The pricing is the tell

The core problem: Incumbent telco vendors say they’re AI companies now, but talk is cheap. Their pricing models reveal the truth. Services billed by the hour and per-token charges betray that vendors are selling effort instead of results, while keeping the compounding learning for themselves. AI makes outcome-based pricing possible for the first time: software executes the work, completes it in seconds, and allows value to be traced back to what created it. Incumbents cannot make the switch because their business model is consulting, i.e. billable human time. Danielle Rios of Totogi argues telco executives should read pricing structure to determine the value of a vendor’s products, and choose vendors whose growth requires the operator’s improvement.

On July 1, Alex Karp went on CNBC and asked an important question: if a frontier large language model (LLM) is really worth what the labs claim, why is it sold by the token instead of by the result? If I could make you a billion dollars, wouldn’t I ask for a share of the billion rather than a fee for the compute?

Now turn that question on the vendor you’re already paying: if their AI is as transformative as the roadmap deck claims, why isn’t it priced by the outcome? Why is it still quoted to you by the hour?

Every time you pay by the token, the license seat, the change request or the hour, your value is quietly leaving the building. It’s the same old pricing for the same old trap. At Totogi, where I’m CEO, we price the opposite way: the more you save, the more we make. When you win, we win. Our growth requires your improvement, not your complexity. So, why doesn’t everyone do it this way?

Pricing reveals the truth

A roadmap can promise anything. But a price is a bet, and vendors only bet on what they believe they can deliver. Marketing tells you what a product wants to be; the pricing page tells you what the vendor knows it is. Nobody prices against a result they can’t reliably produce, and nobody uses a fee model that leaves money on the table. So the way a vendor charges you confesses two things at once.

The first is a claim about capability: can this vendor connect what they sell to an outcome you’d pay for, or can they only bill you for effort and access? The second is a claim about allegiance: does the vendor’s revenue grow when your problem shrinks, or if it persists? You can read both off the price sheet alone. The wrong question to ask is “what does this cost per year?” The right one is “what does this pricing structure confess about where my value ends up?”

Three ways to price AI

Look at the AI contracts on your desk. There are three ways a vendor prices you, and each one is a tell.

1. By the effort (the man-day, the change request, tokens used): You pay for hours worked, changes made, and tokens burned whether or not any of it moved a metric. This is the professional services model that dominates telco, and it is the same confession Karp was pointing at in the labs: the vendor charges for effort because effort is all they can guarantee. But there’s a second admission buried here, too. Every change request you fund also deepens the vendor’s knowledge of how your business runs, knowledge that stays on their side of the table. You pay for the work; they keep the learning (and it’s hard to get it back from them; see “Own your learning loop”).
2. By the license and the implementation: You buy the software, then you buy the small army required to make it do anything. The confession is that the product does not deliver value on its own; the value lives in the professional services wrapper, which is precisely why the wrapper is where the money is. Outcomes, in this model, are quietly your problem.
3. By the outcome: The vendor gets paid when the value actually lands. Vendors who price this way are telling you they believe their system reliably produces results, and, just as important, that they can prove which result they caused. That is a far harder thing to say out loud, which is exactly why so few say it.

Line the three up and the diagnostic writes itself. The first two shift all the risk and all the learning onto you. Only the third puts the vendors’ money where their claims are.

Outcome pricing is all the rage

For decades, outcome pricing was a fantasy in enterprise software. Not because vendors were greedy, but because the delivery mechanism made it impossible. When the work is performed by people, effort is the product. A vendor whose costs are salaries cannot put its fee at risk on your results. It has to bill the hours whether or not the project lands. Human delivery also makes outcomes untraceable. When a retention campaign takes five teams and three weeks, nobody can honestly say which meeting produced which result. And the result arrives so slowly that by the time anyone could bill against it, the people who did the work have been reassigned twice.

AI changes every piece of that at once. The work is executed by software, so the marginal cost of an action is close to zero, which means a vendor can afford to get paid only when value lands. Delivering doesn’t burn payroll. The work completes in seconds rather than quarters, so the outcome shows up inside a billing cycle instead of outside a fiscal year. And the work leaves a complete digital trace, with every action logged and connected to what triggered it. So, for the first time, value can be traced back to specific actions, instead of just guessed at.

Another way to think about it: the unit of delivery is shifting from the hour to the action. Hours get billed if someone works them. Actions can be priced based on how well they work.

What if you paid for value?

At Totogi, we price on outcomes. Our product is not billable time; it is the Totogi Ontology—software that executes against canonical telco semantics and gets smarter with every action. And because the AI takes the action, and we can trace it, we can attribute the value, so the fee can go exactly where the value lands.

Here’s what that looks like in practice:

1. Network operations automation: The fee is a share of the engineer-hours you no longer spend. Incident resolution that took four hours takes 48 minutes; 300 network engineers stop writing reports and start fixing networks. You pay against the reduction in your operations bill, not our headcount.
2. Legacy mainframe decommissioning: The fee is tied to the legacy systems you switch off. We wrap an orchestration layer around decades-old BSS systems and offload customer journeys one by one, so you can wind down the mainframe footprint without risking current revenue. We grow when your stack shrinks.
3. Dormant cell recovery: The fee rides on the revenue you were already losing. Cells that sat dark for months—invisible because no single system had the full picture—get identified, diagnosed, and reactivated. You don’t pay for the discovery; you pay when the network earns again.
4. Enterprise sales acceleration: The fee is tied to cycle time. A rep walks out of a meeting and tells the AI what to sell to the customer; the agent builds and sends the quote before the rep reaches their car. Quotes that used to take weeks of cross-system coordination now close in days, and we get paid according to how much customers save, not the effort it takes us.

Four different fees, three lines of your P&L—cost out, revenue up, growth pulled forward—and one common structure: the fee moves when the outcome does.

And when an outcome doesn’t land? We don’t get paid, or we get paid a lot less. That’s the deal. We can make that offer for the same reason the incumbents can’t: our delivery costs are compute, not payroll, so a miss doesn’t cost us thousands of salaries. It costs us the motivation to make the ontology better.

And yes, outcome pricing creates a new negotiation. Some customers argue the measured outcome down to shrink our fee—and honestly, we take it as a compliment. Nobody haggles over the attribution of value that didn’t arrive. When the argument moves from “did the hours get worked” to “how do we split this improvement,” we know the value exists, we’ve measured it, and it’s on the table. That is a far better fight to be in than auditing timesheets.

Why the incumbents can’t match it

Back to the question from the top: why doesn’t everyone price this way? Because of what they are. Strip away the AI polish and most incumbents aren’t software companies; they’re consulting companies. Amdocs employs roughly 27,000 people, and the majority of its revenue comes from managed services: tens of thousands of humans running billing operations, processing change requests, and remediating fallout by hand. A company whose product is human billing time cannot price by the outcome, because every outcome delivered means fewer billable hours. Its revenue grows by adding people to your account, and shrinks when your problems actually get solved. I’ve written about this before—see “Amdocs’ AI reckoning.”

You could watch the identity express itself when Amdocs launched aOS, its new agentic platform, this past February. Its AI foundation, Cognitive Core—the amAIz suite, rebranded—lists a “flexible consumption-based model” as a headline feature, and Amdocs’ agent announcement touts “optimizing token utilization.” Handed the technology that finally makes outcome pricing possible, the company priced it the way it has always priced people: by the unit of effort consumed. The token is the new billable hour. They are paid whether the AI works or not. Who wants that world?

Read the price sheet

So here is the exercise, and you can try it this week. Pull your three largest AI and BSS contracts and read only the pricing pages. Count the lines billed by the token, the license seat, the change request, or the hour—by now you know exactly what each one confesses. Then ask every vendor on your shortlist one question and watch their faces: does your fee shrink when the outcome doesn’t land? The ones who flinch just answered you.

Karp is right that the enterprises are frustrated. He just left out the fix. It isn’t a better token deal or a tougher data clause. It’s refusing to sign for effort—and instead paying only for results. And only one kind of company can sell you results: the kind whose software can take the action, trace it, and prove the value landed.

Recent Posts

Get my FREE insider newsletter, delivered every two weeks, with curated content to help telco execs across the globe move to the public cloud.