Pega founder Alan Trefler thinks big AI is selling fear

Pega CEO Alan Trefler founded his company back in the 1980s when the enterprise world was very different than it is today, but he's been around long enough and seen enough shifts to speak with an authority that few others can when it comes to enterprise software.

His company is well positioned for the AI era because its workflow software sits at the center of many business processes. But that doesn't mean that Trefler isn't highly critical of how the big AI labs and the largest cloud companies are positioning AI for enterprise buyers. He sees fear mongering, greed and maybe even a little emotional manipulation, and he's not afraid to call them out.

"They're both trying to scare them into thinking it's all powerful, and also scare them into thinking that if they don't use it, they're going to fall behind," Trefler told FastForward. He refers to the big AI labs as being like "drug dealers" giving away tokens for free until users are hooked, then charging big bucks.

The high cost of fear

Token-based pricing has become a growing point of friction for enterprise customers, who want to know what they're paying for before the bill arrives. Trefler has been highly critical of the token-based pricing model, which he believes is deliberately designed to be obscure. "A token is three quarters of a word. It's mathematical. They could have charged by the word, and everybody would have understood what they were paying for," he said. Instead, customers often don't know what their bill will be until it arrives, and even then the costs can vary widely depending on usage.

As he sees it, the AI labs have "declared war on enterprise software" via an "apocalypse narrative." Companies are being told they need to rebuild everything around AI or risk being left behind.

Trefler believes that message serves another purpose as well by inflating the AI companies' addressable market to include all enterprise software to help justify their huge valuations. 

Of course, Trefler has a dog in this fight too. But Pega has taken a different approach to AI usage and pricing, designing its software so it only calls on AI when it's needed, and only in very specific ways to help control costs.

Using AI as a helper

Instead of using AI as the center of everything, an approach that leads to high costs for the customer and the vendor, Pega is trying to take advantage of its strengths as a workflow tool. They define every step of a larger process up front, and only use AI when called for, which he believes will lead to lower costs.

For example, if the workflow hits a step where it needs to read an email or a document to pull some information, the large language model will kick in to take care of that task, but only that task. "By using the language model at runtime for very specific tasks that are under control of a workflow we get a predictability you would just not get if you said to the language model process this email [or some other more general instruction]," he said.

And that specificity is what he says leads to less drift and tighter cost containment. It also lets the company charge for outcomes with confidence that costs won't spiral out of its control even when invoking large language models. "We get the reduction in cost because the amount that we're using the language model is very, very targeted because we've done the heavy lifting when we created the [workflow] recipe at the design time," he said.

From seats to outcomes

He pointed at pricing as the biggest issue when it comes to AI. Like all software companies in the 1980s when he started the company, he sold seat-based licenses, an approach his company took for the next 30 years, but about a decade ago he found that approach flawed. He was selling workflows and they behaved differently than other software. 

"About 10 or 11 years ago, we realized that seat-based licensing was a really stupid way to think about licensing your software," Trefler told FastForward. He observed that his software was based on a unit of work they called a case. That could involve an insurance claim or loan application, but there was a beginning, a middle and an end, and it made more sense to him to start charging based on the outcome. Did they file the claim successfully? Did they complete the loan application?

"We came up with another pricing metric based on the work that the system does, and so Pega is a system that is at its heart a process automation system, and typically every one of these processes are a couple of related processes that operate in something we call a case," he said.

That decision to price based on outcome is something that many companies are suddenly talking about today as a way to price agent usage more fairly, putting Pega well ahead of its time with this approach.