SW: Pricing AI When the Costs Are Real: An Interview with Asana’s Sid Sanghvi
In a single release this September, Asana went from selling one product to multiple products, introduced new pricing bundles, and began selling agents on consumption.
I recently sat down with Sid Sanghvi, Head of Finance Business Applications at Asana, to talk about pricing AI when every move carries a financial consequence: new costs, new revenue streams, new ways for a customer to be surprised by a bill. Plus, the economics can change every few months.
The Launch
Tien: Hey, Sid! Asana went from one product to multiple in a single release. What changed?
Sid: Almost everything. Beyond our core work management product, we now have Agentic Work Management, which adds AI workflows and agents to the core. We launched Asana Service Management for IT, HR, Facilities and Legal, Command by Asana for engineering teams, and Asana Client Management for agencies and professional services.
Each is a bundle of multiple products, and inside those bundles are agentic products sold on consumption. So in one release we went multi-product, introduced bundles, and added usage-based pricing.
Technically everything worked seamlessly, even where we combined consumption with traditional SaaS pricing. We did a lot of co-development with Zuora, and I believe that many of our hairier requirements have since become standard features. For the first time, the technology was the easy part and the process was the hard part.
Why AI Breaks the SaaS Cost Model
Tien: Why is pricing AI different from pricing software?
Sid: In SaaS, the cost of building the product is largely sunk, so margins stay high. With AI, whether you use frontier or open-weight models, a real cost hits every time the product is used. Customers, meanwhile, are excited about AI but skeptical about its value and under pressure not to overspend. So we had to price for all three variables.
Tien: Where did you start?
Sid: With three principles. First, every Asana user should be able to use AI, with no paywall before they understand what it does. Second, build purposeful products that create outcomes. Our moat isn’t the agent. It’s our Work Graph. Third, embed AI in the core offering with a set number of included credits, so customers see the value in the real product, not a trial.
Tien: Free credits are a real cost when every request hits a model. How does that fit the business model?
Sid: SaaS is still at our core. We charge by the seat, and AI usage up to a certain level is included in that seat. Consumption only kicks in when a customer buys an expansion pack or turns on on-demand billing. That matters to CFOs and CIOs. If they buy 50 seats, they know exactly what they’re paying, and they move into variable spend only when they choose to.
Choosing the Unit
Tien: When Asana customers do move into consumption, what are they paying for?
Sid: AI Teammates, our autonomous agents, are priced per request, and we define exactly what a request is and when it counts. AI Studio, our no-code automation builder, uses credits at a set unit price. Before running a workflow, customers can simulate it and see how many credits each run will consume. If they plan to run it 50 times a day, they know the cost up front.
Tien: Lots of software companies facing new token bills are tempted to go cost-plus: pass the token charge through with a margin. Did you consider it?
Sid: Well, we tried it. Our first version of AI Studio was priced on tokens, and it led to some confusion One of our biggest personas is the marketer whose workflow would report 55,000 tokens, and the questions started. Why that many? What went into it? How is it calculated?
So we switched to credits. Say for the sake of simplicity each credit is worth a penny. This workflow uses 50 credits, so it costs 50 cents. Customers get that immediately.
Tien: It also puts token efficiency on your side of the table. Getting smarter about token costs is your job, not the customer’s.
Sid: Exactly, and we pass the gains through. For AI Studio, if we move to a cheaper model or shift work to open-weight models, we recalibrate on the back end. A workflow that used seven credits now uses four. No renegotiation. The customer’s workflows simply get cheaper.
Overages and Buyer’s Remorse
Tien: Overages are where usage pricing usually goes wrong. How did you handle them?
Sid: Nobody wants an overage bill at the end of the month. So customers control on-demand from the admin console. They can turn it off, or turn it on with a cap, say 10,000 credits, after which usage stops. A real-time dashboard shows every credit consumed, who used it, and how often, so they can see where the trend is heading.
We also sell consumption packs, and we proactively recommend the one that’s economically in the customer’s favor. Overselling is its own trap. A customer who pays for a big pack and uses a fraction of it gets buyer’s remorse. So we recommend what they need and let them stack expansion packs as they grow.
Changing Prices Without Losing trust
Tien: You’ve told customers your prices will probably keep changing. Most customers hear that as risk. But it sounds like it in some cases, their prices might actually drop! How do you maintain transparency?
Sid: Every time a new core AI model comes out, the previous one drops in price. Knowing that, we kept our pricing flexible and told customers plainly: if we can deliver the same quality at better economics, we’ll look at lowering the price. That drop is built in, through fewer credits per workflow.
We apply the same logic as a buyer. We’re not signing long-term contracts for AI tools ourselves, because prices fall and the technology moves fast. Our customers are making the same calculation about us. So it comes down to trust: real value, transparent pricing, and customers in control of the commercials.
Tien: What are you weighing when you consider a pricing change?
Sid: Whether the value we create and the price we charge are in balance. If that breaks in either direction, you get churn. What we’ve found is that once customers see AI changing their daily work, pricing barely comes up. Our early adopters have never told us the pricing is out of whack. They ask us to add even more features to the product.
Pricing at the speed of AI
Tien: Experimentation only works if your systems keep up. What does that look like?
Sid: We had an internal mandate not to be afraid of experimenting with pricing. It used to be fine to take eight or ten weeks to change pricing and packaging. Now I get requests to do it in two or three days. Recently I got a Friday request to go live with new pricing on Monday. The foundation we built with Zuora is what makes that possible.
Tien: What do you think is holding customers back from adopting AI?
Sid: Friction. No customer knows their consumption patterns yet, because they’re experimenting too. Boards are mandating AI, so some companies adopt it under pressure and some out of curiosity. Every friction point makes them more hesitant. Remove those, and the only question left is whether the product is good enough to use.
Tien: Any lesson for other companies about to launch new AI services?
Sid: Your unit of measure matters more than you think. Keep it simple and define it clearly. And don’t mistake exposing raw costs for transparency. Real transparency is a unit customers understand, a price they can predict, and the controls to stay within it. Time will tell, but that’s the bet.
Tien: Thanks so much, Sid!
Sid: Of course.
