Most agency terminations begin with a report. Cost per acquisition has risen, pipeline looks thin, and someone in the leadership meeting says the agency is not working. What rarely happens before that meeting is a check on whether the report is true. Flightdeck Advertising, itself an agency, describes the mechanism plainly: if conversion tracking misfires, calls go untracked, forms fail to register or offline sales never reach the ad platform, then every report received has understated or misrepresented actual results (Flightdeck). That is agency advice rather than an industry statistic, but the logic holds regardless of who states it. This guide covers performance and advertising agencies where measurable conversions and ad account ownership are at stake; PR, branding and creative-only engagements need a different test. It gives you a sequence: verify the numbers, locate the fault, then decide whether to repair or replace.
The question is rarely whether the agency is good. The question is whether you are in a position to tell. A large share of switching decisions rest on dashboards nobody outside the agency has audited, on targets set without an acquisition cost ceiling, and on briefs that changed several times in one quarter. GPI's view is that an incumbent should be judged the way GPI evaluates an agency for its directory: on documented evidence, a stated methodology and honest limitations, rather than on promises or a single poor month. Unverified measurement makes the decision unfounded. A fault that sits in your approvals or landing pages is inherited by whoever replaces the agency. Prove the fault first, then act.
TL;DR
The short answer to when to change advertising agency: only after you can prove three things, and not before.
- Do not decide from the agency's own reports until conversion tracking, call and form capture and offline sales feedback have been verified by someone other than the incumbent. If those feeds are broken, every report has been misstating results (Flightdeck).
- Separate three failure modes before assigning blame: broken measurement, client-side bottlenecks (briefs, approvals, budget swings, landing pages) and genuine agency underperformance.
- Fix the relationship when the problems are fixable by you, are recent, and the agency answers a written remediation plan with clear metrics, a stated definition of success and a deadline (WebFX).
- Change the agency when transparency is refused, access to your own accounts is withheld (ClicksGeek), results are guaranteed rather than explained (Oneupweb), or a remediation window passes without measurable movement.
- If you switch, keep the existing ad accounts under your ownership and avoid rebuilding campaigns wholesale; a new account loses history and platform learning (Space Ads).
- Compare alternatives on total acquisition cost against your own cost ceiling, not on retainer size alone.
Note which of the three proofs (measurement, fault, switching cost) you cannot yet supply. The rest of this article works through each one.
Verify the measurement before you judge the agency
Commission a tracking audit by someone other than the incumbent, then re-baseline the last two quarters before any performance conversation takes place. It costs less than any other step in the decision and is the one most easily skipped.
Platforms optimise toward the conversion events they receive. If those events are wrong, the agency has been optimising toward the wrong thing, and you have been judging it against the wrong thing. Both sides are working from bad inputs. Flightdeck's list of failure modes is a useful starting inventory: tracking not firing, phone calls untracked, forms not registering as conversions, offline sales never returned to the platform (Flightdeck). This comes from an agency's own advice page, so treat it as an explanation of how reporting breaks, not as evidence of how often it breaks.
What a tracking audit must confirm
| Tracking check | What must be confirmed | Who verifies it (not the incumbent) | Evidence to request | If it fails, what the reports have been misstating |
|---|---|---|---|---|
| Conversion events | Events fire on the actions the business actually sells, not on page views or button clicks | Internal analyst or neutral third party | Tag configuration export, test conversions matched to CRM records | Conversion volume and CPA |
| Calls and forms | Every lead channel registers as a conversion with a value | Internal analyst with CRM access | Call tracking logs reconciled to platform counts | Lead volume, especially for offline-heavy businesses |
| Offline and CRM revenue | Closed deals and refunds are passed back to the platform | Finance plus analyst | Offline conversion import history | Return on ad spend and quality of leads |
| Deduplication | The same conversion is not counted across several platforms | Analyst | Platform-level totals compared with CRM totals over one period | Inflated total conversions and understated blended CPA |
| Attribution windows | Windows are consistent across platforms and across reporting periods | Analyst | Reporting settings screenshots for each platform | Period-to-period trends |
The access requirement runs through every row. You cannot audit what you cannot see. ClicksGeek frames full account access and reporting that connects spend to business outcomes as something to demand, not negotiate (ClicksGeek). Make it the first governance ask.
Who should run the audit, and why it is not the incumbent
The incumbent may have configured the tracking, so asking it to grade its own work removes the independence the audit exists to provide. An internal analyst, a measurement consultant or the finance team reconciling platform figures to booked revenue all work. Ask the agency at the same time how it defines success and how performance data changes strategy; vague answers are a finding in themselves (WebFX).
What to do if the numbers change after the audit
If performance was understated, the agency may be better than you believed, and the conversation shifts from termination to fixing the data feed. If the agency controlled the tracking and it was broken, that is an execution finding, though not yet proof of overall underperformance. Either way, re-baseline before judging.
Diagnose the fault: agency, tracking, or client-side bottleneck
Fill a diagnostic matrix for your top five symptoms, assign a named owner to each confirming test, and count only the symptoms that land in the agency column toward a switch decision. Deciding when to change advertising agency without this step means paying a new partner to inherit your own bottlenecks.
Assume the tracking audit is complete. Now each symptom has three candidate origins, and each origin has a test.
| Symptom | Likely origin if measurement is at fault | Likely origin if client-side | Likely origin if agency | Confirming test |
|---|---|---|---|---|
| Rising cost per acquisition | Conversion events lost or duplicated after a site or tag change | Budget bursts or cuts that reset platform learning; target set with no cost ceiling | No bid or audience response to the trend; same structure month after month | Overlay CPA against change log of tags, budgets and campaign edits |
| Flat conversion volume | Forms or calls not registering | Landing page, stock or pricing problems; slow approvals delaying launches | No new creative or audience tests in the period | Compare platform conversions to CRM leads; review approval timestamps and test log |
| Creative fatigue | Frequency reported on wrong metric | Brand guidelines or legal review block refreshes | Agency has not proposed or produced refreshes | Count creative submissions versus approvals per month |
| Slow launches | Not a measurement issue | Brief changed mid-flight; assets delivered late | Agency missed agreed dates with assets in hand | Timeline of brief versions and asset delivery against launch dates |
| Missed or thin reporting | Data pipeline broken | Nobody on the client side defined what the report should answer | Reports describe activity, not outcomes; success never defined in business terms | Ask for the written success definition; check whether one exists |
The test approach for unrealistic targets comes from ClicksGeek's advice to set an acquisition cost ceiling from customer economics before evaluating any proposal (ClicksGeek); the agency-column signals draw on Oneupweb's warning about guaranteed results (Oneupweb) and WebFX's evaluation questions (WebFX).
Signals that point to broken measurement
Sudden step changes that coincide with a website release, tag migration or CRM change, rather than with a campaign edit, usually indicate data rather than performance. Discrepancies between platform-reported conversions and CRM records over the same window point the same way. The audit in the previous section should already have surfaced these.
Signals that point to your own organisation
These are the causes most switching guides ignore. Briefs that changed mid-flight force rebuilds. Approval queues delay launches past the moment that mattered. Budget cuts followed by bursts reset platform learning twice. Landing pages, inventory gaps and pricing changes sit entirely outside the agency's control, as does sales follow-up on leads. A target set without a cost ceiling derived from customer economics is unrealistic by construction, and no agency will hit it (ClicksGeek).
Signals that point to the agency
Causes that survive the tests above are the ones that count. No strategy rationale when asked. No visible response to data over several months. A refusal to define success in business terms, or reports that list activity instead of outcomes (WebFX). Guaranteed results instead of explained tradeoffs; Oneupweb's position is that there are no sure things in marketing and guarantees deserve scepticism (Oneupweb).
Using the matrix without bias
Score over a defined period, at least a quarter, rather than a single bad month. Have a named owner for each test who is neither the agency lead nor the internal sponsor who chose the agency. A switch that leaves a client-side bottleneck in place transfers the same result to a new invoice, plus transition costs.
When to fix the relationship: the remediation test
Draft a one-page remediation agreement covering the success definition, the metrics, the access requirements, a review cadence and a dated decision point, and have the agency sign it. Treat repair as a test with a deadline.
Conditions that favour repair
Repair is rational when the diagnosed causes are client-side or shared, when the relationship is young enough that the institutional knowledge you would lose is still forming, or when the agency has responded constructively to data in the past. It is also rational when the agency-column findings are about process rather than honesty: slow testing can be fixed, withheld access rarely is.
Structuring a 90-day remediation plan
The 90-day window is this article's framework, not an industry norm; no approved source quantifies how long remediation takes or how often it succeeds, and none should be implied. The window is long enough for strategy changes to produce measurable movement and short enough to avoid drift. The plan needs:
- A success definition in business terms, agreed by both sides, so the review at day 90 is a reading rather than a debate.
- The metrics that will be tracked and an explanation of how performance data will change strategy. WebFX's evaluation questions for a new agency work equally well as required answers from an incumbent (WebFX).
- Full account and reporting access for the client, with reporting that connects spend to business outcomes (ClicksGeek).
- Named leads on both sides, including the client-side owner of approvals and landing pages if those were part of the diagnosis.
- A review cadence, typically every two to three weeks, and a hard end date.
What counts as evidence the plan is working
Goodwill and capability are different things. An apology, a new account director or a longer weekly call are gestures. Evidence looks like a documented strategy change, a testing log with results, revised bidding or audience structures, and reported outcomes moving against the agreed metrics. If the client-side items were part of the plan, evidence also includes your own approval times shortening. Record both.
Exit criteria agreed in advance
Write the exit criteria before the plan starts. State the metric thresholds that mean the relationship continues, the thresholds that trigger a review, and what happens if measurement itself changes during the window. Agreeing this upfront prevents the day-90 meeting from becoming a renegotiation of the target.
Failure usually looks like this: the plan lapses quietly, targets shift to whatever moved, or the agency reverts to guarantees and generalities. Oneupweb's caution about guaranteed results applies with more force during remediation than during a pitch (Oneupweb): a partner that promises the outcome rather than explaining the mechanism has stopped engaging with the data.
The hidden costs of switching versus staying
Complete the comparison table with your own figures; if switching does not beat staying within 12 months at your acquisition cost ceiling, run remediation first. No approved source quantifies average switching cost across the industry, so this section gives a method rather than a benchmark.
Costs of switching that rarely appear in the business case
Platform learning is the cost most easily left out. Space Ads, describing its own handover mechanics, notes that preserving the same advertising account normally retains its history, but significant campaign edits can still affect platform learning, and a new account creates additional loss of continuity (Space Ads). A replacement agency that rebuilds everything on arrival, as many do, pays this cost with your budget. Add internal time for the review and onboarding, overlap fees when two agencies are billing, notice-period obligations, and the work of re-establishing tags, conversion definitions and data flows. If you do proceed, GPI's guide to running a media agency review sets out the stages and the internal time each one consumes.
Costs of staying that are easy to underweight
Staying has its own bill. Continued spend at a CPA above your ceiling is a direct loss each month (ClicksGeek). Untested channels carry an opportunity cost that never appears in a report. And if the incumbent keeps first-party data and measurement configuration in its own systems, poor data hygiene compounds: Miller Ad Agency argues that as platform signals degrade, the owned CRM becomes the most valuable media asset (Miller Ad Agency), which means every quarter without clean owned data weakens future measurement whoever runs the media.
Building the comparison against your acquisition cost ceiling
| Cost category | Switching path (12 months) | Staying/remediation path (12 months) | How to estimate it | Evidence or assumption |
|---|---|---|---|---|
| Platform learning loss | Rebuilt campaigns or new accounts underperform for a period | Minimal if structure is stable | Historical CPA during past relaunches in your own account | Mechanism from Space Ads; magnitude is your own data |
| Internal time | Review, RFP, onboarding hours at loaded cost | Remediation meetings and audit hours | Hours multiplied by loaded rate | Assumption |
| Fees | Overlap plus onboarding fees, notice period | Existing retainer | Contract terms | Your contracts |
| CPA above ceiling | Reduced after transition if new partner performs | Continues during remediation, then improves or does not | Monthly spend times gap between actual CPA and ceiling | ClicksGeek ceiling method; your own figures |
| Measurement rebuild | Tags, conversions, attribution re-established | Fixes from the audit only | Analyst days | Assumption |
Citations for the table: platform learning mechanics from Space Ads; the acquisition cost ceiling from ClicksGeek. All magnitudes are yours to supply.
A hypothetical illustration, not a measured result: an advertiser with a cost ceiling of 100 per customer, currently paying 130, loses 30 per customer every month it stays. If switching costs the equivalent of four months of that gap in fees, time and learning loss, the switch pays back only if the new partner gets under the ceiling within eight months. The anchor is the ceiling, not the retainer. An agency that charges more but keeps CPA under the ceiling is cheaper than a lower retainer that does not.
In-house, new agency, or hybrid: comparing the alternatives
Score in-house, new agency and hybrid on five criteria, then require any shortlisted agency to answer the evaluation questions in writing before it pitches. The criteria are the same ones you used on the incumbent, which keeps the comparison fair.
| Criterion | In-house team | New agency | Hybrid | What evidence to ask for |
|---|---|---|---|---|
| Total acquisition cost | Salaries, tooling, management time and hiring risk | Retainer or percentage plus onboarding and overlap | Both, at smaller scale each | Full cost model against your acquisition cost ceiling |
| Speed to competence | Slow; hiring and ramp | Faster if the agency has channel depth | Moderate | Named team, their channel history, start date |
| Specialist channel access | Narrow unless you hire per channel | Broad, depending on the agency | Agency covers gaps | Case documentation with methodology and limitations |
| Measurement capability | Close to data, may lack modelling skill | Varies widely | In-house owns measurement, agency executes | Written answers on metrics, success definition, how data changes strategy |
| Control of first-party data | Highest | Lowest unless contracted | High if in-house owns systems | Confirmation that client owns accounts and data from day one |
The cost criterion follows ClicksGeek's instruction to set the acquisition cost ceiling from customer economics before evaluating any proposal (ClicksGeek); the measurement questions follow WebFX; the ownership row follows Space Ads; the data point follows Miller Ad Agency.
When in-house wins
In-house wins on ownership, institutional knowledge and proximity to CRM and sales data. It loses on hiring risk, narrow channel skill and the absence of an external benchmark. It suits advertisers with stable channel mix, enough spend to justify salaried specialists, and a measurement function already in place.
When a new agency wins
A new agency wins when the diagnosis showed genuine agency failure, when you need channels your team lacks, or when speed matters more than control. It only wins durably if it accepts client ownership of accounts and data from day one and provides full access and outcome-linked reporting (ClicksGeek). Any hesitation on that point during the pitch predicts the next transition.
The hybrid most mid-size advertisers end up with
The common landing point is in-house strategy and measurement ownership with agency execution. It requires governance: one owner of the success definition, one owner of the tracking configuration, and a written line between what the agency decides and what it recommends. Without that line, the hybrid drifts back into full outsourcing with an internal observer.
How to evaluate a replacement agency on evidence
Ask for documented work with methodology and stated limitations rather than pitch claims. For an example of what documented, evidence-first agency profiles look like, see how GPI profiles The Influence Agency with stated criteria rather than pitch claims. The profiles of The Goat Agency and ACE Agency illustrate the same format; they are examples of documentation, not recommendations, and GPI makes no claims about their results. If you run a formal selection, the advertising agency RFP template builds the same questions into the process.
How to transition accounts and data without losing history
Build an access inventory covering every platform, who holds admin, and who owns billing, and complete the first two steps below before the termination notice is sent. Order matters more than speed here; the losses from doing this backwards are hard to reverse.
The governing principle comes from Space Ads' handover process: the client should be the long-term owner and administrator of domains, analytics, tag management, core ad accounts and first-party data systems wherever platform rules allow (Space Ads). That is one agency's documented practice, not platform policy, so check each platform's current ownership and transfer rules before relying on it.
Step 1: Secure ownership before giving notice
List every asset: ad accounts on each platform, analytics properties, tag manager containers, pixel and conversion API configurations, domains, creative libraries, CRM integrations and any data warehouse the agency populates. For each, record the legal owner, the current administrators and the billing entity. Where the client is not owner or admin, request the change now, while the relationship is still intact. Do this before notice, because cooperation is easier before termination is on the table.
Step 2: Freeze structure, then transfer access
Freeze campaign structure and conversion definitions for the handover window; no restructuring, no new accounts. Space Ads notes that keeping the same account retains history, that significant edits still disturb platform learning, and that a new account loses more continuity (Space Ads). Then add client administrators to every asset, add the incoming agency at the permission level it needs, and only afterwards begin removing the outgoing agency. Never reverse this order: removing the incumbent first can leave assets with no admin at all.
Step 3: Move measurement and data pipelines
Document tags, conversion event definitions, attribution windows, offline import schedules and CRM feedback loops exactly as they stand. The audit from earlier in this article produced most of this. The goal is a new baseline that is comparable to the old one; if conversion definitions change during handover, no later performance difference can be attributed to the agency change (Flightdeck). Any planned measurement improvements should wait until after the baseline is confirmed, or be recorded as a deliberate break.
Step 4: Run the overlap and confirm baselines
During the overlap, the incoming agency observes and documents; the outgoing agency maintains. Record spend, conversions by platform and by CRM, CPA and attribution settings for the final incumbent period. Agree in writing which figures constitute the baseline. Resist pressure from the new team to show quick wins through restructuring in week one.
Step 5: Revoke access and document the state
Remove the outgoing agency from every asset in the inventory, confirm the client remains owner and admin everywhere, and rotate any shared credentials. Save the final state: account structure export, tag container version, conversion definitions, and the baseline figures. When performance moves in either direction three months later, this record is what lets you explain why.
Contract, notice and governance risks to check before you decide
Send the incumbent contract to legal with three questions: what the notice terms are, what reverts to the client on exit, and how data processing is governed during handover. This section is general guidance, not legal advice; terms and obligations vary by jurisdiction and by contract.
Notice periods and overlap exposure
Read for the notice period, whether termination for convenience is permitted or only termination for cause, and which fees continue during notice. A long notice period with a full retainer running alongside a new agency's onboarding fee changes the switching-cost table materially. If the contract allows termination for cause, the remediation plan and its documented results become relevant evidence, which is one more reason to write the plan down.
Ownership clauses for accounts, data and creative
Confirm that ad accounts, first-party data, tag configurations and creative assets revert to the client, and establish what happens to agency-created assets and any proprietary tooling the agency used. The ownership principle described by Space Ads, with the client as long-term owner and administrator of accounts and data systems (Space Ads), is easier to enforce when the contract already says so. Where it does not, negotiate it before notice.
Data protection obligations during handover
Customer data moving from one processor to another needs a documented lawful basis, processor agreements with both agencies, and a plan for deletion by the outgoing party. Treat this as a legal and compliance review, not a marketing judgement.
Use the review to fix governance for the next relationship. ClicksGeek's stance that full account access and reporting tied to business outcomes should be demanded (ClicksGeek) translates directly into contract language: client ownership, client admin rights and outcome-linked reporting written in from the start.
What privacy and AI shifts mean for the partner you keep or choose
Add two questions to any review, whether of an incumbent or a candidate: where does our first-party data live, and which privacy-safe measurement environments has this team actually operated in? Both questions test durability rather than current performance.
First-party data as the asset that survives a switch
As platform signals degrade under privacy changes, the data you own becomes the asset that persists across agencies and platforms. Miller Ad Agency puts it as the owned CRM becoming the most valuable media asset when the algorithm loses visibility (Miller Ad Agency). This is commentary from a digital agency with no published date, so treat it as an argument rather than a measured trend. The argument still strengthens the case for client ownership of data and measurement regardless of who executes the media.
Measurement capability as a selection criterion
LiveRamp, which sells data collaboration technology, describes clean rooms as an essential technology for responsible customer data analysis given fragmentation and privacy pressures (LiveRamp). Vendor interest is obvious, and the date is unknown, so do not read it as an adoption statistic. Read it instead as a capability question for the agency: has the team operated in clean rooms or similar privacy-safe environments, and can it show the work?
An agency that keeps your data and measurement configuration inside its own systems is a durability risk even if this quarter's numbers are acceptable, because the next transition will cost you the asset that was supposed to survive it.
Vendor claims to treat with care
None of the approved evidence for this article quantifies the effect of AI on agency performance. Agencies will make claims about AI-driven efficiency; ask them to demonstrate on your account, with a defined test and a stated limitation, rather than assert. The same evidence standard you applied to the incumbent's dashboards applies to the candidate's technology pitch.
How GPI would decide: change the agency or fix the relationship
Write down your answers to the three proofs today and schedule the decision date. The question of when to change advertising agency reduces to those proofs, and each one has an owner and a test described above.
First, measurement is verified: conversion events, calls, forms and offline revenue reach the platform and reconcile to the CRM, confirmed by someone other than the incumbent (Flightdeck). Second, the fault sits in the agency column of the diagnostic matrix after client-side causes have been tested and cleared. Third, the switching path beats the remediation path within 12 months at your acquisition cost ceiling (ClicksGeek).
If any proof is missing, run the 90-day remediation with the success definition, metrics and exit criteria agreed in advance (WebFX). If all three hold, switch, but only after securing client ownership of every account and dataset and keeping the existing accounts intact rather than rebuilding (Space Ads).
Whichever path you take, leave with better governance than you started with: client-owned accounts, transparent reporting tied to business outcomes, and a written definition of success.
GPI has not run campaigns and does not claim to. Its directory assesses agencies on documented evidence, stated methodology and acknowledged limitations, as set out in the GPI methodology; a profile such as Agency Jet shows the format without implying a recommendation. That is the same standard to hold an incumbent to: an agency that can be evaluated on evidence can be kept or replaced on evidence.
FAQ
How long should we give an agency to fix performance before switching?
This article uses a 90-day remediation window as its own framework; no approved source establishes an industry norm. The window works only with a written plan, agreed metrics and exit criteria set before day one. Judge the result against those criteria, not against how the relationship feels at the end.
Should we tell the incumbent we are running a review?
Complete the access inventory and secure client ownership of accounts and data first, while cooperation is easiest. After that, disclosure is a judgement about contract terms and the relationship. Check notice and termination clauses with legal before any conversation, since what you say may have contractual consequences.
Can we move to a new agency without them touching our existing ad accounts?
Yes, and it is the preferable route. Keeping the same accounts retains history, while significant edits still affect platform learning and a new account loses more continuity (Space Ads). Add the new agency as a user, freeze structure during handover, and treat any rebuild as a deliberate, documented decision.
What if the agency owns the ad accounts and refuses to transfer them?
Start with the contract and legal review. The principle that the client should own and administer accounts and data systems (Space Ads) is good practice, not law, so enforceability depends on your agreement and platform rules. Where transfer is impossible, plan for a new account and budget for the continuity loss.
Is a lower retainer from a new agency a good reason to switch?
Not on its own. The anchor is total acquisition cost against the ceiling you derive from customer economics (ClicksGeek). A cheaper retainer that leaves CPA above the ceiling costs more than an expensive one that keeps it below. Run the 12-month comparison before treating fees as the deciding factor.
How do we evaluate a replacement agency's claims without a track record with us?
Require written answers on which metrics they track, how they define success and how data changes strategy (WebFX). Ask for documented work with methodology and limitations. Treat guaranteed outcomes as a warning sign (Oneupweb), and confirm they accept client ownership of accounts from day one.