Your positioning problem is usually a language problem

In brief: Voice-of-customer research means collecting the words buyers use for the problem, the alternatives they considered and the result they wanted, then using those words to make positioning decisions. I get most of what I need from recent buyer interviews, reviews and sales-call transcripts. The output is a message map that feeds campaigns, landing pages, scoring and the rest of the acquisition engine.

The first thing I do on a new engagement is read the homepage. The second is listen to three sales calls. They almost never speak the same language.

The homepage says “AI-powered workflow automation for modern revenue teams.” The last three closed-won customers, on recorded calls, said they were sick of copying deals between Salesforce and a spreadsheet every Monday.

Buyers search in the second language and forward the second language to their boss. On a landing page they skim past the first.

Nobody chose this on purpose. Homepage copy usually gets written early, by the people who know the product best. That often means it ends up furthest from how buyers actually describe the problem.

The ads, nurture emails and landing pages then inherit that language. So before I start changing campaigns, I want to know whether the words underneath them are any good.

What positioning actually is

April Dunford, in Obviously Awesome (2019),¹ describes positioning through five components:

• Competitive alternatives. Whatever a buyer would do otherwise, including a spreadsheet and an intern.
• Unique attributes.
• The value those attributes deliver, with proof.
• The customers who care most.
• The market category that frames it all.

Ries and Trout came at the problem from a different direction in Positioning: The Battle for Your Mind (1981).² Their useful point for me is that positioning ultimately exists in how the prospect understands the product relative to the alternatives they already know.

You can influence that perception, but you have to start by understanding what the buyer is actually comparing you with.

That’s where voice-of-customer research comes in. What the buyer would otherwise have done is a question with a factual answer. So is the outcome they cared about enough to buy.

I’d rather collect those answers than settle them in a workshop.

The three sources I use most

Kurt Vonnegut’s advice to writers was to write to please just one person. The marketing version needs a step in front of it: work out who that person is and listen to how they talk.

Three sources get me most of the way there:

• interviews with recent buyers
• reviews of the product and its competitors
• the company’s own sales-call recordings

None of them needs an elaborate research setup. Mostly they need access, a spreadsheet and enough patience not to start rewriting what people said.

Customer interviews: the question bank

I want people who bought recently, plus a handful who evaluated the product and went elsewhere.

The recency matters. Somebody who bought two years ago tends to give you a tidy explanation of why they bought. Somebody who bought last month is much more likely to remember what was actually happening when they started looking.

On sample size, there’s proper research to lean on. Griffin and Hauser’s “The Voice of the Customer” (Marketing Science, 1993)³ found that 20 to 30 interviews in a homogeneous segment surfaced roughly 90% of customer needs, and that a single analyst reading the transcripts identified only about half of them.

The second finding is the one I act on. I code the transcripts twice, and if there’s a second person available I like them to read the same set as well.

For a single-segment SaaS product, I normally start with 10 to 12 interviews and keep going until three in a row produce no phrase I haven’t already got.

The questions are deliberately concrete:

Record everything.

The phrasing is what I’m there for. Notes written up afterwards usually keep the gist and lose the part I actually need.

Review mining: the G2/Capterra/TrustRadius workflow

Reviews are interviews somebody else already ran, at scale, with the useful addition that I can do the same work on competitors.

The workflow takes a spreadsheet and a wee bit of patience:

1. Pick the product plus the two or three alternatives that show up in sales calls. Pull 100 to 200 reviews per product into a sheet, one row per review, with the source URL.
2. Add coding columns: trigger (why they went looking), prior tool (what it replaced), outcome (results, ideally with a number), friction (complaints, gaps), role (who is speaking).
3. Fill the columns with verbatim fragments only. No paraphrase. Every row keeps its URL so any phrase is traceable.
4. Count phrase frequency across the trigger and outcome columns. Phrases that recur across the product and its competitors are usually category language. Phrases that appear mainly in the product’s own reviews are worth looking at as possible differentiation.

Two parts of the reviews tend to be especially useful.

Three and four-star reviews often contain the trade-off somebody accepted to buy the product. Five-star reviews are much more likely to tell you that everything is wonderful.

I also spend a lot of time in the “what do you dislike?” answers on competitor profiles. Those complaints are useful when I later write comparison pages or decide which difference is worth testing in an ad.

Sales-call transcripts: the coding frame

If the sales team runs Gong, Chorus, Fireflies or anything similar, the company already owns the biggest pile of voice-of-customer material it’s likely to get. In my experience, it’s also one of the least likely datasets to have been read with positioning in mind.

I pull 20 to 30 discovery calls, with a mix of closed-won, closed-lost and still-open opportunities.

I start with the first ten minutes.

That’s normally where the prospect is describing what brought them to the call, before the rep has spent much time shaping the conversation around the product.

I use the same sheet structure as the reviews, with two extra columns:

• objection, copied verbatim
• alternative named, meaning whichever competitor or workaround they mention without being prompted

The rule I have to police hardest in my own notes is simple: if it isn’t a quote, it doesn’t go in the sheet.

Once I start summarising what somebody said, I’m usually putting the company’s own language back into it. At that point the research gets much less useful.

Before and after: what buyer language does to copy

Take a vague version written without that research:

“Streamline your revenue operations with AI-powered insights.”

And a version rebuilt from the coding sheet of a Salesforce-adjacent product:

“Stop rebuilding the board pack every Monday morning. Pipeline numbers pull straight from Salesforce, so the report your CEO sees matches the CRM.”

The second version works because it’s specific.

It names the task, the board pack, and the moment the problem shows up, Monday morning. It says what the buyer is doing instead: rebuilding the thing by hand. And it explains where the numbers come from, so there’s something behind the claim.

None of that was invented. Every part came out of the research.

The same material is useful in paid search because buyers also give you the words they use when they go looking. “Salesforce pipeline reporting” is a plausible search. “Streamline your revenue operations” is much less likely to be one.

Google’s Quality Score includes ad relevance and landing-page experience among its diagnostic components.⁴ I’m not trying to optimise a campaign around Quality Score itself, but using the same language across the search, ad and page gives me much better raw material than starting with internal product terminology.

The 95:5 rule and category entry points

Professor John Dawes at the Ehrenberg-Bass Institute published the 95:5 heuristic through the LinkedIn B2B Institute in 2021.⁵

The exact percentage is less interesting to me than the underlying point: most potential buyers in a category aren’t shopping at any particular moment.

That changes what I need from the research.

Jenni Romaniuk’s category entry points give the memory side of this a useful structure (Better Brand Health, 2023).⁶ A category entry point is a situation that brings the category to mind.

“Board meeting on Thursday and the pipeline numbers don’t reconcile” is useful.

“Revenue operations” isn’t really a situation.

This is why the first interview question asks somebody to take me back to the day they realised they needed something. I’m trying to find the circumstances around the problem, not only a description of the product they eventually bought.

For the kind of work I do, I usually finish this stage with a short list of those situations and the buyer language attached to each.

That gives me something useful for both sides of demand.

For somebody already shopping, I can write directly to the problem and the alternatives they’re considering.

For somebody who isn’t shopping yet, I can build a message around a situation they recognise, such as the Thursday board meeting, without pretending they’re already looking for software.

The in-market buyer does a lot of the purchase without a salesperson present. Gartner has reported that B2B buyers spend only a minority of the buying process meeting with potential suppliers, and only a small share with any individual supplier when comparing several.⁷

That leaves the website, comparison content, documentation and whatever other material the buyer can access on their own doing a lot of work.

Increasingly, that material may also be read and summarised by an AI system before somebody visits the page directly.⁸

Either way, the language on the page has to make sense without a salesperson there to translate it.

Where the output goes

What I want out of all this is a message map, normally one page per segment.

It contains:

• the category entry points and the verbatim trigger language around them
• the alternatives buyers actually named
• the outcome metrics they used
• the objections that came up most often
• the proof available for the claims

Then I use it elsewhere in the engine.

• Paid campaigns. Each campaign takes its message from one entry point or query cluster and gets its own conversion action, which is what the demand generation post covers. The frequency counts in the sheet help me decide which claims are worth testing first.
• Lead scoring. The role data and named alternatives help tighten the fit criteria. If the company matches the segment and the title looks like the people who actually described the problem, that gives me part of the scoring model.
• Landing pages. Headline, objection block and proof block come straight off the map. I build these as pages controlled in code, because fixed templates have a habit of making the message fit the available slots rather than the other way round.
• The feedback loop. A free-text “how did you hear about us?” box on the demo form is worth more than it looks. It picks up channels and phrasing analytics can’t see, and it keeps producing material for the next revision. Where that field lives in reporting is covered in the attribution post.

Positioning is the one part of the engine where what I hand over is mostly a document rather than a working system.

Everything after it, from campaigns to scoring to pages, gets built from the language in that document. That’s why I want it done before I start changing the rest.

When I come into an engagement where the ads, pages and nurture emails were written before anybody listened back through a sales call, that’s usually one of the first places I look.

It’s not the worst diagnosis to get.

Words are the cheapest part of a funnel to fix.

I’m engaged with client work just now, so none of this is a pitch. But if you’re building growth-stage B2B SaaS and want this kind of work when capacity opens, tell me what you’re building. I read every note.

Sources

1. Dunford, A., Obviously Awesome, 2019. ↩

2. Ries, A. and Trout, J., Positioning: The Battle for Your Mind, McGraw-Hill, 1981. ↩

3. Griffin, A. and Hauser, J.R., “The Voice of the Customer,” Marketing Science 12(1), 1993. ↩

4. Google Ads Help, “About Quality Score for Search campaigns.” support.google.com (opens in new tab) ↩

5. Dawes, J., “Advertising Effectiveness and the 95-5 Rule,” LinkedIn B2B Institute / Ehrenberg-Bass Institute, 2021. business.linkedin.com (opens in new tab) ↩

6. Romaniuk, J., Better Brand Health, Ehrenberg-Bass Institute, 2023. ↩

7. Gartner, “The B2B Buying Journey.” gartner.com (opens in new tab) ↩

8. Aggarwal, P. et al., “GEO: Generative Engine Optimization,” KDD 2024. arxiv.org (opens in new tab) ↩