IT firms grapple with deal pricing complexity in the age of AI agents

As AI agents begin working alongside humans, IT services firms are rethinking time-and-material contracts, experimenting with outcome-based and hybrid pricing models

The infusion of artificial intelligence (AI) agents across workflow processes to improve productivity and efficiency has brought a new set of challenges for information technology (IT) companies — how to set up a new pricing model that incorporates the value of a human and an agent working alongside each other on projects going forward.

 

It is still early days for enterprise AI adoption, which has lagged expectations, and most of the impact is being seen in areas such as improving customer experience, automating processes, code generation, and upskilling.

 

However, analysts say the traditional pricing model, which has long relied on time and material to determine how much a client pays per employee, needs to be tweaked as agents improve in performance and become more integrated into enterprise systems.

 

The first to speak about this was Mohit Joshi, chief executive officer and managing director of Tech Mahindra, while announcing the company’s Q3 FY26 results. In collaboration with research and advisory firm Forrester, Tech Mahindra has come up with a white paper that examines a pricing model incorporating both human effort and the increasing role of digital labour.

 

Joshi said last week that the company is looking at using some of the parameters discussed in the paper as a base to price some of the large programmes that it is winning.

 

The model distinguishes between human and digital labour, with the latter based on token consumption. Explaining the token model, Joshi said: “When we talk about token consumption, some of these tokens will be consumed by the models that we have built, and hence we will get the benefit of those. Some of these tokens used could be built by the client or by a third-party model and we will be charging a mark-up wherever appropriate on a third-party model. The idea is to bring transparency for the client on what we are charging for human labour and what is being done by digital labour.”

 

Research and advisory firm Forrester said a core challenge lies in aligning pricing models with value as AI agents become more autonomous. “And AI agents have yet to prove their value across all use cases; it’s not always evident that an LLM-infused agent is cost-effective for document automation when a simpler machine learning model could suffice. The industry is grappling with how to quantify and monetise AI agents in a market still defining their true utility,” it said in the white paper published jointly with Tech Mahindra.

 

“That could be one great model, but unless we start winning deals and revenue starts to flow, we will have to think of some other metrics. I don’t think anybody has come up with a metric which is candidly credible and auditable,” Joshi told analysts when asked if the company was planning to disclose AI-specific revenue.

 

Essentially, the new model will be more outcome-driven, with improved agents expected to bargain for better pricing and, in turn, better margins. When humans and AI agents work together, pricing can no longer be anchored to effort or headcount. Time and material breaks down the moment an agent can replace or augment dozens of human hours instantly.

 

“Value that we create will drive the billings, so a lot of that will be based on the traditional way and a lot of it will change as the overall AI market develops,” Infosys chief executive officer Salil Parekh believes.

 

Most large deals, therefore, see a combination or hybrid approach, where humans and agents are priced separately, as a blended team rate of human plus agent, or through a mix of models.

 

“You will also find within the same client multiple models co-existing — some fixed-fee work for managed services, outcomes linked to savings, and for newer builds creating pods that combine humans and agents. We expect this multi-model pricing environment to persist in the future,” said Jimit Arora, chief executive officer of Everest Group.

 

This, analysts say, typically unfolds in three phases. First, firms establish a baseline economics model, where providers and clients agree on today’s cost, cycle time, error rates, and revenue or productivity impact before AI is introduced.

 

Second, pricing moves to a unit-of-outcome construct rather than a unit of labour. Instead of full-time equivalents or hours, deals are priced around metrics such as cost per transaction. Third, providers increasingly use gain-share and risk-share bands, where a base fee covers the platform, governance and minimum service levels, while upside is shared when agents materially improve speed, quality or scale.

 

“The hardest part is not pricing the technology. It is building enough trust and data transparency to agree on outcomes and attribution. Firms that can credibly measure value, not effort, will win. Those that cling to time and material will see margins and relevance erode very quickly,” said Phil Fersht, chief executive officer of HfS Research.