AI Is About to Have to Prove It Works

For most of the software era, companies have paid for access.

A monthly subscription. A per-seat fee. A certain number of credits. More recently, with AI, the unit became tokens, compute, API calls and usage.

But AI is beginning to create a problem for that model.

If software helps one of your employees work faster, charging for access makes sense.

If software starts doing the employee’s work, the customer eventually asks a different question:

What did it actually accomplish?

That question may end up changing the economics of the AI industry.

OpenAI is already experimenting with it

OpenAI CFO Sarah Friar said this week that the company is experimenting with pricing based on business outcomes rather than simply usage, as enterprise customers increasingly demand measurable returns from their AI investments.

That detail appeared alongside another telling number: OpenAI’s enterprise revenue reportedly increased 32% from June to July, faster than the company’s overall annualized revenue growth.

The biggest AI companies are moving deeper into businesses. And the deeper they move, the harder it becomes to justify charging merely for access to intelligence.

The software industry has mostly avoided taking this risk

Traditional software companies have had a beautiful arrangement.

They sell you the tool. You take the risk.

If you pay $300 a month for software and it produces nothing, the software company still gets $300.

If your team barely uses it, the software company gets paid.

If it saves you $100,000, the software company still gets roughly the same subscription fee.

AI makes that arrangement harder to defend because its value proposition is increasingly tied to work rather than access.

Imagine an AI sales agent that books qualified meetings.

Why should it necessarily cost $199 a month?

It might eventually cost $75 per qualified meeting.

An AI collections system could receive a percentage of money recovered.

An AI recruiting system could charge per qualified candidate delivered.

An administrative agent could charge according to completed workflows.

An AI marketing system could be compensated partly according to customers or revenue created.

Once that happens, AI begins looking less like software and more like labor, an agency or a business partner.

That changes who captures the value

Suppose an AI system costs your company $5,000 a year and saves $100,000.

Under conventional software pricing, you capture almost all of that economic surplus.

The vendor eventually notices.

If its software really created $100,000 of measurable value, why is it charging only $5,000?

Outcome pricing is one answer.

The AI company can charge substantially more while still leaving the customer with an attractive return.

That’s potentially very good for the strongest AI businesses.

It is considerably less comfortable for the weaker ones.

Outcome pricing removes somewhere to hide

AI companies have become extraordinarily good at describing activity.

Tokens processed. Workflows completed. Hours saved. Messages generated. Documents analyzed. Agents deployed.

Those numbers can sound impressive while telling an owner surprisingly little about what changed in the business.

Outcome-based pricing forces a more uncomfortable question.

Did it make money, save money or materially improve something the customer values?

If the answer is yes, charging against the result can create an excellent business.

If the answer is difficult to determine, the product has a problem.

This may eventually become one of the mechanisms that separates genuinely useful AI from the enormous amount of AI software being sold primarily on novelty.

The buyer changes too

There is another side to this.

Business owners have spent the last few years being told that they need AI.

That framing encourages tool collection.

A writing tool. An agent. A chatbot. A meeting assistant. An automation platform. An analytics system. Another subscription.

The more useful question is much simpler:

If this works, what number in my business should change?

Revenue?

Cost per acquisition?

Appointments booked?

Hours of labor?

Retention?

Response time?

Errors?

Something measurable should move.

If neither the buyer nor the vendor can identify that number, calculating the return becomes almost impossible.

The companies with the clearest results may become the most valuable

This creates an interesting inversion.

For years, AI companies have competed over intelligence: bigger models, better benchmarks, more context, faster responses.

Business buyers may increasingly care about something more mundane.

Can you reliably produce the result?

That favors companies operating close to measurable economic activity.

A system that can prove it generated $1 million in recovered invoices has an easier pricing argument than one claiming it made a team 17% more productive.

A system that can prove it booked 400 qualified appointments has an easier pricing argument than one reporting the number of emails it generated.

The closer AI gets to the cash register, the easier its value becomes to measure.

And the easier its value is to measure, the easier it becomes to charge for the result.

What owners should do now

You don’t need to wait for AI companies to change their pricing models.

You can start evaluating them as though they already had.

Before adding another AI tool, decide what outcome would justify its existence.

Give the tool a number.

Then watch the number.

If you’re paying for an AI sales tool, measure qualified opportunities created.

If you’re automating support, measure resolution time and staff hours.

If you’re using AI for marketing, measure customers and revenue rather than the amount of content produced.

If you’re using an agent to replace administrative work, calculate how much work actually disappeared.

The AI industry has spent several years asking businesses to believe in what the technology might eventually do.

The next phase may be much less philosophical.

Show me the result.

That could turn out to be one of the healthiest things that happens to AI.