM/order reconciliation.
- Campaign results split by brand, acquisition and customer quality.
- Written strategy, tests, reports and unresolved action items.
- Dependencies in Tag Manager, GA4, Merchant Center, CRM and call tracking.
Red flags that justify escalation
- No company-controlled administrative access. This is a governance risk, although contractual rights still depend on the agreement.
- Reporting cannot be reconciled. Revenue, leads or values materially disagree with the system of record and no one owns the fix.
- The agency optimises to the wrong outcome. Cheap forms are celebrated despite poor qualified or won rates.
- Changes have no hypothesis or record. Activity volume is not the test; decision quality and follow-through are.
- Material errors are concealed or repeated. Honest mistakes happen; hiding them prevents control.
- Conflicts are undisclosed. Incentives, subcontracting or tool dependencies should be understood.
Run a fair performance review
Choose a representative period that includes conversion lag. Compare actual results with the agreed commercial target and decompose the variance into CPC, conversion rate, value, demand volume and downstream quality. Ask the agency to respond with evidence, not a presentation assembled from impressions.
A hypothetical account misses its £120 qualified-lead target at £155. If CPC rose 10%, landing-page conversion fell after a site release, and qualified rate remained stable, firing the agency may not fix the primary cause. If raw-lead CPA looks good only because spam doubled and the agency refuses CRM imports, the management model is the problem.
Use a remediation period when the relationship is recoverable
Set three to five dated outcomes: restore admin access, reconcile tracking, import qualified leads, stop a defined source of irrelevant traffic, and deliver an experiment plan. Specify evidence and owner. A remediation period is not “give it another 90 days”; it is a controlled opportunity to resolve named failures.
Plan the handover before termination
- Secure direct administrative access and exports.
- Map manager-owned conversions, audiences, negatives, bidding portfolios, scripts and billing.
- Appoint the incoming owner and freeze unapproved structural changes.
- Test one conversion and capture a performance baseline.
- Transfer knowledge and replace dependencies.
- Remove old access only after continuity is verified.
Google’s unlinking guide explains permissions and presents an impact preview. Removing a manager does not itself erase the client account’s history, but manager-owned dependencies can be affected.
Use an evidence-based scorecard
| Area | Evidence | Fail example |
|---|---|---|
| Commercial | Qualified CPA, contribution, growth role | Reports stop at clicks or raw forms |
| Measurement | Order/CRM reconciliation and tests | Material variance remains unexplained |
| Strategy | Written choices, hypotheses and trade-offs | Recommendations copied without context |
| Operations | Owners, deadlines, QA and change record | Repeated errors or no follow-through |
| Governance | Access, billing, security and handover | Business cannot inspect its account |
Weight the categories before scoring. A regulated lead business may weight measurement and governance more heavily than creative volume. Require supporting links or exports for each score so the process does not become a popularity vote.
Questions for the agency review
- Which business outcome is bidding optimising, and how was it validated?
- What changed in the last 90 days, and what evidence connects it to results?
- Where is marginal budget currently profitable or unprofitable?
- Which assumptions remain unproven, and how will they be tested?
- What depends on agency-owned tools or manager assets?
- What would you do with 20% less or more budget?
- What are the next three decisions, their owners and dates?
Worked review example
A hypothetical account reports CPA improving from £90 to £62. CRM data shows qualified rate fell from 45% to 20%, so qualified CPA moved from £200 to £310. Change history shows a new broad campaign optimising to all forms. The issue is not simply “broad match”; it is that the agency selected a shallow objective, failed to monitor downstream quality and reported the easier metric.
A fair remediation plan would import qualified outcomes, isolate the new traffic, set quality guardrails and review a mature cohort. If the agency accepts ownership and completes this work, the relationship may be repairable. If it refuses CRM evidence and continues presenting raw CPA, termination becomes a governance decision.
Select the replacement against the failure
If the outgoing model failed on measurement, do not select the next provider from creative case studies alone. If it failed on business context, ensure the next model has an accountable internal owner. Use the in-house, freelancer and agency comparison to decide the operating model before comparing pitches.
Termination risk register
| Risk | Control before notice |
|---|---|
| Loss of access | Verify two company administrators |
| Tracking interruption | Map tags, imports, call tracking and shared actions |
| Campaign drift | Appoint incoming owner and change freeze |
| Billing disruption | Confirm payment profile and invoicing |
| Knowledge loss | Export settings, tests, creative and decisions |
Do not confuse style with failure
A quiet account is not necessarily neglected if automation is stable and analysis occurs outside visible edits. A busy change history is not necessarily good management. Likewise, a strategic disagreement is not misconduct. Define the evidence and commercial consequence.
Communication matters when it changes control: unanswered access requests, unexplained data variance or missed incident deadlines are substantive. Preference for a different report design is not.
Review 30 days after the switch
Confirm access, billing, tracking, scripts, feeds, audiences and automated rules remain functional. Compare spend and outcomes with the handover baseline, accounting for campaign changes and lag. Close or assign every unresolved dependency. Then remove obsolete users and archive the final handover register.