SaaS Says: Not in the Token Business

Not in the Token Business

Ketan Karkhanis prices ThoughtSpot by the question, not by the token, and he explains why with two queries. "And that question results in, let's say, 100,000 tokens getting used. And if you ask another question that results in only 10 tokens getting used. I just charge you two. I don't charge you 100,000 plus 10."

The ThoughtSpot CEO's argument is about whose incentives the customer is buying. "Because the underlying business model of the model companies is to get you to use tokens." His product uses the language model to interpret the request and little else, which he says leaves him "controlled predictable costs for my LLM and I don't pass that on to my customer and it becomes my differentiator." A vendor describing his own product. The pitch still shows where the fight is.

Brad Scott, who runs worldwide sales at Glean, makes the same pitch. Glean routes across the major labs, which he frames as neutrality: "everyone we compete with has a perverse incentive to sell you a token, right? They run their business on selling you tokens, whereas we don't." The buyer's half of the argument: "because it turns out a token has a dollar value to it, and there isn't a money tree in every company's backyard that they can go out and prune when they need new intelligence."

Matt Hedberg has covered software at RBC for more than twenty years, and he gives the analyst's version. "really, the model is commoditized and when we think about the app layer, that could just be like a thin wrapper around data. So in some regards, switching costs could be lower with where we're going." Data is still a moat, but a conditional one. If nobody opens the interface, the incumbent's "toll bridge may not exist forever if it's my customer data, right?" He hears the harder question from customers: "Having a good AI product is like, that's going to save you from churning. What is it that's gonna actually cause me to pay you more money as a vendor."

The systems of record think they have the answer. "It turns out everyone can't vibe code their own salesforce and those systems of record are going to be pretty sticky," Scott says. Chris McDonough, VP of sales at the AI-native ERP Campfire, makes the case for his own category, where the numbers have to be right: "You can't go to the street with bad data. It's not allowed. And so like, no one's going to vibe code it. You can't." Autodesk CEO Andrew Anagnost makes it for vertical software. "Probably right is wrong in our business." And: "It's expensive to use a horizontal tool to solve vertical problems."

Nobody claims to have the price figured out. Dan Barzily, CFO of Tipalti, says his company was among the first to move from seat to platform pricing. Asked what AI does to it next: "customers will have to pay differently. How differently? I do not know yet"

Three weeks ago the models commoditized. Now the software built on them is pricing against the meter.

Sources: Interviews from Metrics that Measure Up (Sep 10, Ketan Karkhanis of ThoughtSpot), Revenue Builders (Sep 10, Brad Scott of Glean), Run the Numbers (Sep 7, Matt Hedberg of RBC), GTMnow (Sep 10, Chris McDonough of Campfire), Masters of Scale (Sep 10, Andrew Anagnost of Autodesk), and CFO Thought Leader (Sep 7, Dan Barzily of Tipalti).