The five decisions
Restate any marketing question as one of five decisions (who, what claim, which channel, what it costs, did it work) and name the number that settles it.
On this page
Marketing questions arrive in many shapes: should we be on short video, is the brand tired, why did sign-ups fall, can the agency have another $30,000 a month. Nearly every one is one of five decisions underneath: who we sell to, what we claim, where we say it, what it costs, and whether it worked. Each has one number that settles it. Argued without its number, a marketing question becomes a debate about taste, and the most confident person in the room wins it.
Five decisions, five numbers
| Decision | The question underneath | The number that settles it | Taught in |
|---|---|---|---|
| Who we sell to | Which segment gets the next year? | accounts × win rate × ARPU | module 1 |
| What we claim | Which alternative do we position against? | win rate against each alternative | module 1 |
| Where we say it | Which channel gets the next dollar? | cost per paying customer at the next dollar | module 3 |
| What it costs | Can we afford to acquire customers this way? | CAC, read as a payback month | module 2 |
| Whether it worked | Did the spend cause the result? | incremental customers against a holdout | module 4 |
Who we sell to. A segment is a group we can reach, that buys for one reason, at a price and sales cycle we can name; "SMBs" fails all three. Module 1's running example: 40,000 small clinics × 2% won × $1,200 a year = $960,000 a year. A smaller segment often wins once the cost of reaching it is in the sum, which is the beachhead arithmetic.
What we claim. Positioning is chosen against what the buyer would do instead, and the number is the win rate against each alternative. A product that wins more than half its demos against paper and phone and one in ten against a $400 practice-management suite, as module 1's clinic scheduler does, has a claim to make to clinics on paper and a fight to avoid. Dunford's method turns that into a positioning.
Where we say it. A channel is chosen at the margin: not "is paid search good" but "what does the next $10,000 there buy", priced per paying customer rather than per click or lead, because a channel's second $100k costs more than its first. Paid media basics builds the model.
What it costs. Acquisition cost against what a customer returns, read as the month the money comes back. For the course's default customer, $40 of margin a month at 5% monthly churn, a $333 CAC pays back in month 11 and a $521 CAC in month 21. Both clear an LTV/CAC of 1.5; only the first meets the course's 12-month payback target, a CAC ceiling of $368. CAC, LTV and payback derives both months.
Whether it worked. The number is incremental: customers the spend caused, counted against a group that did not see it. Attribution allocates credit; it does not measure cause. When eBay switched off search ads on its own brand name, almost all of that traffic still arrived through the unpaid result beneath them (Blake, Nosko and Tadelis, Econometrica, 2015). Holdouts, geo tests and power measures it properly.
Restate the question before answering it
The common failure is answering a question in the shape it arrived. "Should we be on short video?" invites an argument about where the audience spends its evenings, and nothing in it can be checked. Restated, it is a channel decision, settled by what a fixed test budget returns in paying customers against the CAC we can afford.
| The question as it arrives | The decision | The number that settles it |
|---|---|---|
| "Should we be on short video?" | where we say it | cost per customer from a $5,000 test |
| "The brand feels dated." | what we claim | win rate against the named alternative, this quarter against last |
| "Sign-ups fell 18% last month." | whether it worked | the fall against ordinary noise, then its cause |
| "The agency wants another $30,000 a month." | what it costs | cost per customer of the next $30,000, not the average |
| "Should we generate 500 posts a month?" | where, and what it costs | cost per visit with review counted, against the $0.67 a blog visit is worth |
The sign-up fall, worked: sign-ups have averaged 150 a month over the past year, and last month brought 123. Monthly counts of independent events vary by at least their square root, so a month moves by roughly √150 ≈ 12 around that average, and a fall of 27 is a little over two of those. Worth investigating, not proof that anything broke; the next step is a cause, such as a page that dropped out of search, not a new channel. The $2 click and the review cost come from the arithmetic a marketer carries; why a blog visit is worth a third of a click, $0.67, is in content and SEO.
The order the decisions run in
A channel cannot rescue a claim that loses, and a cheap CAC in the wrong segment shows up as churn in month three. The stages also multiply. If a demo costs $200 in media and the claim wins 10% of demos, a customer costs $2,000. Halving the media price gets that to $1,000; so does doubling the win rate to 20%, and that applies to every channel at once. A 10% lift is worth the same at any stage, so the useful question is which stage is cheapest to move, which the funnel and AARRR works through.
What AI changed, decision by decision
The five decisions are unchanged. Their inputs changed, and for two of them, so did what can still be measured.
- Who. Language-model personas draft interview questions and hypotheses in minutes; they cannot size or price a segment, the line synthetic research draws.
- What claim. A competent claim costs nothing to write, so claims separate on proof: customer evidence, proprietary data, a number the buyer can check.
- Where. Channels whose value came from scarcity fill up when a post costs under a dollar to generate, and answer engines cite pages instead of ranking them; see when content costs nothing.
- What it costs. A post falls from $400 to under $1 plus a $50 review. The cost moves to review and distribution.
- Whether it worked. An answer read on the results page produces no click to attribute, so weight moves to experiments and mix models: measurement when the click disappears.
In practice
When a question arrives from a co-founder, the board, an agency or a vendor, the first useful sentence names which of the five it is and which number would settle it: "That's a channel question. What would the next $10,000 there buy in paying customers?" It turns opinions about the medium into an estimate someone can check, and tells the asker what to bring back.
Ask a vendor which of the five their product moves, and by how much. A tool that writes ad copy moves production cost, rarely the number holding a company back; a promise of better "performance" should name the stage and the rate.
The mistake that costs money is working on a downstream decision while an upstream one is open: buying reach for a claim that wins one demo in ten. At $200 a demo every customer costs $2,000 in media, and no bid tuning does what doubling the win rate does to every channel at once.
Check yourself
1. Organic sign-ups have averaged 400 a month and came in at 340 this month, and your co-founder wants to start paid search to make up the gap. Which decision is this, and do you start paid search?
Whether it worked, first. Noise around an average of 400 is at least √400 = 20, so a fall of 60 is three of those and probably real. Find the cause: a page that lost its search position or a broken form is cheaper to fix than 60 sign-ups a month at $67, about $4,000 of paid media. Buy the channel only for a loss you cannot recover, priced per paying customer.
2. For a sales-led product, not the course's $50-a-month default customer: an agency quotes $40,000 to rework your messaging. You run 800 demos a year, all against your main alternative, win 15% of them, and a customer returns $3,000 of lifetime margin. How many points of win rate must the work add in a year to cover its fee?
$40,000 ÷ $3,000 ≈ 13.3 extra customers, and 13.3 ÷ 800 ≈ 1.7 points: from 15% to about 16.7%. That is a small lift, so the work is plausible. Agree only with the win rate measured on the same demo source before and after, or the result will be argued rather than counted.
3. A writing tool promises to cut your cost per post from $400 to $50. You publish 8 posts a month. Which decision does it move, and by how much?
What it costs, and only production: 8 × $350 × 12 = $33,600 a year. It does not change whether the posts bring customers, and if every competitor now publishes more, each post is worth less, so the saving is the ceiling of the benefit, not the floor.