when AI changes your costs, i fix your pricing.

Every software CFO right now has two priorities that directly contradict each other. Cut software costs. Increase AI usage. Fair use is how you hold both.

As companies scale, LLM usage scales with them. Even after negotiating unit costs down, you're looking at 40–50% margin at best. So how do you get back to 80%+ margins?

The reality is 5–10% of users are driving 90% of your costs. A price increase for 100% of your base when 90% aren't the problem is a tough sell. So design a lever that only the 10% will feel.

The math

You have 1,000 customers on a $50/month plan. Your AI feature costs $2/customer/month at median usage — manageable. But the P95 customer is using $25 in compute. The P99? $80. That top 1% — just 10 accounts — might cost you more than the bottom 500 combined.

You have two options: raise prices across all 1,000 customers to cover the tail, or set a fair use threshold at the 95th percentile and only trigger a conversation for the 50 customers above it. Option one risks churn from 950 customers who were never the problem. Option two protects your margin without touching anyone using the product normally.

I've used fair use to cut product COGS by 33% in six months. The framework: Inspect your usage distribution, Analyze the P95 threshold, Apply it as a hard or soft limit depending on your product type.

You've rebuilt the nurture sequence twice. You've tested new creative, swapped channels, thrown a webinar at it. Conversion rate hasn't moved. Here's the thing nobody tells demand gen leaders: pricing problems disguise themselves as marketing problems.

The 6 symptoms

1/ Pricing page bounces qualified traffic. That's not a design problem — it's a packaging architecture problem.

2/ Free-to-paid conversion is flat. The gap between free and paid is a cliff, not a step.

3/ "Talk to sales" gets clicks but deals don't close. They clicked because self-serve didn't make sense.

4/ Discounting closes every deal. When value isn't obvious from the tier structure, the only lever left is "make it cheaper."

5/ MQLs are up but pipeline stays soft. Your pricing might be attracting the wrong segment entirely.

6/ Self-serve isn't scaling. If buyers need a calculator to figure out what they'll pay, you've broken the loop.

How much is too much to charge? Wrong question. How you charge is more important than what you charge — price points are easy to adjust, but unit choice requires an internal and external overhaul once it's set.

The 3 WTP question types

Alternative: If our solution didn't exist, how would you solve this? This is the baseline to position against.

Relative: Rank our value as a % of a tool you already pay for. That percentage × their spend = your anchor price.

Value-based: Quantify outcomes (revenue up, cost down, risk reduced), then apply a value-to-price ratio to find the ceiling.

The best companies re-evaluate pricing every quarter. The question isn't whether your model is right — it's whether you're asking often enough.

Every concession feels like it costs you once. It doesn't. Give something because a buyer asked for it, and you've trained them to ask again — then set up a renewal you're structurally built to lose.

Discounting isn't the problem. Discounting without a system is. The rule most CFOs run on: keep total concession under 10–15% of deal value. Stay under it and the discount is a rounding error. Go past it and you're not defending a discount anymore — you're defending your judgment.

The math nobody models

A concession is never just price — the roadmap feature, the liability cap, net-90 terms, the seats you "threw in." Add them up and a "10% discount" is often a 25% concession. And at renewal they expect it again. On a three-year deal, one concession is $450K.

The fix: give to get — always. No concession leaves the table without something coming back.

A price increase isn't a blanket bump everyone resents. Done right, it's value capture — concentrated where you've earned it. I once raised prices 5–20x across a book of contracts and lost under 5% of it.

Most teams treat a raise like a thermostat: everyone's a degree too low, so nudge the whole building up. That's the version that gets you churn — you've charged 100% of customers to fix a problem 20% created, and the 80% who were never underpriced flinch first.

The move

Rank accounts by the dollar value your product drives for each. It's never flat — the top 20 pull 10x the value of the bottom 80. So don't touch the 80. Raise the 20, from $50K toward $150–250K, concentrating the increase where switching cost is highest.

I ran this on a synthetic monitoring product: 25 contracts renegotiated, prices up 5–20x, churn under 5% — roughly $6M in two quarters, including saving the largest customer everyone assumed we'd lose.

Your average customer costs $25 a month to serve. Your median customer costs 25 cents. Real numbers from an AI product I launched pricing for — same plan, same price, a hundred times apart underneath.

For twenty years you could underwrite a deal off the org chart. In AI, cost tracks behavior, and behavior has nothing to do with the size of the company writing the check. Roughly 20% of your users drive 80% of your cost, and you can't tell which 20% from the contract.

Why the old model breaks

Every generous cap you wrote was safe because a person gets tired. Call it the attention-span subsidy — for a decade your margin was funded by human fatigue. Point an agent at the same endpoint and every soft limit gets found and flattened by Tuesday.

Then price off your cost per call and you've promised, in the structure itself, to hand every efficiency your engineers ship straight to the customer. Cost tells you when to walk away from a deal. It has never told you what to charge.

What worked

At Grafana Labs I shipped a hybrid seat model into exactly those conditions — no competitive anchor, self-serve usage nothing like contracted. $1M ARR in 30 days, 15.2% free-to-paid against a ~9% industry average, 12,000 orgs in month one. Early on, don't overindex on margin. Set a logo target, close them, make it repeatable.