How to test a new marketing channel and get a real answer
Founders keep running channel tests that end in a shrug: some engagement, no verdict, and the spend quietly continues. A test only returns a yes or no if you make four setup decisions before you spend the first dollar.
By Stacey Tallitsch | July 21, 2026
A peer, a podcast guest, or a rep from a platform you have never used tells you the growth is somewhere you are not. LinkedIn. YouTube. A new ad network. A booth at the regional trade show. You carve out a few thousand dollars to try it. You have done this before. You spun up the channel, watched it for a couple of months, and got nothing you could actually rule on. Some clicks. A handful of likes. One lead that might have come from there, or might not have. The test never ended. It faded into the general spend and renewed itself every month, because nobody could prove it worked and nobody could prove it did not. That is not a test. That is a subscription you forgot to cancel.
A channel test is not a marketing activity. It is an experiment, and an experiment that cannot return a clear yes or no is not an experiment. It is a purchase with a story attached. The reason most channel tests end in a shrug has almost nothing to do with the channel and almost everything to do with how the test was built. The founder picked a round number, turned it on, and waited to see how it went. There was no threshold that would end it, no way to read the result, and no fixed date to decide. So the outcome was always going to be ambiguous, and ambiguity always resolves in favor of continuing to spend.
This is not a small-company problem. It is everyone's problem. By The CMO Survey's long-running tracking out of Duke's Fuqua School of Business, marketing analytics inform well under half of the decisions marketers actually make, and focusing data and analytics on the most important marketing problems ranked as a top challenge for marketing leaders in the Spring 2025 wave. Companies with entire analytics departments still run on gut. If they cannot read their own data, the solo founder eyeballing a new channel has no chance unless the test is engineered to produce a signal a human can actually see.
So engineer it. Four decisions, all made before the money goes out.
Decide the answer before you spend
Write the kill number first
Before you fund anything, write down the single result that will make you stop. Not a vague hope. A number, on paper, with a date attached. If this channel has not produced 3 booked jobs by September 15, I kill it. If cost per qualified lead is above $400 after 60 days, it is dead. The kill number is the most important part of the test and the part founders skip most often, because writing it down forces you to admit what a failure would look like. A test with no defined failure can only succeed, which means it can never end.
Set the kill number off your own economics, not a benchmark from a blog. You know what a customer is worth and what you can pay to acquire one. If a new channel has to come in under $500 per acquired customer to beat what you already run, then $500 is the line. A channel that lands at $520 is not close. It lost. The whole reason you write the number in advance is that you will be tempted to move it once real money and a little hope are on the table, and you cannot referee your own game after kickoff.
Size the test to reach signal
The most common way a test dies is starvation. A founder puts $500 against a channel, gets two conversions, and cannot tell whether the channel is good, bad, or ordinary, because two conversions is noise. You need enough events to separate the pattern from the luck. For most paid channels that means budgeting to reach at least 30 to 50 conversions before you read anything, which, at a realistic cost per action, is usually several thousand dollars and several weeks, not five hundred dollars and a fortnight.
Work it backwards. Estimate the cost per conversion, multiply by the number of conversions you need to trust the result, and that product is your minimum test budget. If that number is more than you are willing to spend, you have learned something valuable before spending a dime: you cannot afford to test this channel properly right now. A half-funded test is worse than no test, because it costs real money and hands back a false verdict. Underfunding a test does not shrink the risk. It just guarantees the result will be unreadable.
Wire the attribution before launch
If you cannot tell which channel produced a lead, no budget or duration will save the test. Attribution has to be built before launch, not reconstructed after. For a home-services or trades business that can mean a dedicated phone number used on that channel and nowhere else, so every ring is unambiguous. For a business taking web inquiries it can mean a landing page that only this channel points to, plus a how-did-you-hear-about-us field that a human actually reads. The mechanism does not have to be sophisticated. It has to be exclusive: one path that only this channel can travel, so a result there means what you think it means.
The discipline that breaks most tests is changing two things at once. You launch the new channel the same week you redo the website or run a promotion, and now every number is contaminated. If you want a channel test to be readable, it has to run against a stable background. Change one variable. This is the same reason website traffic can climb for a year while inquiries stay flat is so hard to diagnose after the fact. When several things move together, the data cannot tell you which one moved the outcome.
Set the clock and leave it alone
Pick the end date when you fund the test, and pick it long enough to clear the noise. Most channels need 60 to 90 days before the numbers mean anything: paid channels toward the shorter end once they hit conversion volume, organic and relationship-driven channels toward the longer. Write the date next to the kill number. Then, and this is the hard part, leave the test alone until that date arrives. Do not cut it in week 2 because the first weekend was quiet. Do not optimize it every three days into something you can no longer read. And do not extend it a month past the deadline because you are not ready to hear the answer. The clock is there to protect you from your own reactions in both directions.
When testing is the wrong move
Two founders should not run this play at all, and for them the prescription is to walk away, not to run the test better.
The first is the founder whose current channel is not tapped out. If referrals or your one working ad channel still have room to grow, testing a new channel is not diversification. It is procrastination wearing a lab coat. New channels are attractive precisely because they are unproven, which means they cannot yet disappoint you, while scaling the thing that already works forces you to confront its ceiling. Exhaust the headroom in what works before you go shopping. The founders watching their cost to win each customer keep climbing every quarter often have the opposite problem, a saturated channel they genuinely should be diversifying away from, and for them a disciplined test is exactly right. Know which one you are before you spend.
The second is the business that structurally cannot reach signal. If you close 20 deals a year at six figures each, no channel will ever hand you 30 conversions in a quarter to average against. The math that makes a channel test readable does not exist at your volume, and forcing it produces false confidence off a sample of three. That does not mean you fly blind. It means you evaluate the channel qualitatively and by pipeline: did it put you in front of the right buyer, did a real opportunity move, would that room have happened without it. Run that review with the same discipline you would bring to a win-loss review on a deal you expected to win. The instrument changes. The rigor does not.
What to do today
Before you spend a dollar on whatever channel someone talked you into this week, do one thing this afternoon. Write the single number that would make you kill it, and the date you will check. Put both somewhere you will actually look. If you can write that number, you have a test. If you cannot, if every result you can imagine would keep the channel alive, then you do not have a test at all. You have a hope with a budget line, and a hope renews every month until you cancel it.
— Stacey Tallitsch, Stronghold CMO
About the Author
Stacey Tallitsch is the President of Stronghold CMO, a Fractional AI CMO service operating under Talisman Capital, Inc. He is a 30-year tech veteran and the author of 21 books on systems thinking, operator-grade decision-making, and personal sovereignty, with more than 30,000 students across his Udemy course catalog.