How to test marketing channels one at a time and know in 90 days if the first one is worth keeping.

Annual plans tend to start several channels in the same month, and a quarter later it's hard to tell which one worked. As you plan next year's tests, the way to test marketing channels without that fog is to run them one at a time, each with its own number, its own owner, a fixed budget and a date when you decide.

That approach paid off when I built an email program from nothing at a company that sells workforce training. The database held years of event attendees and inquiries, and emailing all of it would have been the fast start. I started with one group instead, operations managers who had attended a past event, because I could describe them best and they had a recent reason to hear from the company.

Every email could speak to the one problem they had raised at that event, and results from one group were easy to read within weeks. What worked for them shaped the emails for the next group. Group by group, the program grew to more than 25 qualified leads a month, and each new group took less time to get right.

Why one channel at a time beats four at once.

Claude Hopkins described the idea in Scientific Advertising in 1923.

"Before we spread out, we prove our undertaking absolutely safe." Claude C. Hopkins, Scientific Advertising, chapter 15, "Test Campaigns"

Hopkins tried a new project "in four or five towns" before spending nationally, and he judged every test on money: "our final conclusions are always based on cost per customer or cost per dollar of sale." For one food advertiser his agency tried more than fifty plans over five years, and the best cut the cost of selling by 75 percent.

Marketing budgets have held near 7.7% of company revenue for three years in Gartner's CMO Spend Survey (the data is charted in setting next year's budget), and smaller companies will see different percentages. The point holds at any size: a new channel usually gets paid for by moving money, and moved money has to beat what it replaced.

The prospect's side points the same way. In a Gartner survey of 632 B2B decision makers (August and September 2024), 61% said they prefer a buying experience without a sales rep, and 73% said they actively avoid suppliers who send irrelevant outreach. Prospects research out of sight and tune out anything off their problem, which favors one group with one problem.

Measurement adds a reason. In 2023, SparkToro and Really Good Data sent more than 1,000 test visits from 11 networks through about 100 participants and checked how Google Analytics labeled them. Every visit from Slack and WhatsApp came in as "direct", as did 14% of visits from public LinkedIn posts.

When four channels start in the same month, the one people pass around in private messages often shows up as "direct", and the one with the cleanest tracking tends to collect the credit.

How to test marketing channels in five steps.

1. Pick the channel from 10 client conversations.

Ask 10 of your best-fit clients where they looked before they found you, and pick the place that comes up most that you don't use yet.

Rand Fishkin of SparkToro has named three problems with "how did you hear about us" (2023). People remember the last step and forget the colleague who first named you, the question doesn't say which moment it means, and it only reaches people who found you. A conversation about the whole search reduces the first two, and adding 3 lost prospects widens the view. People who never heard of you stay out of reach, which makes the tally a list of candidates, not proof.

At a larger company, pull the last 20 closed deals and 10 lost ones from the CRM, have account managers ask on regular calls, and search recorded sales calls for "heard about" and "found you".

2. Choose one number that decides it, and a date.

Pick the single number that decides if the channel stays, like qualified leads per month or cost per meeting, and a date to judge it, usually 90 days out. Write both down before you start, while you have no result to defend.

Then check the volume. A count of outcomes swings by roughly 2 times its square root, which is about plus or minus 40% at 25 outcomes and 63% at 10. At 25 you get a direction, and the fix band in step 5 absorbs the swing. If the channel can't produce 25 of your decision number in 90 days, judge an earlier number, like meetings booked, and give the revenue question more time.

Slow channels like SEO, events or a podcast need this most. Les Binet and Peter Field's IPA report The Long and the Short of It (2012) drew on about 1,000 campaigns from 30 years of IPA effectiveness data, mostly consumer brands, as their 2019 B2B report notes, which makes it a warning for B2B and not a benchmark. The IPA's summary says that "very short-term online metrics" used as primary measures carry "dangerous implications for long-term success." Judge the first 90 days by an early number and set a later date for the full decision.

At a larger company, use pipeline created or sales-accepted opportunities as the early number, and compare against the same quarter last year to take seasonality out.

3. Give it a fixed budget and one owner.

Multiply the cost you'll accept per outcome by 25, and if that's more than you can move, judge a cheaper outcome or run longer.

One person owns the test, by name, writes one line a week with spend and outcomes, and makes the call on the decision date.

Before the first dollar goes out, give the channel a campaign name with the year, like "2027-q1-association-webinars", in your UTM tags and in the CRM (a campaign in HubSpot or Salesforce, a tag in Mailchimp, a list in Constant Contact or a spreadsheet column). If you're unsure your tracking catches every lead, fix conversion tracking first.

At a larger company, give the test its own budget line that finance can see, and fix any agency's scope for the full 90 days.

4. Keep other changes small during the test.

Hold steady what you can: the website, the offer, pricing, the sales follow-up and the other channels' budgets. Keep a dated change log of anything you couldn't hold. A redesign in week six turns a channel test into a guess. Weigh a paid platform's early results lightly, because Google says Smart Bidding usually settles within 1 to 2 conversion cycles, the time from click to conversion.

Watch the total as well as the channel. In large field experiments on eBay's paid search ads, returns were a fraction of non-experimental estimates, and brand-keyword ads had no measurable short-term benefit (Blake, Nosko and Tadelis, Econometrica, 2015). That's consumer ecommerce, and the same paper found non-brand ads did move new and infrequent shoppers. A new channel can take the same credit for prospects who were coming anyway, which is why the check is total qualified leads against the same quarter last year, read with the change log for growth and price changes.

At a larger company, the stronger version is a holdout. Run the channel in some regions or account lists and not others, then compare, the way Hopkins compared towns.

5. Keep it, fix it or drop it, then pick the next one.

Write the rule before you start, and keep the channel if it meets the target. Fix it if it comes in within about twice the target and the owner can name one cause, then change only that and run it 90 more days at the same weekly spend alongside the next channel's test, with the change noted in that test's change log. Drop it if it's worse than twice the target or couldn't spend its budget, and use the same rule at day 90 of a fix. Then pick the next channel from your tally.

Money from the channels you drop then funds the ones you keep, which makes next year's budget easier to set.

A worked example with numbers.

Example numbers: a company gets most qualified leads from referrals and search ads, at about $250 per qualified lead from paid channels. The owner interviews 10 clients and 3 lost prospects, and 6 of the 13 name webinars run by their industry association, which the company has never used.

• Decision number: qualified leads from association webinars.
• Target: $300 or less per qualified lead.
• Budget: $7,500 for 90 days, moved from the lowest-performing ad campaign, which costs about $350 per qualified lead. At $300, that's the 25 qualified leads the volume check asks for.
• Rule: keep at 25 or more qualified leads, fix at 13 to 24 ($313 to $577 each), drop at 12 or fewer.

After 90 days the channel brings 18 qualified leads, about $417 per qualified lead, which means fix. Most leads came from one session topic, and the next 90 days test only that topic.

Total qualified leads for the quarter went from 84 in the same quarter last year to 81. The moved ad money used to bring about 21, and 84 minus 21 plus the channel's 18 is 81, which suggests most of the channel's leads were new. A swing of about 18 is normal for about 80 leads, and the owner treats the match as support, not proof.

The channel test plan, ready to copy.

Common mistakes and the pushback a specialist would raise.

The usual mistakes are starting two or three channels in the same month, judging by impressions, opens or clicks, and ending early after one great or terrible week.

"Channels work together, and testing one alone misses that." They do, which is why step 4 checks the total. After counting the moved money's old leads, a channel that lifts the total by more than its own count and the normal swing was helping others, and one that reports more than the rise was borrowing.

"One a quarter is too slow." It's four answers a year, and tests inside a channel, like an email A/B test or a Google Ads experiment, can run alongside because each splits one audience at random. A new channel has no such group, which is why each needs its own quarter.

A test you can run in one week.

Run the ten-client interview from step 1 before you commit next year's money. It takes about five working days of 15-minute calls. Measure how many of the 13 people named each place. A result is any place you don't use that 3 or more people named, and that becomes your next channel test. If nothing reaches 3, test a new group inside a channel you already run, which is the list-first rule at work. What a lead is worth gives you the target for step 2.