This series started with a question from Lenworth Henry at Google Cloud, who works with early-stage founders on the Founders Advocates / Startup Ecosystem team. He asked how founders can figure out how much to build, how to price it, and how to make sure the numbers actually work. This is post 2 of 3. If you missed the first one on how to actually know what to build, read it here.
Ask any early-stage founder how they priced their product, and you’ll usually get some version of the same answer.
They looked at what it cost them to deliver. They added a margin they thought was reasonable. Twenty percent. Fifty percent. Someone once told them software should have eighty percent gross margins, so they added eighty percent. They called it a price and moved on.
That’s the wrong instinct, and it’s a particularly expensive one in the AI moment because your cost structure is volatile enough that a cost-plus model can wipe out your margin without warning. But the deeper problem is that cost-plus pricing tells you almost nothing about what your product is actually worth to the person buying it.
The right question isn’t “what did this cost me to make.” It’s “what is this worth to the person receiving it.”
Why value-based pricing is a magic trick
The reason value-based pricing works, and the reason the best tech companies rely on it almost exclusively, is that it thrives on a gap that most other business models can’t exploit.
What something costs you to produce and what the customer receives from having it are usually two different numbers. Often wildly different.
If a customer pays you $50,000 a year for software that saves them a full-time employee, the value they’re capturing is not $50,000 minus your delivery cost. It’s the fully-loaded cost of that employee (probably $150,000+ once you factor benefits and overhead), plus whatever secondary benefits come from the work being done faster or more consistently. Your delivery cost might be $2,000 a year. The value they’re capturing might be $200,000 a year. Your margin between cost and value is not eighty percent. It’s ninety-nine percent.
That gap exists in almost every well-priced software product. It’s not exploitation. It’s how the value equation actually works when you’re delivering something the customer couldn’t produce themselves.
Which brings us to the harder question. When can you actually capture value that way?
Value-based pricing only works if you’re actually differentiated
Here’s the part most founders get wrong. They hear “value-based pricing” and think it just means “charge more.” So they raise their prices, watch conversion collapse, and conclude that value-based pricing doesn’t work in their market.
Value-based pricing depends on a specific condition. You have to be delivering something the customer could not get without you. Not something they could get slower without you. Not something they could get more expensively without you. Something they genuinely could not get at all.
The trap most tech founders fall into is thinking that being cheaper or faster than the alternative is a defensible position. It isn’t. Cheaper and faster are commodity dimensions. Someone else will be cheaper and faster next year. And even if they aren’t, being cheaper and faster doesn’t let you charge premium prices, because the customer can always fall back to the slower and more expensive option.
The number one thing you hear in AI companies right now is some version of “we can do X in 30 seconds instead of a week.” That’s cool. It’s also not really a business. It’s an efficiency improvement. Efficiency improvements get priced against the labor cost of the thing they replace, which puts a hard ceiling on your pricing power.
The real question, once you have the technology in hand, is what becomes possible that wasn’t before. Not what becomes cheaper or faster. What becomes possible.
Example from a Probate Startup
An AI company we’ve been watching built a system to handle all the paperwork required to submit an estate into probate court. On the surface, that sounds like a straightforward efficiency play. Do the forms faster, charge less than an attorney, take the market.
What actually happened is different, and it illustrates the value-based pricing point better than any framework could.
Because the software handles the paperwork automatically, the company is able to work with attorneys who take the cases at a discounted rate. The attorneys make more money because they can process more cases with less work. The customer gets probate resolved faster and for less money. That would be the expected value.
The unexpected value, and the thing that turned out to be worth the most, is what the AI enables that no attorney had ever offered. The system will actually listen to the customer. The customer can talk to it about what they’re going through as they work through probate. It listens. It gathers the information it needs from the conversation. It processes the grief alongside the paperwork.
That’s the thing no human attorney can offer because no human attorney has the time to sit and listen to a grieving client at that depth. It’s not a feature. It’s an unlock. It’s something the customer couldn’t have gotten anywhere else at any price.
And that’s what they’re actually willing to pay premium prices for.
The lesson isn’t that you should build AI grief counselors. The lesson is that the sharpest pricing power comes from finding the thing your technology makes possible that nobody has ever been able to offer before. That’s the thing worth building the business around. Everything else is a feature.
The one-of-one asset
The ideal end state of this thinking, and the thing tech companies specifically should be looking for, is what you could call a universally unique asset. One of one. Something nobody else can do because nobody else has the specific combination of technology, data, insight, or context that you have.
Most businesses will never reach that state, and that’s fine. But even short of it, the mental exercise is worth doing. What is the thing about our approach that literally could not be replicated? Not “would be hard to replicate.” The thing that could not be replicated at all because it requires an insight or an asset that only exists inside our company.
If you can find that thing, everything else in your pricing conversation gets easier. You’re no longer negotiating against a competitor who’s offering the same thing cheaper. You’re negotiating against the version of the world where the customer doesn’t have your solution at all, which is a completely different math.
One more thing, for AI-native companies specifically
Once you’ve solved the pricing side, the cost side becomes worth engineering hard. This is where AI-native companies leave a lot of money on the table.
Most of the AI products we’ve seen use significantly more tokens than they need to. Sometimes because the founder threw the largest available model at every problem. Sometimes because they never went back to optimize once the first version worked. Sometimes because they built with an AI where a deterministic solution would have worked just as well.
You can typically cut your token costs by 80 to 90 percent after your first working version, without any degradation in quality. Model selection alone often accounts for half of that. Better prompt engineering and harnessing gets you most of the rest. And genuinely asking “does this need to be AI at all” often reveals that half of your pipeline should just be code.
None of that matters if your pricing is broken. If you’re charging cost-plus on inflated costs, you have a bad business either way. But if you’ve done the value-based pricing work and you have real margin available, then the cost engineering compounds. Your margin goes from ninety percent to ninety-eight percent. That difference funds everything else you want to do.
The pricing model is the strategy. It tells you what kind of company you’re building.
Cheaper and faster are fine, as features. They are not a great business. The companies really worth building are the ones that unlock something that wasn’t possible before. Those are the ones that can charge what the value is actually worth because the customer has no other way to get it.
Everything else is an efficiency improvement, priced against labor costs, and headed for a race to the bottom.