Positioning and Pricing as Product Decisions

Your homepage says: the intelligent platform that empowers modern teams to work smarter.

Nobody wrote that sentence badly on purpose. It went through four drafts and two rounds of feedback, and each round made it broader, because every narrowing prompted someone to say "but that excludes the customers who—".

Now read three things your actual customers said last month, taken from support tickets and a renewal call:

"We use it to catch the invoices that don't match the PO before they hit accounts payable."

"Honestly it replaced a spreadsheet Devon built that nobody else could maintain."

"It's the only thing that tells us which of our suppliers is quietly slipping."

Those are narrower, stranger, and more specific than anything on the homepage. They are also more interesting. Somebody reading them who has that problem would stop.

Here is the awkward part. The homepage sentence is true. It is true of your product and it is true of forty others, and a buyer skimming it learns nothing they can use — including whether to keep reading.

The team knows the homepage is weak. What they cannot agree on is what to replace it with, because every specific version excludes somebody they hope to win.

So: where does a positioning statement actually come from, and why does the right one always feel like a loss?

Positioning is context, and you choose it

A buyer landing on your product does not evaluate it in a vacuum. They slot it into a category — oh, it's one of those — and that slot determines what they compare you against, what they expect it to cost, and what counts as good.

They will do this in about four seconds, with or without your help.

Positioning is the context you deliberately place a product in so that its value is obvious to the people who should buy it. Not a slogan. Not a description of features. A claim about what kind of thing this is. Change the frame and the same product becomes obviously excellent or obviously overpriced, without a line of code changing.

How to use it. Work from evidence rather than aspiration. Five inputs, in this order, because each one constrains the next:

1. What would customers use if you didn't exist? The real alternative, which is usually a spreadsheet, a contractor, or nothing — not the named competitor your team argues about in meetings.
2. Which of your capabilities do those alternatives lack? Concretely. Not "better UX."
3. What value do those capabilities enable? Stated as an outcome for the customer, not as a feature of yours.
4. Who cares about that value disproportionately? There is always a segment that cares much more than the average. Find it.
5. What market frame makes the value obvious to them? This is the sentence.

How to spot it going wrong. The characteristic error is doing this backwards: inventing the frame first, then hunting for evidence. It produces language that is fluent, category-shaped, and true of everyone.

The test is one question and it takes ten seconds: could a competitor put their name on this sentence? If yes, it is not positioning. It is a description of a market.

Why the right answer feels like a loss. Honest positioning is almost always narrower than the team hoped, and it excludes buyers. That is not a side effect to be minimised — it is the mechanism. A statement that excludes nobody informs nobody. The discomfort you feel choosing it is the same discomfort a buyer would have felt failing to understand the broad version, moved from them to you.

One thing that has changed. Positioning statements are the single easiest artifact to generate and among the hardest to generate well, because the inputs are customer language nobody wrote down. Ask a model for positioning and you get something fluent and category-shaped, assembled from how the category already talks about itself — which is precisely the thing you are trying to differentiate from. The failure signature is a sentence that would fit forty products, and the test above catches it in one read.

Pricing is a choice, not a forecast

"What should we charge?" gets treated as a question about the world — as though there is a correct number and the job is to estimate it. That framing produces research designed to answer it: willingness-to-pay surveys, competitor tables, value-based calculators.

Then the number arrives and the argument continues anyway, because the research did not answer the question anyone actually had.

A price is not a prediction. It is a decision about which customers you want, what behaviour you want from them, and what you are willing to give up to get it.

How to use it. State what the price is for before choosing it. Real answers look like: land more accounts at the cost of margin; raise revenue from existing customers at the cost of some churn; deter a segment you serve badly; signal seriousness to enterprise buyers.

Then name what each candidate price sacrifices — volume, margin, positioning, the ability to raise later, the complexity of the sales conversation. Then ask what would have to be true for this number to be wrong, and how you would find out.

How to spot it going wrong. A pricing recommendation that arrives as a single number with a rationale and no stated sacrifice. Chapter four's costless tradeoff, in the one place where the tradeoff is the entire content of the decision.

On willingness-to-pay research. Treat it as a hypothesis generator only. "Would you pay £X" is a forward-looking question asked of someone who has not had the problem cost them anything — chapter two's rule, and it does not stop applying because the subject is money. What is evidence: what they pay today for the alternative, what they paid for the workaround, what they have cancelled.

One thing that has changed. A model will produce a confident price with a plausible rationale and a competitor table, none of which is evidence, and the fluency of the output substitutes for the judgment that was never made. The useful division of labour is exactly the one from the next section: let the model enumerate options and mechanisms, keep the part about what you are willing to trade.

Prediction is cheap; judgment is the scarce half

This is the clearest place in the domain to see a distinction that runs under the whole course.

Think of AI economically as a large drop in the cost of prediction — filling in missing information. When something gets cheap, demand rises for the things that complement it. Prediction's complement is judgment: knowing what outcomes are worth, what to do when the prediction is wrong, and what the decision is actually for.

How to use it. Split any decision into its two halves. The prediction half: what will happen? The judgment half: what do we want, what would we trade for it, what does an error cost? Automate toward the first. Keep and invest in the second.

Pricing shows this cleanly. A model can forecast which accounts will churn at a given price, and that forecast can be genuinely good. It cannot decide what churn is worth preventing, at what price, in which segment — because that is a question about what you want, and there is no fact about the world that answers it.

How to spot the confusion. Any argument of the form "AI can now do X, so we should Y" where the leap from X to Y crossed the prediction/judgment line without stopping. The tell in a document: a recommendation whose support is entirely a forecast, with no statement of what the forecast is being traded against.

Why the bar rose rather than fell. When everyone's prediction is equally cheap, the differentiated part of the work is the part that was never prediction. That is the thesis of this whole product, and pricing is where it is easiest to see: two companies with identical churn models still make different pricing decisions, and the difference is entirely judgment.

Back to the homepage

Run the five inputs over those three customer quotes and something specific appears. The real alternative is Devon's spreadsheet. The capability the spreadsheet lacks is matching invoices against purchase orders automatically and flagging supplier drift. The value is catching bad invoices before payment. The people who care disproportionately are finance teams at companies with many suppliers and thin AP staffing.

The resulting sentence is narrower than "modern teams." It excludes buyers. Someone in the room will say so, and they will be right.

They will also be describing the point.

The rewrite that lost buyers on purpose

A team sells a scheduling tool. Homepage: smart scheduling for growing teams.

They collect what customers actually say, which takes an afternoon of reading support tickets and two renewal-call recordings. A pattern shows up immediately and nobody had noticed it: almost every enthusiastic customer runs shift work across multiple locations, and the thing they mention is not scheduling. It is handling last-minute swap requests without a manager having to arbitrate.

Run the inputs. Real alternative: a group chat and a manager's phone at 6am. Capability the alternative lacks: rule-based swaps that auto-approve when coverage rules are satisfied. Value: managers stop being the bottleneck for shift changes. Who cares most: multi-location retail and hospitality with hourly staff.

New sentence: shift swaps that don't need a manager, for multi-site hourly teams.

What this costs, and it is not nothing: professional-services firms — currently 20% of customers — no longer see themselves on the homepage. They will keep renewing, because they already have the product, but the top of the funnel stops producing them.

That is a real loss, and it is the decision. The team accepts it because the professional-services segment converts at a third the rate and churns at twice — which they knew separately, in a dashboard, and had never connected to the homepage.

The general move: positioning that excludes is a choice about who you are for, and it is usually the same choice your retention data made for you a year ago without telling anyone.

Taking apart the AI analyst's pricing recommendation

A founder asks a model to recommend pricing for a new tier. What comes back is good work: a competitor table with eight products, a value-based rationale, a recommendation of £79 per seat per month, and a projected revenue impact.

The document is well organised, internally consistent, and confidently worded. It is also not a pricing decision, and taking it apart is instructive because the failure is not in any individual claim.

The competitor table. Real products, real prices. But two of the eight price per workspace rather than per seat, which makes the numbers non-comparable, and the table does not say so. Chapter three's measured how, in a pricing document.

The value rationale. "Customers save an estimated 6 hours per week, worth £180 at average loaded cost." Where did six hours come from? Nowhere traceable. It is a plausible number in the shape of a finding — and it is load-bearing, because the entire price follows from it.

The projection. Assumes adoption at a rate with no stated basis, and the revenue impact is that assumption multiplied by the price. Two unsupported numbers multiplied together, presented as a forecast.

What is missing entirely. What the price is for. Nothing in the document says whether £79 is meant to land more accounts, raise revenue per account, or push buyers up from the tier below. Nothing says what it sacrifices. Nothing says what would make it wrong.

The rebuild. The founder keeps the model's genuine contribution — it enumerated the competitive landscape and the mechanisms faster than a person would — and redoes the part it could not do:

What the price is for: pull existing £29 customers upward, because the base has grown and the top of it is under-monetised.

What each candidate sacrifices: £59 converts more of the base and makes raising later harder. £79 converts fewer, positions against the enterprise tier awkwardly, and leaves room. £99 signals a different category than the product currently delivers.

What would make it wrong: if fewer than 15% of eligible accounts upgrade in the first quarter, the value story is not landing and the problem is the packaging, not the number.

What is actually evidence: what those customers pay today for the tools this would replace. That data exists in sales-call notes and nobody had pulled it.

The final number might well be £79. That is not the point. The point is that it is now a decision somebody made rather than a number that arrived.

The narrower frame that raised the price

A company sells a document-review tool at £40 per seat, positioned as faster document review for busy teams. Growth is flat and every deal is a price negotiation.

Reading the customer base: three-quarters of revenue comes from legal teams doing contract review, and those accounts churn least and negotiate least.

Reframing the product as contract review for in-house legal changes what buyers compare it against. Against generic document tools it looked expensive at £40. Against paralegal hours and outside counsel it looks inexpensive at three times that.

Nothing about the product changed. The comparison set changed, and the comparison set is what "expensive" means.

The costs, and they are real: every non-legal customer is now off-category, and some will leave. The sales motion has to change, because selling to in-house legal is a different conversation with a different buyer. And the reframe is expensive to reverse — going back to the general position after claiming the specialist one means giving up the specialist credibility, which is the thing that took longest to build.

That last cost is the one teams routinely miss. Positioning changes are not free to undo, which is why they deserve the same treatment as any other one-way decision: name what would tell you it was wrong, before you commit.

The pricing recommendation, taken apart and rebuilt

An invented composite. The artifact is the kind of document a competent model produces today; the company is not real.

The situation. A three-year-old company selling supplier-risk monitoring to mid-market manufacturers. £3.2m annual revenue, one product, flat pricing at £29 per seat per month with a 20-seat minimum. Growth has slowed from 6% to 2% monthly over three quarters. The board has asked for a pricing review.

The founder, short on time, briefs a model with the customer list, the competitor set, and last year's win/loss notes, and gets back an eleven-page pricing strategy.

What the document contains. A three-tier structure at £29 / £79 / "contact us." A competitor comparison across eight products. A value-based justification building to £79. A migration plan for existing customers. A projected £740k incremental annual revenue. A risks section listing execution risks: migration complexity, sales-team retraining, billing-system changes.

It is, genuinely, a competent artifact. A year ago it would have taken an analyst three weeks.

First pass — traceability rather than structure. Three load-bearing claims hold the whole thing up:

1. Customers realise £180/week in value. Traces to a "6 hours saved" figure with no source. It is not in the win/loss notes and not in any customer document. It is the number the entire price is derived from, and it was generated.
2. Competitors price at £65–£95 for comparable capability. The prices are real. "Comparable" is not: two of the eight are per-workspace, and three include a data feed this product does not have. The comparison is between different things at different units.
3. 35% of eligible accounts will upgrade in year one. No basis given. Multiplied by the price, this is the £740k.

Two of the three are fabricated in the specific way that matters: not invented from nothing, but shaped like findings — a plausible figure, a real competitor list — so they pass the cheap checks a busy reader runs.

Second pass — what the document is not. It has no statement of what the price is for. No sacrifice named for any option. No disconfirming condition. Its risks section lists execution risks only, which is chapter four's tell arriving intact.

It is a well-formatted prediction where a judgment was required.

The rebuild, and what it keeps. The founder does not discard it. Two parts are genuinely useful and were fast: the enumeration of the competitive set, and the list of packaging mechanisms — per-seat, per-supplier-monitored, tiered by alert volume — several of which the team had not considered.

What she adds:

What the price is for. Growth has slowed because the product is under-monetised at the top, not because it is over-priced at the bottom. The goal is revenue per account from the largest customers, without disturbing the small ones who are the funnel.

The evidence that exists. What these customers pay today for the alternative: one full-time analyst per company doing this manually, or a consultancy retainer. That is in the sales-call notes, untouched by the model because it was in prose, not in a spreadsheet. It is also the only real evidence in the entire exercise.

The options, with sacrifices. Keep £29 and add a per-supplier-monitored component: aligns price with value and complicates the sales conversation. Move to £29/£79 tiers: simple, and the tier boundary is arbitrary, so it will be negotiated in every deal. Raise the seat minimum: raises revenue per account and closes off the smallest customers, who are the funnel.

The disconfirming condition. If the largest ten accounts push back on per-supplier pricing in the first five conversations, the value story does not survive contact and the tier structure is the safer move.

Where it lands. Per-supplier component, piloted with five accounts before any public change, with the pushback condition written down first.

What the case is about. Not that AI-generated pricing work is worthless — it saved real time and produced an option the team had missed. It is that the document answered the prediction question and presented it as though it had answered the judgment question, and it did so fluently enough that the substitution was invisible. The founder's contribution was not analysis. It was knowing which half of the decision she was still responsible for.

Where a good decision could still go wrong. The five-account pilot selects the five friendliest accounts, so the pushback condition is tested against people who will not push back. The per-supplier component makes revenue harder to forecast, which the board will notice next quarter. And the "6 hours saved" figure, having been written down once, reappears in a sales deck six months later with the original document as its source.

Look for what they already pay for or hack around. Existing spend and existing workarounds are revealed preference — someone already paid a cost to solve this. In positioning it tells you the real alternative; in pricing it is the only cheap evidence about value that is not a forecast. Tell: the artifact names current tools, spreadsheets, contractors, or manual processes and what they cost. Absent: the problem described only in adjectives.

Say what the chosen option costs. In pricing this is the whole decision, not a caveat on it. A recommendation naming only benefits has not made a choice. Tell: an explicit "what we give up" — the segment excluded, the margin traded, the ability to raise later, the reversibility lost.

Establish the benchmark before arguing about the number. A competitor price table is worthless without the units and the inclusions. Five minutes checking "comparable how" reframes the discussion. Tell: the comparison states its unit and what each price includes. Absent: a table of numbers with no definitions.

Hypothetical validation

What it looks like. Willingness-to-pay surveys, "would you pay £X," and feature-request rankings collected from people who have never had the problem cost them anything — presented as evidence for a price.

Why smart people do it. It scales, it produces charts, and it arrives fast. Real behavioural evidence about price is slow, small-n, and awkward to collect because it means asking people about money they have actually spent.

The correction. Convert every forward-looking question into a backward-looking one. What do they pay today, for what, and what did they cancel? Treat surveys as hypothesis generators, never as validation — and note that the presence of a number makes this failure harder to see, not easier.

Naked metric

What it looks like. A price comparison table with no units, no inclusions, and no definitions. "Competitors charge £65–£95." A value claim with no traceable source: "customers save 6 hours per week."

Why smart people do it. The numbers are real, or real-adjacent, and supplying the definitions is extra work that can only make the comparison messier. A clean table is more persuasive than an honest one.

The correction. Compared to what, at what stage, in what category, measured how. In pricing the fourth question does most of the work: two products priced per different units are not comparable, and that is the most common way a competitor table misleads.