This guest post is from Amy Konary, the Founder and Chair of the Subscribed Institute at Zuora.
Here’s a question for every finance and product leader in B2B SaaS: Do you know what it actually costs to serve your heaviest AI user? Not the average. The specific customer. The one whose agentic workflow fires 200 API calls every time they click a button.
If the answer is that you’re still working on it, you’re in good company. You’re also in danger.
In 1973, fuel was a stable, negligible cost for American trucking companies. Then the Arab Oil Embargo hit, diesel prices tripled overnight, and carriers with fixed-rate contracts went bankrupt. They were legally bound to deliver goods at prices that now cost them more in fuel than the total contract value.
The survivors invented something specific: the fuel surcharge. The industry created a standardized index that isolated fuel as a separate, floating line item. The base rate covered the service, labor, equipment, and insurance. The surcharge covered the volatile input. It adjusted automatically, weekly, tied to a published benchmark.
That mechanism is still the financial backbone of the global supply chain. FedEx, UPS, Maersk—they all use it. It taught an entire industry a powerful principle: decouple the service from volatile inputs. Let the input float. Keep the core contract stable.
Knight-Swift vs. Yellow
The surcharge solved the pricing problem, but it created a new one: the information lag.
The trucking industry’s recent history makes the lesson concrete. The losers had no idea they were losing money on a specific customer until the quarterly report came out. Yellow Corporation’s 2023 bankruptcy was driven by an inability to see costs in real time.
The winners look very different. Knight-Swift, the largest truckload carrier in North America, treats cost data as a real-time product feature. Their AI-driven systems tell drivers which specific fuel station along a route offers the best price, continuously, per trip, per customer.
Motive (formerly KeepTruckin), which powers fleet operations for thousands of carriers, has built a real-time cost infrastructure to help combat information lag. They process billions of data points a day to automate decisions that used to take hours and arrive too late. Their Fuel Hub uses AI to flag wasteful idling and inefficient driving. Their Savings Finder analyzes fuel prices at nearby stations and helps drivers find the cheapest options. Their integrated Motive Card cross-references vehicle location with transaction data, automatically declining fuel purchases when the truck isn’t physically near the station where the card is being swiped.
SaaS is having its Trucking Moment
If seats were SaaS’s stable fuel, tokens are its post-embargo oil price. For two decades, SaaS economics assumed that the cost of serving one more customer was close to zero. AI inference has rewritten that. A heavy AI user can now consume 50% of their MRR in inference costs, while a light user consumes a fraction of a percent. A misconfigured agent can move your gross margin five points in a week.
Most SaaS companies today discover these margin shifts the same way Yellow discovered its fuel losses — long after the damage was done. That’s the information lag.
Closing the lag is what makes everything else possible. Knight-Swift routes trucks to cheaper fuel stops because it can see fuel prices in real time. SaaS companies can route simple queries to smaller, cheaper models — but only if they can see query costs in real time. Knight-Swift uses index-linked surcharges that adjust automatically. SaaS companies need usage-based credits or token surcharges — but a surcharge built on stale data is just a guess with a formula around it.
And here’s what the information lag hides. Token bloat from inefficient prompts acts as a hidden discount to your LLM provider. Agentic loops where a single click triggers hundreds of downstream calls can turn a profitable customer into a loss in minutes. These margin killers only become visible when you close the lag — when you instrument token costs at the customer and feature level, in real time, the way Motive instruments fuel costs per mile.
Close the lag
Trucking learned this fifty years ago: when your most important input cost becomes volatile, you either build the surcharge and the visibility to operate it in real time, or you discover the damage in the quarterly report.
The companies that will define B2B SaaS in the AI era are the ones closing the lag today. Tagging token costs at the customer and feature level. Reviewing unit economics weekly. Routing inference to the cheapest adequate model. Feeding what they learn into pricing structures that capture value while preserving margin.
Knight-Swift closed the lag. Yellow didn’t.
Which one are you?
For more on revenue architecture design, see the Revenue Architecture Report 2026.
