Positioning
Find the alternative your customer really uses, then fill Dunford's five components for an AI feature whose real competitor is doing nothing.
On this page
Positioning is the context a buyer reads a product in, and it decides who they compare it with, what they expect it to do and what they think it is worth. For an AI feature the tempting context is the other AI vendor, and it is usually wrong: in 31 of 40 interviews, the drafting feature's real competitor was the macro library the agents already had. April Dunford's method in Obviously Awesome fills five components in a fixed order, starting from that competitor, and each can carry a number.
The mistake: positioning against the rival
The first launch draft read "Smarter drafts than the other AI tools, in half the time." Sales had lost four deals to a rival drafting tool last quarter, so the copy was written at the loudest competitor in the building.
That line tells every buyer who was not shopping for an AI tool that this is a purchase among AI tools. It invites a head-to-head on draft quality, which the buyer cannot run and will settle on a demo. And it says nothing about what the agents would give up, the habit that decides whether the feature gets used.
The fix starts with a count. The weekly interviews had asked one story-based question all quarter: "Tell me about the last time the queue was still full at 6. What did you do?" Forty answers, sorted by what the team did.
Competitive alternatives: what they do today
Dunford's first component is what the customer would do if the product did not exist. Here it is "do nothing", which is never nothing: 200 macros learned by shortcut, a reply that takes about 3 minutes and, for the 14-agent team in jobs to be done, 130 tickets carried overnight.
Doing nothing is the hardest competitor because it appears on no invoice: staying costs nothing the buyer can see, and switching costs a new motion for every agent. So the component carries the status quo's cost in the buyer's units.
The other two alternatives differ in kind. The five teams paying for overtime already price the problem in hours and will compare the add-on with an hour of temp time. The four using a rival have already accepted the category; for them, and only them, a comparison belongs on a sales battlecard.
Unique attributes: what the macros cannot do
An attribute is unique relative to the alternative, not relative to the rival. Against macros the drafting feature has three:
- It reads the whole ticket and the help-centre article it cites, so the draft fits the case and the agent edits less.
- It starts from the macro the agent picked, so the 200 macros stay: it rides the habit force rather than fighting it.
- The first words stream onto the screen in about 500 ms, never slower than finding a macro.
"Built on a large model" is not on the list: the rival has one too, and the buyer cannot observe it in a trial.
Value: one minute, in the buyer's units
Attributes are what the product has; value is what the buyer gets, and here it is one number: a reply falls from 3 minutes to 2. For the 14-agent team, 560 replies a day × 1 minute = 560 agent-minutes, about 9.3 hours, which clears the 4 pm queue before 6.
Per seat, 40 replies × 1 minute is 40 agent-minutes a day. At an assumed loaded cost of $30 an agent-hour, that is $20 a day, about $440 a month across 22 working days, against the $5 a seat in pricing an AI feature. Against a competitor that costs nothing, the price is argued in the buyer's time, not against another vendor's price.
Value needs proof from the buyer's own queue. The one-day prototype timed 60 drafted replies against the 3-minute baseline and measured about 2.6 minutes, not 2, so the minute stays an assumption until live tickets measure it. "2 minutes instead of 3, measured on your tickets" is proof. "Up to 50% faster" is a claim.
Best-fit customer: who cares most
The minute is worth something only to a team with push, tickets carried overnight, and only where the draft is usually right, the short-reply 70% of tickets. So the best-fit customer is criteria with numbers: carries 50 or more tickets overnight on most days, answers at least 60% of tickets from the help centre, keeps a maintained macro library.
The 40-agent team with an empty queue at 6 pm, from the same jobs lesson, fails the first criterion: nobody hires a tool to clear a queue that clears itself. Positioning for them blurs the message for the teams that will renew. The best-fit list is also the first stage of the launch, because those teams' numbers move first.
Market category: the frame that sets expectations
The category tells the buyer what to compare the product with and what it should do. Two candidates: "AI support agent" and "reply drafting inside your help desk".
"AI support agent" puts the feature beside tools that answer customers on their own, an expectation it never meets: 8% of drafts are not acceptable on the golden set, so an agent reads every one. A buyer who expects an agent and gets a drafting tool feels short-changed by a product that works.
"Reply drafting inside your help desk" puts it beside macros, which it beats on all three attributes, and beside a per-seat add-on line the buyer already approves. Choose the second: a category is chosen for the expectations it sets, not the size of the market it names.
The statement, and where it goes
Filled in order, the five components produce one paragraph:
For support teams that carry tickets overnight, the drafting add-on writes each reply from the ticket and the macro your agent picked, so a reply takes 2 minutes instead of 3. Unlike macros alone, it fits the reply to the ticket. Unlike tools that answer customers for you, an agent reads every draft.
That paragraph is not copy; it is the brief every piece of copy is checked against: the launch announcement, the first sales slide, the onboarding email. The marketing course works the method in depth in Dunford's method.
In an interview
In a product-sense round ("who is the competitor?"), what is being tested is the order: alternatives found in interviews and counted, attributes relative to them, value in the buyer's unit with its proof, best-fit customer as criteria, category chosen for the expectation it sets. "31 of 40 teams used macros, so the competitor is doing nothing, and it costs them 3 minutes a reply" is the sentence the rest of the answer follows from.
The mistake that loses points is naming the loudest AI rival as the competitor and listing features against it: it skips the alternative most buyers use, picks a category the product cannot live up to, and leaves value as an adjective. The second is value without proof, "saves time" with no minute and no stopwatch.
Check yourself
1. Sales lost four deals to the rival last quarter and wants the homepage to compare against it by name. The interviews say 31 of 40 teams use macros. What goes on the homepage?
The macros comparison. The homepage is read by the 36 in 40 who were not shopping for AI; a rival comparison tells them they should be, then loses them to a demo contest. The rival goes on a battlecard. What would change it: a quarter where most lost deals go to the rival rather than to no decision, which the CRM's loss reasons show.
2. A 30-agent team handles 25 tickets an agent a day and carries 10 tickets overnight. Best-fit customer or not?
750 replies a day × 1 minute is 12.5 agent-hours of capacity, but the push is 10 tickets, which about 20 minutes of it clears. Not best fit for this job. Ask whether they pay for overtime or temps: that is a different alternative with its own number, and if they do, the hours are money.
3. Marketing proposes "AI support agent" as the category because search volume for it is ten times higher. Do you take it?
No. The category promises a product that answers customers by itself; at 8% unacceptable drafts it never will, so every buyer that search brings in is disappointed by a product that works. Take the smaller audience. Revisit only if the feature ever sends on its own, which needs zero wrong facts in at least 600 checked drafts first, the fewest that bound the rate under 0.5%.