Stopping a channel is a change to the plan, not a deeper cut
A budget cut and a channel exit can look alike in a spreadsheet, but they answer different questions. The cross-channel guide to cutting a marketing budget works when the total has to fall: it ranks every line by what its last money earns, removes the top slices where they earn least, keeps slow-to-rebuild work alive at a lower level and agrees in advance when money will come back. An exit asks whether one channel should hold a place in the plan at all, and the total may not change, because the money can move to another line. The Channel MixA set of channels with assigned jobs that together cover the customer journey for the chosen segments, and the share of effort each one carries. entry draws the same distinction: shifting weight among the channels a plan already runs is routine, while adding or dropping one alters its structure.
An established channel is not a trial either. The guide to choosing marketing channels stops a channel whose test failed rather than leaving it on a token budget, and the diagnosis for a new marketing channel that has not delivered tells a trial that never ran as designed apart from one whose economics cannot work, closing that second case by stopping the channel and writing down the reason. This guide deals with a channel that has run long enough to gather dependants: capture channels drawing on the demand it creates, remarketing audiences on other platforms partly filled with its visitors, signed orders and retainers with notice dates, an agency or creator who runs it, customer lists uploaded to it and a tag loading on the site. The decision belongs to whoever owns the performance plan, with finance involved wherever money is committed.
The evidence an exit needs: role, margin and a holdout
Three questions come before the wind-down begins, in this order.
Is the channel doing the job the plan gave it? The channel mix entry gives every channel one job and its own measure of success. With an exit at stake, judge the channel against its role, not against whichever channel shows the best reported return. A channel given demand creation looks poor on last-click sales whatever it achieves, so the question is whether the measure written for that job, such as new customers in the business's records, moved as intended. As general practice, if the job itself has gone, because the product it served was withdrawn or its audience has moved elsewhere, that can be the case for an exit, though the dependency checks below still apply.
What does the whole channel buy at the margin? A cut asks what the last slice of a line earns; an exit asks what the business loses if none of the channel runs, and whether the same money would buy more on another line. Put the answer in whatever measure the business runs on and set it beside what a customer is worth and beside the marginal evidence for the line that would receive the money.
What does a holdout show? Platform reports cannot settle the exit, because they credit the channel with conversions that would have happened without it. An IncrementalityConversions caused by marketing. test can. The incrementality testing guide withholds a channel from comparable people or places and reads the difference in outcomes, which makes the held-out group a small version of the stop itself. Where the channel's effect reaches sales the platform cannot see, or several platforms claim the same customers, a Geo ExperimentA test that runs a campaign in some regions and holds it back in comparable ones, then compares total outcomes, where people cannot be randomised one by one. read from the business's own records gives the platform-neutral answer; the comparison of geo holdouts and conversion lift studies sets out when a platform's own lift study fits instead.
The hierarchy of evidence for marketing claims explains why a holdout outranks the alternatives: attributed figures cannot rule out buyers who would have bought anyway, and a before-and-after readingA measure read before and after a change with no comparison group, so the season and everything else that moved in those weeks is credited to the change. of a pause cannot rule out the season or other changes in the same weeks. The framework of one-way and two-way doors lists leaving a channel among the decisions that call for research and a pilot first, and among its ways of making such a decision smaller are holding some customers back and keeping the old channel running at reduced weight while its replacement proves itself.
This guide treats that reduced weight as a step inside the holdout, with an end date written before it starts, and never as a standing token budget, which keeps it consistent with the choosing guide's rule. As general practice, write the exit's decision rule before the holdout begins: the outcome, the result that keeps the channel, the result that stops it, and the date of the reading. Where no holdout is possible, say so, label the strength of the evidence used, and stage the stop so that its early steps can still be walked back.
What leans on the channel: created demand, remarketing pools and other platforms' numbers
Check each dependency before the stop, in the held-out regions where a holdout has run.
Demand it creates for capture channels. The cross-channel planning guide records that capture channels such as search draw on the demand that creation channels generate, so that taking money from creation shows up later as falling search volume rather than as a loss on the day. The comparison of brand and performance marketing explains why that effect is hard to see: last-click and platform attribution hand the credit to brand search and retargeting, and the creation side's contribution surfaces as other channels' conversions and as cheaper results later on. A creation channel that looks weak in its own report can therefore be holding search up.
Remarketing pools on other platforms. The prospecting and remarketing comparison explains that the RemarketingAdvertising to previous visitors or engagers. pool is fed by prospecting and by any other route that brings people to the site, and that remarketing spend should track how large the pool is rather than what it reports. A channel that sends visitors fills the audiences other platforms build from their own tags, so once it stops, those pools take in fewer people. As general practice, estimate the channel's share of recent visitors from analytics and plan remarketing budgets elsewhere to fall with their pools.
Other platforms' reported numbers. The cross-channel planning guide notes that platforms each claim conversions the others claim too. When one channel stops, the other platforms go on claiming the buyers their own ads reached, as they did while it ran, and one whose remarketing drew on its visitors may report fewer. No platform's report measures what the stop cost the business. As general practice, read the stop in the business's own records and treat a shift in another platform's report as a change in what that platform claims until a test says otherwise.
| What leans on the channel | Why it depends on the channel | Read before the stop | Watch after the stop |
|---|---|---|---|
| Search and Shopping volume | Searches prompted by the demand the channel creates | Branded and category search in held-out regions against live ones | Branded search and capture costs, on a lag |
| Remarketing on other platforms | Pools filled partly by the channel's visitors | The channel's share of recent site visitors | Pool sizes, frequency and remarketing results |
| Other platforms' reports | The same buyers claimed by more than one platform | Platform claims against the business's own counts | Reported conversions against the business's records |
| Exclusions and seeds on the channel | Customer lists uploaded to it | The list register's rows for the channel | Nothing, once removed and recorded |
| The job the plan gave it | No other line may do that job | Whether another line takes the job, or the job ends | The job's own measure, under its new owner |
Winding down: commitments, agencies, campaigns and the order of switch-off
Once the decision is made, the channel comes out in sequence: what is promised to it, then the people who run it, then the campaigns.
Commitments. List everything signed for the channel, such as insertion orders, private deals, a creator or agency retainer or a tool used for this channel alone, with each one's value and next break point or notice date. The cutting guide's table of fixed costs, variable media and commitments shows that a commitment comes free at its next break point or notice date and carries a penalty if ended sooner, and its sequence of cuts gives notice where a contract allows rather than breaking it; an exit follows the same rule. The guide to tracking spend against budget supplies the arithmetic: the headroom free to move is the budget less what is committed, not less what has been invoiced, so money tied up until a notice date is not yet available to another channel.
Agencies and creators. Where an agency, freelancer or creator runs only this channel, the relationship ends with it, and the agency offboarding checklist covers the weeks from notice to the final day of access. Where the agency keeps other work, record the change of scope in writing and remove its access to the stopped channel only.
Campaigns. As general practice, the switch-off runs in steps that can each be dated and read:
- Start nothing new. From the decision date, no new campaigns, bookings or creative briefs on the channel.
- Move or end each campaign's job. A job that moves, such as excluding recent buyers, is already running in its new place before the old campaign pauses.
- End the reduced-weight step on its written date, not when someone remembers it.
- Pause in stages, across the days or regions the plan sets, noting each step's date. The cutting guide's warning against making every change on the same day holds here too, because a staggered stop lets each step be read.
- Pause, never delete. Campaigns, ad sets and ads stay in the account, paused; uploaded lists and the tag are separate decisions, set out in the next section.
The guide to pausing or closing a market covers the parallel case for one market's whole presence, with its own switch-off order in 'Campaigns, budgets and ad accounts' and its own list in 'What to keep for a return'. This guide takes one channel out of every market. One point carries straight across: pausing campaigns costs nothing, while deleting an account loses its history.
If the business also posts organically where it advertised, leaving that presence is a separate decision, which the guide to choosing social platforms sets out under 'When to leave a platform'. Ending the paid activity does not by itself mean leaving the platform.
What to keep for a return, and what to remove
A stopped channel may come back. Keep what would make a return quicker and better informed, and remove what has no purpose once the campaigns stop.
Keep the account, in the business's name. The new ad account setup checklist sets the ownership a dormant account needs as much as a new one. Review access before the account goes quiet and record who still holds it, as the market closure guide does for a market's dormant accounts. As general practice, also file the holdout's design and result, the final settings and structure, the conversion setup and the creative that earned its place, so a return can be judged against the same baseline.
Decide on uploaded customer lists with no remaining purpose. A list uploaded for exclusions or seeds has no job once the campaigns stop. The Personal DataAny information relating to an identified or identifiable living person, such as a name, email address, location data or an online identifier. in it stays personal data when hashed, and the guide to personal data in analytics describes the GDPR's data minimisation principle as collecting only what the purpose needs, with storage limitation adding that data which identifies people is kept no longer than the purpose needs. As general practice, and not as legal advice, decide with whoever owns privacy whether each list is removed, record the decision, and take its rows out of the register kept by the guide to running customer lists across ad platforms. How each platform deletes an uploaded list is outside this guide.
Remove a tag with no remaining purpose, deliberately. The channel's pixel goes on loading after the campaigns stop, and can go on sending data to a platform the business no longer buys from. The tutorial on retiring a marketing tool has the tag and access steps that fit; follow them for the channel's pixel. Without the tag, no new site visitors join the platform's audiences ahead of a return, and that trade belongs in the record.
Plan a return as a relaunch. The tutorial on bringing paid media back after a budget cut reports Meta's documentation that an ad set brought back after an extended pause counts as a significant edit, one able to restart learning. As general practice, treat any return as a launch with its own learning period, a fresh test against the filed baseline and newly built customer lists, rather than as switching the old campaigns back on; for a platform other than Meta, read that platform's own section of the site first.
Moving the money and reading what the stop changed
Move the money on marginal evidence. Released money is not owed to whichever channel reports the best return. Record the move in the budget reallocation memo, including the marginal evidence for the line that receives the money and whether it is tested or platform-reported. As general practice, move it in steps and read each one. Money the receiving lines cannot spend at the target can be returned to finance.
Log the stop as a measurement change. A report that includes the channel can show a step on the day it stops. The measurement change log lists campaign pauses big enough to shift totals among the changes it records, each with a date, an expected effect and, later, the observed one. Enter the stop, each switch-off step and the removal of the tag, and mark the break in trend on the dashboards that carry the channel.
Read the business's own records, on a lag. Start with the business's own count of orders, qualified leads and new customers, then read the dependants on their own timescales. The cutting guide warns that a smaller brand flight may surface only later, in Branded SearchSearch queries that contain a company's or product's brand name. and in what capture channels cost, or not at all, so read branded search against the period before the stop for as long as the business's own lags suggest, and set other platforms' reported conversions beside the business's own counts.
Know what the reading can and cannot show. After the stop, the comparison is a before-and-after reading, which the hierarchy of evidence places below a holdout because the season and other changes are mixed into it, so the holdout read before the exit stays the stronger evidence. The incrementality testing guide adds a caution: a test measures a channel's contribution given everything else that was running, so removing one channel can change what the others add. As general practice, retest the line that received the money once it has settled.
Record the decision and what would reopen it. Write the exit into the Decision LogA running record, one row per decision, of what changed and why, what was expected, who decided and when the result will be checked. with its evidence and that evidence's strength, its classification as a one-way door, the date of each step and a check date. The marketing decision log template has fields for the evidence, the reversibility and the check date, and its guidance on reversibility notes that leaving a channel needs research or a test first; the hierarchy of evidence asks that each result carry a label of how strong the evidence behind it is. Add the conditions that would reopen the channel, such as the audience moving onto the platform or a product that suits its role, so that the question comes back on evidence rather than on enthusiasm.
A ukulele shop stops its paid video channel (illustrative)
Talmordique Ukuleles takes TikTok out of its plan
Talmordique Ukuleles, an invented online seller of ukuleles and ukulele accessories, spends £21,000 a month on paid media: £7,000 on search, £4,000 on Shopping, £6,000 on Meta and £4,000 on TikTok, whose job was demand creation among people taking up an instrument. The shop can pay £36 to win a new customer. After two quarters TikTok's report claims 150 orders a month, but new customers in the shop's records, about 800 a month, have not moved. A regional holdout switches TikTok off in 6 of 18 matched regions for 8 weeks, with a rule written beforehand: stop if the incremental cost per new customer is above £36.
The held-out regions record 2.5% fewer new customers than the live ones, with a range of 1% to 4%: between 8 and 32 new customers a month nationally, or £125 to £500 each on the media alone, more with the retainer. Branded search in the held-out regions is flat against the live ones. TikTok brings 11% of recent site visitors, so the Google and Meta remarketing pools will shrink by about that share, and a £900-a-month creator retainer, paid on top of the £4,000 of media, has a notice date at the end of the following month.
The wind-down takes a month. Notice goes to the creator on the decision day, and the last £900 counts as committed. As the last step written into the holdout's design, with its end date set before it began, the held-out regions stay off while the live regions run at £2,000 for four weeks and Meta prospecting takes the first £2,000; then every TikTok campaign is paused, and remarketing on Google and Meta, £1,350 a month between them, is planned down by £150 a month, about 11%. The ad account stays paused in the business's name with two staff administrators and the creator's access removed. The uploaded list of 9,400 customers is removed once the owner and the person responsible for privacy agree it has no remaining purpose, and the pixel comes out through the tag manager with a version note. Each step goes into the measurement change log with its date.
The reallocation memo moves the £4,000 into Meta prospecting as two £2,000 steps, where a spring step test had put the marginal cost at £30 per new customer. The first adds 64 new customers at about £31 each; the second adds 50 at £40, above what the shop can pay, so it is withdrawn and £2,000 is handed back to finance at the next reforecast. Three months on, branded search sits 3% below the quarter before, inside its normal range, and new customers are 5% higher. The decision log records the reopen conditions: a fresh holdout showing an incremental cost under £36, or a beginner's kit aimed at younger players.
Common mistakes
- Keeping a token budget to stay present. It costs money and settles nothing; stop the channel, or hold it at reduced weight only as a dated step of the holdout.
- Deciding on the platform's own report. It credits the channel with buyers who would have bought anyway; the exit rests on a holdout in the business's records wherever one is possible.
- Forgetting the other platforms' pools. Remarketing budgets elsewhere stay sized for visitors who no longer arrive.
- Breaking commitments or missing notice dates. List every signed order and retainer with its break point before the first campaign pauses.
- Deleting the account. Pausing costs nothing, and a deletion takes the history a return would need.
- Leaving lists and tags behind. An uploaded customer list goes on holding data, and a live pixel can go on sending it, with no purpose; decide on each with whoever owns privacy and record the decision.
- No record of why. Without the evidence and the reopen conditions in the log, the channel comes back on enthusiasm or never comes back at all.
The channel playbooks hub sets this exit beside the channel roles, the cross-channel plan, the prospecting-remarketing split and customer lists across platforms.
