The Uncharted Territory of AI Pricing

Companies are facing significant challenges in managing and pricing artificial intelligence services due to the unpredictable nature of AI's computational costs.

Will Venters, Associate Professor of Digital Innovation and Information Systems at the London School of Economics, highlighted that businesses can incur substantial expenses as staff experiment with or implement AI internally, leading to rapid consumption of "tokens," a unit of measure for AI processing. "People are finding it really hard to manage that cost… it's a non-deterministic output, so it's a non-deterministic value," he said.

Some smaller organizations are attempting to circumvent these issues by utilizing personal accounts with flat fees, a practice Oliver King-Smith, founder of engineering software firm smartR AI, believes the major AI vendors will eventually restrict. "This has to end at some point in time, because the big guys are taking a bath on those accounts," King-Smith stated, predicting increased oversight once AI platforms face pressure from shareholders to demonstrate profitability.

To mitigate escalating costs, King-Smith advises companies to be more judicious in selecting AI models and to refine their prompts for greater precision. Rob Steele, CFO at UK accounting software firm iplicit, used a shopping analogy, emphasizing the need for detailed instructions when requesting AI outputs, similar to how one would brief a family member on grocery shopping.

Scaling AI Costs

The complexity of cost control intensifies when AI is integrated into products intended for widespread distribution. Venters noted that beyond core software development, AI tokens may be required for tasks such as testing, security implementation, and the creation of "guard rails."

"It's particularly hard when you're looking at agentic processes," Venters said, referring to AI systems that can perform tasks autonomously. The ease of deploying multiple AI agents with a single click contrasts sharply with the deliberate processes involved in expanding a human workforce.

However, Venters also suggested that increased token expenditure might correlate with enhanced value, likening AI's output to an enhanced calculator rather than a simple tool. "The more you give it, the more expensive it is, but the better the result may be."

The Pricing Conundrum

Despite potential value, businesses must find ways to pass these costs onto their customers. Bill Peterson, senior director of product marketing at Sumo Logic, acknowledged that a universally accepted solution remains elusive. "Nobody's really figured it out," he said.

Sumo Logic is exploring pricing models for its new security services that leverage agentic AI. Potential strategies include across-the-board price increases, performance-based billing, or charging for "bundles" of services. However, Peterson pointed out that any chosen pricing structure could become obsolete if the underlying large language model providers alter their own pricing. "You get into variable pricing, and it's changing every couple of months," he noted. "Customers don't like that. That's not how anybody builds a budget."