AI’s ‘Token-Maxing’ Era Is Over As End Of Subsidies Forces Corporate Reckoning, Says Sabio AI Chief

The era of businesses burning through artificial intelligence with little regard for cost has ended, according to a senior industry figure, as the withdrawal of vendor pricing subsidies forces organisations to confront how little value much of their AI spending has delivered.

Sabio Group’s Chief AI Officer, Stuart Dorman, told City AM’s Business As Usual podcast that the AI sector has entered a new phase – one triggered not by the slide in technology share prices but by a fundamental change in how AI is priced and paid for.

“They called it the token-maxing era, where people were encouraged to use as much AI as possible,” Dorman told the podcast. “We’re definitely through that era now.”

The subsidy cliff

Tokens are the units of usage that determine what companies pay to run AI models. Dorman likened the system to a pay-as-you-go mobile phone, where heavy use quickly runs down available credit.

Until recently, that usage was heavily subsidised by the leading AI vendors. When those subsidies fell away around late May and early June, organisations could suddenly see the true cost of their consumption for the first time.

The result has been a wave of internal scrutiny, with businesses now questioning which use cases justify the spend and which AI models are fit for each task.

Market nerves meet pricing reality

Dorman’s comments come against a backdrop of falling AI-related share prices. One benchmark semiconductor fund has dropped almost 10 per cent over the past month, and investor sentiment around AI stocks has turned markedly cautious.

Dorman argued that the sell-off reflects expectations that were pushed too high rather than a fundamental problem with the technology itself.

“The market has been primed for perfection,” he observed, pointing to industry predictions that all jobs would disappear or that work would become optional.

Research suggesting enterprise AI adoption has lagged behind the hype has sharpened investor scrutiny, he added. Meanwhile, new Chinese AI models that are “dramatically undercutting” their American rivals on price have introduced a fresh source of competitive pressure, making it harder for US vendors to sustain their pricing.

Wasteful spending and misaligned incentives

Dorman was candid that much of the early corporate spending on AI had been wasteful. Some firms, he explained, had even rewarded staff for maximising their AI usage – an approach that in many cases produced significant expenditure with little measurable return.

The path forward, he argued, lies in enablement and training rather than raw consumption. That means helping employees identify the specific tasks within their roles that AI can support, rather than encouraging blanket adoption.

“AI doesn’t automate jobs; it automates tasks,” he stated.

The distinction matters because it reframes how organisations should approach AI deployment. Rather than seeking to replace entire roles, businesses stand to gain more by breaking down workflows and applying AI selectively to the parts where it can deliver a clear benefit.

Contact centres as a proving ground

Dorman singled out customer service within the contact centre industry – Sabio’s core market – as an area where the return on AI investment can be clearly measured. That measurement, he indicated, comes either by cutting the time taken to handle a customer interaction or by removing that interaction from human agents entirely.

Sabio Group, which positions itself as an AI-first customer experience specialist, has been deploying voice AI agents across contact centre operations for enterprise clients. The company operates an outcome-based pricing model for its voice AI platform, where clients pay per resolved call rather than through upfront licences or per-seat fees.

The company’s voice agents handle customer calls across functions including booking, billing, support and sales. Sabio claims its platform can process over 120 action types, supports more than 30 languages and operates with sub-300 millisecond latency.

The voice AI system integrates with contact centre platforms from vendors including Genesys, NICE, Amazon Connect, Five9, Twilio and Avaya, and is designed to write back to CRM systems in real time. Sabio has also built in compliance certifications spanning SOC 2 Type II, ISO 27001, PCI-DSS, GDPR and HIPAA readiness.

Enterprise case studies

Sabio has pointed to several client deployments as evidence that its approach can deliver measurable results in production environments.

Work with British Airways involved the analysis of 35,000 calls and the identification of 120 distinct call intents. According to Sabio, the airline was able to absorb a 70 per cent increase in call volume during disruptions caused by Heathrow strikes, adding over 60 new resolution pathways without increasing headcount. The company reported a 22 per cent reduction in contact centre workload.

A deployment with HomeServe UK covered 150 separate intents across transactional calls such as engineer rescheduling, claims and billing. Sabio reported that 85 per cent of customers completed their journey within AI-handled interactions, with typical resolution times of around 60 seconds.

Work with Marks and Spencer focused on AI-powered interactions and intelligent routing, with Sabio claiming 70 per cent routing accuracy compared to the retailer’s legacy IVR system and a 95 per cent customer engagement rate.

A deployment with Vodafone in Spain’s business-to-business segment produced a 27 per cent uplift in Net Promoter Score and a 36 per cent reduction in repeat contacts, according to Sabio. First call resolution improved by 5 per cent.

The enablement gap

The broader issue Dorman raised – that organisations have consumed AI without adequately training their workforces to use it – is one that extends well beyond the contact centre sector.

The token-maxing era, as Dorman characterised it, saw companies encouraged and even incentivised to burn through as much AI capacity as possible. That approach treated consumption as a proxy for progress, when the reality was that many organisations lacked the internal capability to direct that consumption toward useful outcomes.

The withdrawal of subsidies has exposed that gap. With the true cost of AI usage now visible, businesses are being forced to make harder decisions about where to deploy the technology and how to equip their people to use it.

An optimistic long view

Despite the near-term turbulence, Dorman struck an optimistic note about the longer-term trajectory of AI adoption.

He predicted a “significant acceleration” in enterprise uptake, although he acknowledged it would not arrive at the pace the largest technology firms would prefer.

Dorman cited a Gartner forecast that AI will become a net creator of jobs by 2029, a projection that runs counter to the more alarmist predictions that have dominated public discourse.

“This will be a technology that drives productivity and drives job creation,” he predicted.

For Sabio, the shift from raw consumption to measured deployment represents a commercial opportunity. The company’s outcome-based pricing model – where it does not invoice unless the AI achieves an agreed result – is designed to align its interests with those of clients who are now, for the first time, paying full price for their AI usage.

Whether the broader market follows that model or settles on a different approach, Dorman’s central argument is clear: the era of unchecked AI spending is over, and the companies that thrive will be those that move from consumption to capability.

Last Updated on August 6, 2026 by Nick Ross