How Do You Decide Which Marketing Channel to Cut?

How Do You Decide Which Marketing Channel to Cut?

Cut the channel you cannot explain, not the one with the worst last click numbers. Attribution flatters channels that capture demand and punishes the ones that create it. A disciplined cut starts by separating what a channel is for from what your dashboard happens to measure.

This decision usually arrives badly. Budget tightens, someone asks for a list ranked by cost per lead, and the bottom two get removed in a meeting. Six months later pipeline is down and nobody can prove why.

Here is the framework we use to make the call properly, and the questions that stop a bad cut.

Why Is the Obvious Method Wrong?

Because ranking channels by attributed conversions measures proximity to the conversion, not contribution to it. The channel sitting closest to the form will always look best, and the channel that made someone search for you in the first place will always look worst.

That is not a modelling flaw you can configure away. It is what last touch attribution does by design. Multi touch models redistribute the credit but still only see the touches they can observe, which excludes podcasts, communities, dark social, word of mouth and every conversation that happens where you have no pixel.

So the first rule is that the report is evidence, not a verdict. We go into the mechanics of this in our piece on B2B SaaS attribution.

What Should You Ask Before Cutting Anything?

Four questions, in this order. What job is this channel doing. How would we know if it stopped. What would it cost to restart it. And what breaks somewhere else if it goes.

The second question is the sharpest. If you cannot describe the signal that would tell you the channel died, you were never measuring it, and cutting it is a coin flip rather than a decision. If you can, you now have your test.

The third question catches the expensive mistakes. Paid search can be switched off on Monday and back on in an hour. An SEO content programme that took two years to build authority cannot be restarted at the same level, and a community you abandon does not reassemble.

What Kinds of Jobs Do Channels Actually Do?

Three, roughly, and you should be able to name which one each channel is doing for you. Some create demand, making people aware they have a problem worth solving. Some capture demand, catching people already looking. Some convert demand that already exists into a meeting.

Cutting a capture channel is usually safe and reversible, because the demand still exists and something else will catch it. Cutting a creation channel is slow poison, because the effect lands a quarter or two later, at which point your capture channels quietly get more expensive and nobody connects the two.

That lag is the single most common way a rational looking cut damages a business. It is also the reason this decision benefits from the framing in brand versus performance marketing.

Which External Changes Should Trigger a Review?

Changes in how people find things. The clearest current example is AI summaries in search. Pew Research Center tracked 900 United States adults across 68,879 unique Google searches in March 2025, publishing in July 2025, and found people clicked a traditional result in 8% of visits when an AI summary appeared, against 15% when none did. Clicks on links inside those summaries ran at 1% of all visits.

A shift of that size changes the economics of a content channel without anything about your content changing. That is a legitimate reason to review, and it is a poor reason to cut on its own, because appearing in the answer still puts your name in front of the buyer.

The right response to an external shift is to re examine what the channel is for. If a blog existed purely to generate sessions, that assumption needs revisiting. If it existed to be the thing buyers and answer engines cite when they evaluate you, it is doing better than the traffic chart suggests.

How Long Should a Channel Get Before You Judge It?

As long as its sales cycle plus its ramp, and you should write that number down when you start the channel rather than when you are deciding to kill it.

Paid search can be judged in weeks. Outbound in a quarter. SEO and content in a year, sometimes more. Community and events in longer still. Judging a twelve month channel on a three month report is not discipline, it is just measurement of the wrong thing.

If you inherited a channel with no stated horizon, set one now and commit to it publicly. A channel on probation with a defined end date and a defined test produces a real answer. A channel under permanent quiet suspicion just decays.

What Does Expansion Revenue Have to Do With This?

More than most channel reviews account for. If a large share of your growth comes from existing customers, your acquisition channels are carrying less of the load than the dashboard implies, and the marketing that supports retention and expansion deserves protection.

ICONIQ's 2025 State of Software reporting describes net dollar retention settling into a healthy range of roughly 110 to 120%. Public results show the same pattern at scale: for the quarter ended 30 June 2026, Datadog reported revenue of 1.12 billion dollars, up 36% year over year, alongside about 4,720 customers with annual recurring revenue of 100,000 dollars or more, up 23% from about 3,850 a year earlier.

Revenue growing faster than the count of large customers is expansion inside accounts. If that describes your business too, then customer marketing, documentation, onboarding content and community are revenue channels, and cutting them to protect acquisition spend is the wrong trade.

What Is the Safest Way to Test a Cut?

Pause rather than delete, in a way you can reverse. Turn a channel down to a fraction of its spend for a defined period instead of switching it off, and hold everything else steady while you do it.

Watch the leading indicators rather than closed revenue, because closed revenue arrives too late to tell you anything useful. Branded search volume, direct traffic, demo requests that cite no source, and the share of new opportunities that already know who you are will all move before your revenue does.

Geographic or segment holdouts work well if you are big enough. Keep the channel running in one region and pause it in a comparable one. It is the closest thing to a controlled experiment most marketing teams can actually run.

Which Channels Are Usually Safe to Cut?

The ones that are running because nobody stopped them. Sponsorships renewed out of habit. A second rate event you attend because you attended last year. A paid channel bidding on terms your organic results already own. Content formats produced on a schedule that nobody has read the performance of in a year.

Tooling counts too. Marketing stacks accumulate, and a platform bought for one campaign often survives three budget cycles past its usefulness. Cutting a tool is easier politically than cutting a channel and often frees comparable money.

The common thread is that nobody can name the job. If the answer to "what is this for" is a history rather than a purpose, it is a candidate.

What About the People?

Be explicit, because this is where channel decisions get emotional and therefore get avoided. Someone built that channel, and cutting it can read as a judgement on them even when it is a judgement on market conditions.

Say what the decision is and what it is not. Name the horizon, the test and the conditions under which the channel comes back. A person told "we are pausing this for two quarters and here is what would bring it back" can redeploy their energy. A person whose channel quietly loses its budget starts updating their CV.

It also protects the decision. A documented pause with criteria is reversible on evidence. An undocumented cut becomes permanent by default, which is how companies lose channels they meant to revisit.

What Would We Do With a Week and a Spreadsheet?

List every channel with four columns: the job it does, the horizon it needs, the signal that would prove it works, and what restarting it would cost. Most teams cannot fill this in, and the blanks are the finding.

Then cut the rows where nobody can name a job, pause one creation channel as a real test with a holdout, and leave everything else alone for a quarter. That is usually a bigger saving than the meeting expected, and it is reversible.

If you are weighing up a channel cut and want a second opinion before it becomes irreversible, we are happy to work through the list with you. You can find us at phoenix.studio.