A framework with sound logic and an unsound test
The Bullseye framework, from Gabriel Weinberg and Justin Mares's Traction, addresses a real problem: businesses default to the channels their team already knows, then conclude that acquisition is hard.
Its structure is good. Its testing protocol is not.
The book's guidance is to test three shortlisted channels with small experiments — figures in the region of $200 to $500 per channel, over roughly a month, under $1,000 in total — and then pick the winner.
That budget cannot distinguish a working channel from a failing one. A test with a few dozen conversions has confidence intervals wide enough to contain almost any conclusion, which means the result reflects variance rather than channel quality. Businesses following the protocol exactly will pick a channel, focus on it, and have no idea whether they picked the right one.
The fix is not to abandon the framework. It is to keep the process and replace the test.
What the framework gets right
The 50% rule. Spend half your time on product and half on distribution, from the beginning. Weinberg and Mares are blunt that poor distribution rather than poor product is the more common cause of failure, and the discipline of treating distribution as a parallel workstream rather than a later phase is correct.
The full-channel brainstorm. Nineteen channels, and the requirement to generate an idea for each — including the ones that feel irrelevant. This exists to interrupt defaulting, and it works.
The contrarian principle. Founders neglect channels that are uncomfortable, such as public speaking, or unfamiliar, such as trade shows. That discomfort is a reason to consider them, because competitors are avoiding them for the same reason. This is the single most valuable idea in the book.
Focus after testing. Once a channel shows promise, commit to it rather than spreading across several. Diluted effort produces results too small to learn from anywhere.
What a real channel test requires
The pilot design principles apply directly here.
Set a minimum detectable effect from your economics. The question is not whether the channel produces customers. It is whether it produces them below your target acquisition cost. That target is the threshold the test must be able to detect.
Size the test from that threshold. Detecting whether cost per acquisition is above or below a target requires enough conversions for the estimate to be stable. Dozens will not do it. The requirement is usually hundreds, which means the budget is a multiple of the book's figure.
Run for at least one full purchase cycle. Channels have different lags. A channel that produces conversions three weeks after first exposure will look dead in a two-week test.
Do not compare channels on platform-reported numbers. Each platform reports favourably on itself, and the reported figures are not comparable across platforms because the attribution rules differ.
Accept that some channels cannot be tested cheaply. Brand-building channels with long lags do not produce readable results in a small test, and concluding they failed is a measurement artefact.
Qualify before you test
The practical answer to expensive tests is to test fewer things, which means qualifying harder before spending.
Four questions, answered from research rather than from experiment.
Are our buyers there in volume? Not whether the channel is large, but whether your specific category buyers use it.
Does the channel match the buying situation? Search captures existing intent. Social creates awareness among people not looking. A channel mismatched to where the buyer is in the cycle underperforms regardless of execution.
Can we operate it competently? A channel requiring skills the team does not have is testing execution, not the channel.
What do the economics need to be? Estimate the cost per acquisition the channel would need to deliver before spending anything. Some channels can be ruled out on arithmetic alone.
Qualifying properly usually reduces a shortlist of three to one or two, which is what makes an adequately powered test affordable.
Channels are depreciating assets
Andrew Chen's law of shitty click-throughs: over time, every marketing channel produces worse response rates.
The mechanism is straightforward. A new channel is effective because few advertisers are using it and audiences have not developed resistance. As more pile in, noise rises, attention falls, and auction prices climb. Email in the 1990s, display in the 2000s, social in the 2010s — each was extraordinarily efficient and then was not.
Two consequences.
Channel selection is a repeating process, not a one-time decision. The channel that works now will decay. Weinberg and Mares are explicit that the framework should be re-run periodically.
Early entry is worth more than optimisation. A mediocre operator in an uncrowded channel usually outperforms an excellent one in a saturated channel. That is an argument for the contrarian principle, and it is why the willingness to test unfamiliar channels compounds.
The evidence
Documented practitioner method. Traction is built on interviews with roughly 40 founders, mostly of venture-backed technology companies, plus the authors' own experience. There is no controlled research comparing businesses that used the framework against those that did not.
Two honest limits.
Survivorship. The founders interviewed had achieved traction. The channel stories are drawn from cases where the channel worked, which tells you little about the base rate.
Context. The book assumes resources and a market structure typical of venture-backed software. The nineteen channels are not equally available to a consumer brand, a services business, or a company selling into a small market.
The framework's value is in its process discipline — enumerate, qualify, test, focus, repeat — rather than in its specific channel guidance.
Where channel selection goes wrong
Testing everything at once. Effort spreads, each test is underpowered, and nothing produces a readable result.
Concluding from an underpowered test. The most consequential error, because the conclusion feels earned. A channel abandoned on thin data may have been the right one.
Optimising a channel that is decaying. Improving efficiency in a channel whose costs are rising structurally is running to stand still.
Never revisiting. A channel chosen three years ago and never re-evaluated is a legacy decision, not a current one.
Treating channel performance as fixed. Channel-product fit changes as the product, the price, and the audience change. A channel that failed at an earlier stage can work later.
Diagnostic: is channel selection deliberate?
Six tests.
All available channels were enumerated, including ones nobody wanted to consider.
The shortlist was qualified on buyer presence and economics before any spend.
Each test was sized to detect a difference against a stated acquisition cost target.
Tests ran for at least one full purchase cycle.
Results were assessed on measured incrementality rather than platform-reported figures.
There is a date for re-running the process.
Test three is where most channel programmes fail, and the failure is invisible because the test still produces a number.
What this produces
A channel decision that can be defended and revisited.
That is the honest output. Not certainty — channel performance shifts with competition, price, and audience — but a decision traceable to evidence rather than to what the team already knew how to do.
The larger value is the discipline of enumeration. Most businesses never seriously consider more than three or four ways of reaching customers, and the channels they skip are frequently the ones their competitors are also skipping. That is where the unusual returns sit, and it costs nothing but the willingness to list them.
Frequently asked questions
What is the Bullseye framework?
A five-step channel selection process from Traction: brainstorm all nineteen traction channels, rank them into inner circle, potential, and long shots, prioritise three, test them cheaply, then focus on the winner.
How much should a channel test cost?
More than the book suggests. Its guidance of a few hundred dollars per channel over about a month cannot produce a result that distinguishes a working channel from a failing one. Size the test from the acquisition cost target you need to detect, which usually requires hundreds of conversions.
How long should a channel test run?
At least one full purchase cycle. Channels differ in lag, and a test shorter than the cycle will report failure for channels that simply convert later.
Why do marketing channels stop working?
Andrew Chen's law of shitty click-throughs: as more advertisers enter a channel, noise rises, audience resistance grows, and auction prices climb. Every channel that has been extraordinarily efficient has eventually become ordinary.
Should we test many channels or focus on one?
Qualify many, test few, focus on one. Testing several simultaneously with a fixed budget produces several underpowered tests and no usable conclusion.
How often should channel selection be revisited?
On a set schedule, because channels decay predictably. A channel chosen years ago and never re-examined is a legacy decision rather than a current one.
What is the most underused idea in the framework?
The contrarian principle. Founders avoid channels that are uncomfortable or unfamiliar, which is exactly why competitors are also avoiding them. Early entry into an uncrowded channel usually beats optimisation in a saturated one.
Sources
Weinberg, G. and Mares, J., Traction: A Startup Guide to Getting Customers — the nineteen channels, the Bullseye framework, and the 50% rule, built on interviews with approximately 40 founders
Chen, A., The Law of Shitty Clickthroughs
Published commentary on the adequacy of the framework's recommended test budgets and durations
Structure your next phase
Zerologic builds acquisition systems channel by channel — qualified before spend, tested at a size that can produce an answer, and re-examined as channels age.
Talk to us: partners@zerologic.io · zerologic.io
