AI companies need billing decisions in milliseconds, not monthly invoices: Flexprice CEO Manish Choudhary

Traditional software billing systems are struggling to keep pace with artificial intelligence (AI) companies, where the cost of delivering a product can change with every customer request, according to Manish Choudhary, co-founder and chief executive of Flexprice.

Unlike conventional software-as-a-service (SaaS) companies, which typically charge customers for a fixed number of seats or a monthly subscription, AI companies must account for tokens, model usage, GPU txime, caching and calls to external tools.

The issue is becoming relevant as AI products move from pilots to large-scale deployments, exposing companies to volatile model and computing costs that traditional monthly billing systems were not designed to track.


“Pricing is no longer just a commercial decision, it has become part of the product’s runtime behaviour,” Choudhary told The Economic Times in an interview. “Two customers on the same $99 plan can cost wildly different amounts to serve.”

Also read: Tokenomics 2.0: The battle against AI costs

Flexprice, which recently raised $1.5 million from investors including Shastra VC, TDV Partners and Anupam Mittal, provides open-source billing and metering infrastructure for AI-native and API-first companies.

Choudhary said traditional billing platforms were designed around a relatively straightforward flow where a customer purchases a subscription and receives an invoice at the end of the month.

For an AI product, however, the invoice is generated only after a series of technical processes involving usage metering, aggregation, credit enforcement and dynamic pricing, he said.

“The first place old systems break is metering,” he said. “AI teams need to capture billions of events with full metadata, model, tokens, cached versus uncached, GPU seconds, customer and feature and older platforms were never built for that.”

Billing systems must also make decisions while a customer is using the product, Choudhary said. These include determining whether the customer has enough credits, whether a request should be allowed and if it should be routed to a cheaper model.

“Those decisions have to happen in milliseconds, not at the end of the month,” he said. “Billing has become production infrastructure, not back-office reconciliation.”

Also read: Tokens, tokens everywhere, but not a cent to spare

Measuring profitability request by request

AI companies cannot determine whether a customer is profitable by looking only at subscription revenue, according to Choudhary. They must also account for the variable cost associated with individual requests, features and models.

“Every request in the product gets tagged with a customer, a feature, a model and a cost,” he said. “The metric they actually track is contribution margin, revenue minus variable cost, at three levels, per request, per feature and per customer.”

Such analysis can show that a small group of users accounts for most of a company’s infrastructure costs or that one heavily used feature is eroding the margins generated by the rest of the product, he said.

Accurate usage data is therefore critical. Delayed, duplicated or incorrectly attributed usage events can lead to customers being overcharged or companies losing revenue.

“Metering is the ledger. If the meter is wrong, the invoice is wrong and no amount of downstream logic will fix that,” Choudhary said.

Also read: Let's minimise tokenmaxxing: India's next AI breakthrough will come from efficiency, not bigger models

Balancing predictability and consumption

According to Choudhary, there is no single pricing model that works for every AI company.

Usage-based pricing works when the value received by a customer rises predictably with consumption. Credits can make pricing easier to understand when a product uses several types of resources, while subscriptions remain useful for customers seeking fixed and predictable budgets.

“Most real deployments end up mixing two or three of these,” he said. “The answer is customer behaviour. Predictability, margin and ease of selling all matter, but they are downstream of how the pricing model changes what customers actually do.”

For enterprise contracts, one of the most common structures include a minimum spending commitment with charges for usage above that threshold.

“The customer commits to a floor for the term, which gives them budget certainty and gives you a revenue guarantee,” Choudhary said. “Usage above the floor is priced at a rate agreed on upfront, so nobody is renegotiating mid-term,” he added.

Flexprice uses a hybrid model for its own product. It offers tiered monthly plans with event and revenue limits, a free tier, customised enterprise plans and an open-source version that companies can host themselves.