Price the Work, Not the Software

How should we price this? It's a common question from founders in the AI era, and one I get a lot. Here's the answer I send.

Charge for the outcome. Not seats, not tokens.

Align incentives

Always align incentives. It's the oldest rule in business; Munger said it best: "Show me the incentive and I'll show you the outcome." Price on outcomes and you'll build the product that produces them. Every improvement you ship lands in your own revenue. Growing an account means making the product better, not selling harder, and the whole company ends up optimizing the same metric your customer does.

Advertising learned this decades ago. Advertisers never wanted impressions; they wanted customers. CPM gave way to CPC and then to pay-per-install because each step tied the platform's revenue to the advertiser's result, and nobody misses paying for impressions.

AI pricing is at the same fork. Sierra charges a negotiated rate when its agent resolves a support case; escalation to a human is free. Intercom charges 99 cents per conversation its agent resolves.

Pricing is also a decision about who holds the risk. Outcome pricing puts it on you. Good. You control whether the product works.

Don't price on usage

"Our costs scale with usage, so our price should too." Tempting, and wrong. Tokens don't measure value any more than lines of code measured productivity. Worse: a usage-priced product earns more by being inefficient. Your incentive points backwards. Price the outcome and efficiency becomes your margin.

The test: does your AI assist the work or do the work? If it assists, seats can still make sense. If it does the work, charge for the work.

The measurability test

One requirement: the outcome has to be measurable. Where does yours sit?

• Binary and attributable (ticket resolved, meeting booked): post a price.
• Measurable but negotiable (leads qualified, revenue influenced): negotiate the outcome per customer, or keep a base price and add an outcome bonus when a measurable result lands.
• Not yet measurable: charge hybrid for now and build the product toward measurability.

About 7% of AI products priced outcomes at last count. Early is the point.

Incumbents can't follow you

Incumbents have too much inertia to price this way. They built their entire organizations around the old model: sales comp paid on bookings, revenue recognized on subscriptions, forecasts and board expectations wired to committed ARR. Changing the pricing model means rewiring the whole company, and most will never try.

The rare exception shows why. Intercom made the switch, and it cost them: sales comp rebuilt around logo counts, committed revenue traded for consumption revenue, tens of millions in legacy revenue knowingly destroyed. Their CFO calls his own comp plan "heart-stopping." That's the bill for changing pricing inside an organization built on the old model, and it's why so few will pay it.

You don't have that bill. Start with the model they can't get to. Outcome pricing isn't just better aligned; it's a moat with inertia on your side.

Won't the work commoditize?

The best objection, argued well by David Peterson: as models improve, the work becomes a commodity and prices deflate. He's right about the deflation. HubSpot's customer agent just went from $1 per conversation to 50 cents per resolved conversation.

But deflation hits every pricing model. The outcome-priced vendor keeps the offsets: efficiency gains land on your margin, rising resolution rates turn unbillable conversations into billable ones (Intercom went from 25% to nearly 70%), and contracts wired into how a customer measures their business are hard to rip out. The usage-priced vendor faces the same deflation pinned to a collapsing input cost. Buyers are already demanding outcome terms. Your only choice is whether you set them. And if you don't, your competitor will.

The playbook

1. Name your customer's outcome function: the metric they'd pay for if they could. If you can't name it, that's a product problem, not a pricing problem.
2. Place it on the measurability spectrum. Pick the matching model.
3. Price against the human cost of the work, not software comps. A support ticket runs $5-20 with humans; an SDR is $70-90k. That's your ceiling and your ROI story in one number.
4. Make efficiency your problem, proudly.
5. One public price for volume. Negotiated outcomes for enterprise.
6. Move now. The window is open while the incumbents are stuck.

Price the work.