The Bullseye framework for channel selection: how I decide what to scale and what to drop

The problem Bullseye solves

Most businesses I meet do not have a product problem. They have a distribution problem. The product is decent, the offer is fair, and the marketing is a scatter of half efforts: a bit of Google Ads, a LinkedIn post a week, a stand at one show, an agency retainer nobody can account for. None of it gets enough attention to work, and none of it gets enough budget to fail clearly either. So it drifts.

The Bullseye framework is a way out of the drift. It is a short process for choosing the one marketing channel to concentrate on, built on a single observation: at any stage of a company's life, one channel usually dominates how customers arrive. The authors of the framework put it plainly in their own excerpt of the book: most businesses get zero channels to work, and if you can get one to work properly you have a business.¹ The job is therefore not to be everywhere. The job is to find the one channel that moves the numbers, prove it with small tests, and then put everything behind it.

That is the whole idea. The rest of this article is how I run it, with the numbers written out.

Where the framework comes from

Bullseye comes from the book Traction: How Any Startup Can Achieve Explosive Customer Growth by Gabriel Weinberg and Justin Mares, published by Portfolio, a Penguin imprint, on 6 October 2015.² Weinberg founded the search engine DuckDuckGo, so the book is written by someone who had to find a channel for a product nobody was asking for. The publisher's own description says the book teaches "the nineteen channels you can use to build a customer base" and offers a framework in three steps, called Bullseye, to work out which of them will work best for your business.²

I mention the provenance because the framework is often presented as folk wisdom, with the ring names muddled and the channel list rewritten to suit whoever is selling something. The original is tidy and worth knowing exactly. An earlier edition of the book existed in 2014 under a different subtitle, but the Portfolio edition is the one I quote.

The 19 channels, and why the list is longer than you think

The outer ring starts with every channel that could possibly bring customers. Weinberg and Mares count 19 of them, and the list is the single most useful thing in the book, because almost everyone's mental list has about six entries. The 19, as the book's chapters name them:³

• Viral marketing
• Public relations
• Unconventional public relations
• Search engine marketing
• Social and display ads
• Offline ads
• Search engine optimisation (SEO)
• Content marketing
• Email marketing
• Engineering as marketing
• Targeting blogs
• Business development
• Sales
• Affiliate programmes
• Existing platforms
• Trade shows
• Offline events
• Speaking engagements
• Community building

A few of these need a word of explanation. Engineering as marketing means building a small free tool, a calculator or a checker, that attracts the people you want and shows them what you can do. Existing platforms means riding on somewhere your customers already are, an app store, a marketplace, a large social network. Targeting blogs means getting in front of the small, specific publications and newsletters your customers already read, before you can afford the big ones. Unconventional public relations is the stunt, the contest, the handwritten thank you note that gets talked about.

Notice what is not on the list, because this is where the old version of this article went wrong. Influencer outreach is not a channel of its own in the book, it sits inside targeting blogs and social ads. Partnerships are business development. Direct mail is offline ads. Podcast sponsorship is offline ads or targeting blogs depending on how you do it. Display is already in social and display ads. The list is deliberately coarse so that you are forced to think about every category rather than a fashionable tactic inside one of them.

Before you start: know what a customer may cost

The book begins with brainstorming. I begin one step earlier, because a test can only be judged against a number, and the number is the most you can afford to pay for a customer. Without it, every test result looks either thrilling or hopeless depending on your mood.

The number comes from two things you already know: how much contribution a customer leaves you each month, and how many months you are prepared to wait to get your acquisition money back. Contribution here is the monthly price times the gross margin, in other words what is left after the direct cost of serving the customer. The payback period is a business decision, not a formula. A company with cash in the bank can wait 18 months. A company living on this quarter's revenue cannot wait six.

$$\text{CAC}_{\max} = (P \times g) \times T$$

Here $\text{CAC}_{\max}$ is the most you can spend to acquire one customer (the ceiling on the customer acquisition cost, CAC), $P$ is the monthly price the customer pays, $g$ is the gross margin as a fraction (so 80% is $0.8$), and $T$ is the payback period in months. The bracket, $P \times g$, is the monthly contribution. If you want the groundwork on margins as percentages, it is in Foundations: percentages.

Worked example: Rotaline sets its ceiling

Every number in this article is made up, and the company does not exist. Rotaline is a small business to business (B2B) SaaS in Limerick. Sinéad, who founded it, sells rota and timesheet software to independent home care agencies. The plan costs €120 a month. Hosting, support and payment fees come to about 20% of that, so the gross margin is 80%.

Monthly contribution is therefore $120 \times 0.8 = 96$ euro. Sinéad has a modest cash cushion and decides she can wait six months to earn back what she spends on winning a customer. Her ceiling is:

$$\text{CAC}_{\max} = 96 \times 6 = 576$$

So $\text{CAC}_{\max}$ is €576, the most Rotaline may pay for a customer. Any channel that cannot get under that number, with room to spare, is not a channel for Rotaline right now, however glamorous it looks. Every test from here on is read against €576.

The outer ring: one idea for every channel

The first step in the book is the outer ring, which Weinberg calls "what's possible". You go through all 19 channels and write down, for each one, the single best idea you can think of for how that channel might work for your business.¹ Not a vague "we could do SEO", but a specific channel strategy: which keywords, which publications, which event, which tool you would build.

The point of this step is to counteract your own bias. Everyone has a favourite channel, usually the one they understand or the one that worked at the last place. The outer ring forces you to think honestly about the channels you would otherwise dismiss. The book is clear that you should not skip any channel, and that founders most often fail here by not thinking long enough to produce a real idea for each one.¹

Then you rank. For each channel, the idea goes into one of three columns: the inner circle (the ideas that look genuinely promising now), potential (ideas that could plausibly work), and long shots. The next step promotes the inner circle to the middle ring, and the book advises stopping where the excitement visibly drops off, which in its experience tends to be around the third channel.¹

Worked example: Rotaline's outer ring

Sinéad spends an afternoon on the 19 channels. A few of her ideas:

• Search engine marketing: Google Ads on phrases like "care rota software" and "home care timesheets", sending people to a page with a 14 day trial.
• Sales: a part time caller ringing the managers of independent home care agencies across Munster, offering a 20 minute demo.
• Trade shows: a table at a regional care providers' conference in Cork.
• Targeting blogs: a guest piece in the two newsletters care managers actually read, about rota mistakes that cost agencies money.
• Engineering as marketing: a free "rota cost calculator" that shows an agency how much unfilled shifts cost them.
• Existing platforms: a listing in the marketplace of a popular payroll product care agencies use.
• Viral marketing: a referral credit for agencies that bring in another agency. She writes it down honestly and gives it a low rank, because care agencies do not talk to each other much and a rota tool is not something you share for fun.

She ranks everything. Search ads, sales and trade shows land in the inner circle, because they are specific, she can start each within a week, and she can picture how each would deliver a customer. Targeting blogs and the calculator go into potential. Viral, public relations and offline ads are long shots. Three channels go to the middle ring.

The middle ring: three small tests, run in parallel

The middle ring is "what's probable". For each promoted channel you design a small test that answers three questions, which I take directly from the book because they are exactly right:¹

1. How much will it cost to acquire a customer through this channel?
2. How many customers are available through this channel?
3. Are the customers this channel brings the ones you want right now?

The tests are deliberately small. You are not trying to grow yet. You are trying to find out, as quickly and inexpensively as you can, whether the channel could ever matter. The book's rule of thumb is that a rough read on a channel should cost at most about a thousand dollars and a month of time, and often less.¹ In my practice I round that to €1,000 and four weeks per test, run the three tests at the same time, and refuse to extend any of them. Running them in parallel matters: tests take time once set up, and sequential testing is how a quarter disappears.

Cost per customer is the only one of the three questions that produces a number, and the number is simple:

$$\text{CAC} = \frac{S}{N}$$

Here $\text{CAC}$ is the customer acquisition cost, $S$ is what the test spent in euro and $N$ is the number of paying customers it produced. It is a rate, events over opportunities turned upside down, and the groundwork is in Foundations: ratios and rates. The other two questions are answered in words, and they matter just as much.

Worked example: Rotaline runs three tests

Sinéad runs all three tests in the same four weeks of March. Every figure is invented.

Test one, search ads. She spends €900 on Google Ads across her phrase list. The average cost per click comes to €3, so she gets 300 clicks. Of those, 18 start a trial (6% of visitors) and three convert to paying customers at the end of the 14 days. Cost per customer: $900 / 3 = 300$ euro. Volume: the keyword planner suggests the whole phrase list is searched a couple of thousand times a month in Ireland, so there is more to buy, but not endlessly more. Fit: the three customers are exactly her target, small independent agencies with 20 to 60 staff.

Test two, sales. A part time caller works 30 hours at €30 an hour, €900, plus €100 of phone and tooling costs: €1,000. She reaches 200 agencies, books 14 demos, and two become customers. Cost per customer: $1{,}000 / 2 = 500$ euro. Volume: the list of agencies is long enough that volume is not the constraint. Fit: good, but the two customers are larger agencies with longer sign off, and the sales cycle runs three to six weeks, so the test only just caught them.

Test three, a trade show. A table at a regional care conference in Cork costs €700, print and a banner €150, travel €100: €950. Sinéad has about 60 conversations, books nine demos and wins one customer within the month. Cost per customer: $950 / 1 = 950$ euro. Volume: there are perhaps four such events a year within reach, so the channel is capped. Fit: excellent, these are warm, well matched buyers, but slow.

Each bar is one channel tested. The height of the bar is what Rotaline paid for one paying customer in that test, in euro. Sinéad's ceiling of €576 sits between the second and third bars: search ads came in well under it, sales under it with less room, and the trade show well over it. The chart says a great deal, but hold your judgement for a paragraph, because three, two and one customers are very small counts, and the next section is about what to do with small counts.

Reading the results with expected value, not the raw count

Here is where most people go wrong, and the book does not spend much time on it. Three customers from one test is not a cost per customer of €300. It is a cost per customer of roughly €300, with a wide band of doubt around it. Had one more trial converted the figure would be €225. Had one fewer, €450. One customer from the trade show tells you almost nothing about what the next show would do. The arithmetic is right, the certainty is not. If you want the maths of why small counts are so noisy, I have written it up in how much data you actually need, and the groundwork is in Foundations: sampling.

So instead of ranking channels on a single noisy number, I ask a different question for each one: if I put a real budget into this channel next quarter, what range of outcomes do I believe is possible, how likely is each, and what is the result worth on average? That average is the expected value, and the idea is simple enough to do on the back of a beer mat.

$$E[X] = \sum_{i} p_i \, x_i$$

Here $E[X]$ is the expected value of the outcome $X$, the sum runs over each scenario $i$ you can imagine, $p_i$ is the probability you give that scenario (all the $p_i$ add up to $1$), and $x_i$ is the outcome if that scenario happens, in this case the number of customers won. Multiply each outcome by its probability and add them up. The idea, with a campaign example, is in Foundations: expected value, and the textbook definition is the long run average of a random quantity.⁴ I wrote about using it for bigger founder decisions in decision theory for founders.

The probabilities are judgements. They come from the test, from what you know about the channel, and from honesty about what could go wrong. That is grand. A written down guess you can argue about is far better than a single number you treat as fact.

Worked example: Rotaline scores the three channels on expected value

Sinéad has €6,000 to put into one channel over the next quarter. A customer is worth, in first year contribution, $96 \times 12 = 1{,}152$ euro. That ignores customers who leave during the year, which flatters every channel equally, so it is fine for comparing them. For a proper lifetime figure, the method is in survival analysis, churn and customer lifetime value.

For each channel she writes three scenarios for what €6,000 would buy, with a probability for each.

Search ads. The test cost €300 a customer, but a bigger budget means bidding on more and slightly weaker phrases, so the auction may get dearer (there is a whole article on why in auction theory and bidding). She says: 50% chance the cost holds at €300, so $6{,}000 / 300 = 20$ customers. 30% chance it rises to around €430, so about 14 customers. 20% chance it climbs to €750 as the good phrases run out, so eight customers.

$$E[\text{customers}] = 0.5 \times 20 + 0.3 \times 14 + 0.2 \times 8 = 10 + 4.2 + 1.6 = 15.8$$

Expected first year contribution is $15.8 \times 1{,}152 = 18{,}201.60$ euro. Take off the €6,000 spent and the expected net first year result is €12,201.60.

Outbound sales. More hours means reaching further down the list, into agencies that are a worse fit, and the sales cycle means some deals land after the quarter ends. She says: 50% chance of €500 a customer, 12 customers. 30% chance of €750, eight customers. 20% chance of €1,200, five customers.

$$E[\text{customers}] = 0.5 \times 12 + 0.3 \times 8 + 0.2 \times 5 = 6 + 2.4 + 1.0 = 9.4$$

Expected contribution $9.4 \times 1{,}152 = 10{,}828.80$ euro, so an expected net result of €4,828.80 after the €6,000.

Trade shows. €6,000 buys perhaps two larger stands in a quarter. She says: 30% chance of six customers across both, 50% chance of four, 20% chance of two.

$$E[\text{customers}] = 0.3 \times 6 + 0.5 \times 4 + 0.2 \times 2 = 1.8 + 2.0 + 0.4 = 4.2$$

Expected contribution $4.2 \times 1{,}152 = 4{,}838.40$ euro, which is less than the €6,000 spent. Expected net result: a loss of €1,161.60 in the first year.

The ranking is now clear and, more importantly, it is honest about doubt. Search ads win on expected value with a comfortable margin. The worst scenario, €750 a customer, would breach the ceiling, but the expected cost per customer is $6{,}000 / 15.8 = 380$ euro, well inside €576. Sales is a sound second, but its expected cost per customer is $6{,}000 / 9.4 = 638$ euro, already over the ceiling, so it is worth remembering rather than funding. Trade shows come out at $6{,}000 / 4.2 = 1{,}429$ euro a customer and are not a core channel for Rotaline at this stage, however pleasant the conversations in Cork were.

One caution that Foundations also makes: expected value says nothing about the one scenario you cannot survive. If a 20% chance of eight customers would have sunk the company, the answer is a smaller budget or a staged one, not a different channel.

The inner ring: focus, and let the other channels feed it

The inner ring is "what's working". Once one channel has clearly outperformed, the book's advice is unambiguous: direct all your traction effort and resources at it, and keep experimenting inside it to wring out every bit of growth until saturation or rising costs stop you.¹ This is the step founders most often get wrong, and I have got it wrong myself. The other two tests also sort of worked. It is very tempting to keep doing a bit of each. The book calls this a mistake, and it is: the second and third channels are distractions that take attention from the channel where the next improvement is worth most.¹

There is a subtlety that resolves most people's objection to this. Focusing on a core channel often means using other channels in its service. A search engine optimisation strategy needs links, and a good way to get links is publicity. A viral loop usually runs on email or an existing platform.¹ So the other channels do not disappear. They stop being strategies of their own and become tactics that feed the one strategy.

Worked example: what Rotaline does with its bullseye

Sinéad puts the €6,000 into search ads and treats everything else as support. Over the quarter she improves the landing page (the trial rate goes from 6% to 8%), adds the phrases that convert and removes the ones that only attract job seekers, and builds the free rota cost calculator from her outer ring as the thing the ads send people to for the colder phrases. That is engineering as marketing, but it is in service of the search channel, not a second strategy. The caller is kept on for four hours a week, but only to ring trials that have gone quiet, which is sales feeding the core channel rather than competing with it.

She also sets a tripwire. If the blended cost per customer rises above €500 for two months running, the channel is saturating and it is time to run Bullseye again with the new knowledge. That is the book's final point: Bullseye is not a one off. When the core channel stops moving the numbers, you go round again.

When no channel wins

Sometimes all three tests come back poor. The book's answer is to repeat the process with what you learned, looking at where each channel failed to deliver, and at the messaging you used.¹ It also makes a harder point: if you have gone round several times and nothing takes, the product may need more work. The bucket is still too leaky.¹

In my experience this is the most valuable outcome Bullseye can produce, and the least welcome. A channel test that fails is often a message test that failed. Before changing channel, check whether the page said something a care manager actually cares about, and whether the price and the promise match the person who arrived. I have a whole piece on getting message, market and model to agree before spending a euro, and it belongs before a second round of Bullseye, not after it.

The flow, start to finish

Here is the whole process as I run it, top to bottom.

Reading from the top: the first box is mine, not the book's, the cost ceiling that every later number is judged against. The next four boxes are the book's outer and middle rings. The scoring box is where I replace the raw test count with expected value. The decision splits into the inner ring, where one channel gets everything and the rest serve it, or a loop back through message and offer. The last box is the reminder that a core channel wears out, and the process is meant to be repeated.

Where I depart from the book

I follow Weinberg and Mares closely, but three things in how I run it are mine, and I want to be clear about which.

First, the ceiling comes before the brainstorm. The book treats cost per customer as a question each test answers. I think you need the answer to "compared with what?" before you spend the first euro, or you will talk yourself into a channel because the test went better than you feared rather than better than you need.

Second, I score with expected value rather than with the test's single number. Three customers is not a measurement, it is a hint. Writing down scenarios and probabilities makes the doubt visible and lets two people disagree about something specific.

Third, I set the saturation tripwire on day one of the focus phase, as a number, so that the decision to go round again is taken by the data and not by boredom. In the shops and software businesses I have worked on, the core channel has typically carried things for one to two years before the cost curve turned. When it turns, the honest move is the uncomfortable one: back to the outer ring.

What I have dropped from the old version of this piece is a tidy "80% of budget to the bullseye" rule I used to quote. It is not in the book, and I cannot defend the number. The book says focus solely on the core channel, with other channels only in its service, and that is both more demanding and more correct.

What to do this week

If your marketing is a scatter, take the 19 channels and an afternoon. Write one honest, specific idea for each. Work out your cost ceiling from your margin and the payback you can live with. Pick three ideas, give each at most €1,000 and four weeks, and run them at the same time. Then sit down with the three results, write the scenarios, and let the expected value choose. Put everything behind the winner, give the runner up a supporting role, and set the tripwire.

It is not clever. It is just a process that stops you doing a bit of everything, which is the one approach that reliably produces nothing. Fair play to anyone who gets through the 19 honestly, because a good few never do.

Footnotes

1. Gabriel Weinberg, The Bullseye Framework for Getting Traction, an excerpt from Traction published by the author on Medium, 27 September 2015, archived copy: https://web.archive.org/web/20241008072934/https://medium.com/@yegg/the-bullseye-framework-for-getting-traction-ef49d05bfd7e

2. Gabriel Weinberg and Justin Mares, Traction: How Any Startup Can Achieve Explosive Customer Growth, Portfolio (Penguin Random House), 6 October 2015, ISBN 9781591848363, publisher's page: https://www.penguinrandomhouse.com/books/319121/traction-by-gabriel-weinberg-and-justin-mares/

3. Traction: A Startup Guide to Getting Customers, book description listing the 19 traction channel chapters, Goodreads, https://www.goodreads.com/book/show/22091581.Traction_A_Startup_Guide_to_Getting_Customers

4. OpenStax, Introductory Statistics 2e, section 4.2 Mean or Expected Value and Standard Deviation, Rice University, 2023, https://openstax.org/books/introductory-statistics-2e/pages/4-2-mean-or-expected-value-and-standard-deviation