Revenue is up eleven percent. Gross margin is down four points. The accounts consuming the most support, the most storage and the most compute are paying almost exactly what the light users pay. Sales keeps closing deals, but only after the same three discount conversations. Nobody in the company thinks pricing is the problem, because the price list looks reasonable and nobody has complained loudly.
That is the shape of the problem. The hard part is learning to know when your pricing model no longer works, as opposed to when your price is simply a little low, or your sales team is a little soft, or your costs had a bad quarter. These failures feel identical from the inside. They need completely different fixes.
The reason so much pricing advice is useless is that it treats pricing as one dial. It is three.
Three different things people call “pricing”
Before you can diagnose anything, you have to separate the layers, because a symptom in one layer is often blamed on another.
Price level
The number. Fourteen dollars instead of twelve. This is the only layer most companies ever touch, because it is the easiest to change and the easiest to reverse. It is also the layer least likely to be the real problem when something structural has gone wrong.
Packaging
What sits inside each tier, and where the boundaries fall. Packaging decides which customers get pushed up, which feel over-served, and how many conversations your sales team has to have about a single feature trapped in the wrong plan.
The pricing metric
The unit you charge for. Per seat. Per month. Per gigabyte. Per resolved conversation. This is the layer almost nobody audits, and it is the one that quietly decides whether your revenue grows when your customer’s value grows, and whether your revenue grows when your cost to serve grows.
A pricing model is really a theory about what the customer is buying. Per-seat pricing says: you are buying access for people. Usage pricing says: you are buying work done. When the theory and the reality drift apart, the number on the invoice stops meaning anything, and no amount of adjusting that number will fix it.
The U.S. Small Business Administration’s lean business plan format puts value proposition and finances in the same short document. That adjacency is the whole point. Pricing is the hinge between the two. If the value proposition has changed and the finances have not, the hinge has broken.
Look at the spread, not the average
Here is the single most useful diagnostic I know, and it takes an afternoon.
Pull your last twelve months of customers. Calculate gross margin per customer, not in aggregate. Then rank them. Do not look at the mean. Look at the distance between your most profitable customer and your least profitable one, at similar price points.
If every customer sits in a reasonably tight band and the whole band is too thin, you have a price level problem. Raise prices. That is a real answer and it is often the right one.
If the band is wide, and two customers paying the same amount produce wildly different margins, you have a pricing metric problem. Your unit is not tracking the thing that drives your costs. Raising the price now makes it worse, because you will raise it most on the customers who were already fine and least on the ones destroying your economics. You will lose the profitable accounts and keep the unprofitable ones. This is how a price increase can reduce profit.
Averages hide model failure almost perfectly. Dispersion exposes it. If your reporting only shows blended gross margin, you have built a dashboard that cannot see the problem you most need to find. Identifying the most profitable part of your business is the same exercise pointed in a friendlier direction.
Symptoms, and where they actually point
Discounting that clusters
Every company discounts. The question is whether the discounting is flat or lumpy.
If reps give roughly the same discount across most deals, your list price is simply above what the market will pay without a nudge. That is a level issue, and it is manageable.
If discounting clusters in one segment, one company size, or one deal shape, something structural is wrong. Reps are not negotiating badly. They are manually repairing a model that does not fit those buyers. A fifty-person agency being quoted per seat when only six people touch the product every day will always need a discount, because the price is answering a question the buyer did not ask. The discount is the patch. The metric is the bug.
Watch for the version of this that senior leaders miss: the biggest accounts getting the deepest discounts. That is usually read as enterprise buying power. Sometimes it is. Often it means your pricing scales faster than your delivered value at the top end, and the discount is the market correcting you.
The tier people actually wanted
Track two things that rarely appear on a growth dashboard: what customers downgrade to, and what single feature triggered an upgrade.
If upgrades are repeatedly driven by one feature stranded in a higher tier, your packaging boundary is in the wrong place, and you are taxing customers for things they did not want in order to get the thing they did. That works until a competitor prices that feature correctly.
If downgrades cluster at a specific tier, the tier below it is quietly doing the same job for less. Contraction revenue is more diagnostic than churn, because the customer stayed. They told you exactly which part of your packaging they refuse to pay for, and they did it with money instead of a survey response.
The thirty-second explanation test
Ask a rep hired in the last two months to explain your pricing to you as if you were a prospect. No spreadsheet, no calculator, no pricing page.
If they cannot do it in half a minute, your packaging has accumulated damage. Pricing complexity is usually sedimentary. Each layer was a sensible response to one deal, one competitor, one feature launch. Nobody ever removes a layer. Eventually procurement cannot model their own spend, and a simpler competitor wins deals on predictability alone.
Grandfathered customers hiding the truth
This one deserves more attention than it gets. Run your margin and retention numbers using only customers acquired in the last twelve months. Then compare that to the full book.
If the new-cohort-only view is noticeably worse, your current model is being flattered by legacy customers on old terms. Those customers are not coming back to rescue next year’s cohort. If the new-cohort view is better, your legacy base is a drag you have been tolerating, and the repricing conversation you keep postponing is the one with the most money in it.
Atlassian’s FY25 pricing changes are instructive here. The company made list price increases of roughly 15% to 25% on Data Center products effective February 2025, and separately changed advantaged pricing by around 30%, alongside a Cloud pricing and packaging update notified in September 2024 and effective the following month. The legacy and advantaged pricing changes sit right next to the packaging changes. Mature vendors do not just move numbers. They go back and fix the plans that stopped reflecting the economics.
What the AI shift actually broke
Two things changed at once, and they pull in opposite directions.
The first is cost to serve. For two decades, software’s marginal cost of an extra heavy user was close to irrelevant. That assumption is what made unlimited plans and flat per-seat pricing safe. When delivery involves inference, that assumption weakens. A per-seat plan with an AI feature inside it can have a user who costs five times another user on the same price. You will not see this in blended margin. You will see it immediately in dispersion.
The second is expectation. AI capability has moved from premium add-on to assumed inclusion in the categories where it became ordinary. Google says the best of its AI is included in Workspace Business and Enterprise plans rather than sold only as a separate add-on, and the Workspace pricing page merchandises Gemini features inside plans priced at $7, $14 and $22 per user per month on annual billing. Slack went further and removed its AI add-on from self-serve purchase in 2025, folding AI features into paid plans while raising Business+ from $12.50 to $15 per person per month on annual billing. Microsoft says Copilot Chat is included at no extra cost for organizations with an eligible Microsoft 365 subscription.
So the costs argue for charging by consumption, and the market argues for bundling. That tension is exactly why so many pricing models broke at the same time.
My position on add-ons
An add-on is a pricing experiment, not a pricing model. It is an excellent way to learn whether anyone will pay separately for a capability. It is a poor permanent structure once the capability becomes part of what the product obviously is, because at that point the add-on stops isolating willingness to pay and starts adding friction to every renewal conversation.
The exception is real and worth naming. Keep separate pricing when the value is discrete, countable and attributable. Intercom prices its Fin AI Agent on outcomes rather than access: $0.99 per resolution, $0.99 per procedure handoff, $0.99 per disqualification, $0.99 per self-serve routing, and $9.99 per qualification. Intercom has explained that it broadened from a resolutions-only framing because the old metric only recognized value when a conversation ended in a full AI resolution. That is a company noticing its pricing metric was undercounting the work the product was doing, and fixing the unit rather than the number.
Notice the direction of that change. It was not a price increase. It was a redefinition of what counts as a billable event. Those are the changes that actually move a business.
Seats, usage, or outcomes
The honest state of the market is a split, not a winner. Stripe’s own documentation supports flat rate, per-seat, tiered, usage-based, and fixed fee plus overage, because all of them remain in live use.
If you asked me what to build, I would push toward a floor plus a meter. A committed base fee that procurement can budget and forecast, plus a metered component tied to whatever actually drives your cost or your customer’s value. Stripe’s fixed-fee-plus-overage pattern is the plumbing for this and it is not exotic.
The reason is simple. Pure seat pricing is wonderfully easy to buy and increasingly dishonest about cost in AI-heavy products. Pure usage pricing protects your margin and terrifies finance departments, who will choose a predictable competitor over an unpredictable bill even when the unpredictable one is cheaper on average. The floor buys you predictability. The meter buys you margin. Most businesses need both, and the argument should be about where the floor sits, not about which philosophy wins.
One warning. A metered component only works if the customer can see the meter and believes it is fair. If your unit is something the customer cannot control or does not understand, usage pricing becomes a trust problem rather than a pricing model. Pick a unit the buyer would use to describe their own success.
On benchmarking competitors
Benchmark structure, not price. Their numbers tell you little, because you cannot see their costs, their discount floors, or how much of their book is on legacy terms. Their units tell you a great deal, because units are what buyers get trained on. When three vendors in your category move to a new unit, your prospects start arriving with a new mental model of what they are buying, and your pricing page begins to feel strange to them before anyone can explain why.
Changing the model without burning trust
Three things matter more than the announcement email.
Give notice with room in it. Atlassian notified Cloud customers of pricing and packaging changes about a month before they took effect, and Slack’s 2025 changes applied to renewals after a stated date. Tying changes to renewal dates rather than a single cutover is the gentler mechanism, because every customer meets the new model at the moment they were already reviewing the relationship.
Grandfather with an expiry, not forever. Permanent legacy plans are how you end up unable to read your own economics three years later. A defined transition window protects the relationship without permanently splitting your business into two companies with different unit economics.
Change one layer at a time when you can. Moving the price level and the pricing metric simultaneously means you will never know which one the market reacted to. If you must move both, move the metric first on new customers only, and watch win rate and profit dispersion before you touch the existing base.
Two pieces of advice worth ignoring
The first is the claim that a 1% price increase reliably produces some fixed jump in profit. That figure circulates with different numbers attached and always assumes volume holds steady. It is a thought experiment, not a law, and it has talked a lot of companies into raising prices on customers they should have repackaged instead.
The second is the idea that a specific churn rate, margin, or win rate proves pricing is broken. There is no credible universal threshold. A 7% annual churn rate means something very different for a $40-a-month self-serve tool than for a three-year enterprise contract. Thresholds imported from someone else’s business model will tell you to act when you should wait, and to relax when you should move.
What is not context-dependent is the direction of travel. McKinsey’s 2024 analysis of 29 of the largest global consumer companies found that input-price increases around 2022 squeezed margins, which is a tidy reminder that cost structures can move faster than pricing architecture. If your model is reviewed once a year and your costs move quarterly, the gap is not a risk. It is a certainty, and the only variable is how long it takes you to notice.
The question to carry into your next review
Pick your single largest customer. Ask two questions about the last twelve months. Did the value they got from you increase? Did your cost to serve them increase?
Now look at what changed on their invoice.
If the answer to both questions is yes and the invoice barely moved, your pricing metric is deaf in both ears. It is not responding to value and it is not responding to cost. At that point you do not have a pricing problem you can solve with a number. You have a model that has stopped describing your business, and every month you leave it in place, the gap between what you sell and what you charge for gets a little wider.
That gap is where revenue growth quietly stops being profitable growth, and it is the first thing worth checking before you conclude that the answer is more customers. If your cost to serve now scales with usage, it is also worth revisiting whether the model itself still scales, because unlimited plans built on a near-zero marginal cost assumption are the first thing to break when that assumption dies. Add profit dispersion to the questions you review each quarter. It is a cheaper habit than discovering the answer through a bad renewal season.
