B2B AI PlatformB2B AI Platform
AI business models need a total reboot
B2B AI Platform

AI business models need a total reboot

Emma RobertsWords by Emma Roberts

The measured size of it

$2.7 trillion
Worldwide AI spending
up 49.5% annually
37%
AI contribution to EBIT

Almost every large enterprise now runs AI agents in production and has a flat profit line to show for them. The cause is not model quality. Each agent inherits a funding, governance and liability structure built for buying software. Delegated work needs a different one. Fix one of the three and the other two pull the result back. The reboot is structural, and the pilots are already running.

Fragmentation is a budgeting failure before it is a technology failure

Most AI estates are not portfolios. They are accumulated purchases. IDC | Trusted Tech Intelligence reports in its FutureScape briefing on AI's business model reinvention that 99 per cent of CEOs expect AI to rewrite their business model inside three years, that 95 per cent of enterprises have at least one agent workflow live with an average of eleven in flight, and that 61 per cent of organisations remain stuck in the bottom two stages of its adoption scale, which it labels ad hoc and opportunistic. That estate has grown faster than the operating model around it.

The financial picture explains the impatience. The National reported that McKinsey's Global Survey on the State of AI found only 37 per cent of respondents able to attribute any AI contribution to earnings before interest and tax, a share unchanged on a year earlier. The same article records Gartner's forecast of $2.7 trillion in worldwide AI spending this year, up 49.5 per cent annually. Spend is compounding. Evidence of business model change is not. That gap is the reboot case.

The unit of reboot is the workflow

An agent changes a business model only when it owns a complete unit of work. A chatbot that answers questions inside an unchanged process is a faster version of the old model. An agent that owns a claims file, a renewals conversation or a procurement decision changes what the company sells, how it delivers and how it charges. IDC's briefing draws the same line between speeding up existing work and rewriting the commercial shape of the business.

MIT CISR's research briefing Business Models in the AI Era predicts that business models will shift toward outcomes and autonomous AI, and it sorts the options into four. Existing+ uses AI to improve an established process while the company assists the customer. Customer Proxy uses AI to act for the customer within predefined rules. Modular Curator uses AI to assemble reusable modules into a tailored bundle. Orchestrator uses AI to pursue an outcome across an ecosystem on the customer's behalf. The same briefing records that 81 per cent of the companies MIT CISR tracks led or joined a digital ecosystem in 2025, up from 30 per cent in 2013. The direction is clear. The operating model has to catch up.

Funding follows ownership, so split the budget deliberately

If the business owns the outcome, the business should own the budget line. IDC's briefing puts the central IT share of agent inference and related spend at 39 per cent, with business units paying the rest directly. That is not a leak to be plugged. It is a signal that the sales, finance or customer experience team already owns the result. IT's role shifts to platform, guardrails and shared services. Spend visibility becomes the harder job, because AI costs no longer sit on one line.

The practical response is a single AI profit and loss view with three columns: central platform cost, business unit experiments, and production agents. Each production agent gets a named business owner, a funding source and a revenue or cost line it is expected to move. Without that view, the board sees a total spend figure and no way to judge whether the portfolio is becoming a business model or a collection of subscriptions.

Governance starts with outcome quality

Uptime is not a governance metric. A system can run perfectly and produce work that is wrong, late or unauthorised. Boards now ask whether AI spending is producing measurable business value, and technical teams need a scorecard that answers them. The National reported that Gartner's Denis Torii said AI adoption is "not an IT story" and that the organisation itself has to adapt. He is right. Governance belongs to the business process owner, with IT supplying the controls.

The scorecard should measure outcome quality, the cost of rework, the revenue at risk when an agent fails, and the time a human needs to correct the output. Those measures connect the agent to the P&L. They also give the board a way to compare one agent against another without arguing about model benchmarks.

Prove trust with evidence

Trust is an evidence product. It is provenance, data lineage, audit trail and a named person who can explain what the agent did. IDC | Trusted Tech Intelligence reports that its research vice president for AI-fuelled business strategies, Andrea Siviero, says the winners will "govern it, fund it, and prove it can be trusted". That is the standard. A policy document is not proof. A demonstration is not proof. An exportable record of inputs, decisions, tool calls and human approvals is proof.

Insurance belongs in the same conversation. If an agent can make a purchase, change a price or authorise a refund, the company needs to know what is covered and what is excluded. The evidence pack should be ready before the agent goes live. Assembled after the first incident, it documents a loss the company has already taken.

Accountability must be settled before autonomy

An agent that can act is a liability with a latency problem. When it makes a faulty purchase or oversteps its authority, three parties can end up answering for it: the deploying company, the vendor that built it, the platform provider. IDC's briefing notes that the answer is unclear today and that even the EU AI Act leaves the question open. That uncertainty changes how companies budget for risk and document how their AI was built.

Fix accountability before scaling. The sequence:

  1. Name the accountable executive for the agent.
  2. Set the agent's authority limits in writing.
  3. Define the escalation path when a tool call fails.
  4. Record the human who approves an exception.
  5. Run a tabletop exercise that assumes the agent has already caused a loss.

The exercise is cheap. A live incident without a named owner is expensive in every direction.

Human service becomes a product line

As AI absorbs routine customer interactions, a person on the line becomes a premium tier. IDC's briefing projects that rising automation raises the value of a real human. That reverses a long-standing assumption in service economics. Human contact turns into a line item customers can choose and pay for, at a price the company sets deliberately. The companies that get this right will price human access, staff it for judgement, and measure it as a product line with its own margin.

This is not a reason to slow automation. It is a reason to stop treating every human interaction as waste. The business model question is which conversations deserve a person and what the customer will pay for that access.

Policy risk is now a line item

AI business models are exposed to policy choices. The AI Journal reported that OpenAI paused its £31bn Stargate UK project in April, citing regulatory uncertainty. For a UK board, that is a reminder that infrastructure, data residency and model access can change faster than a three-year plan. A business model that depends on AI capacity needs a policy assumption in the same way it needs a demand assumption.

The response is to write that assumption into the business case, next to the demand forecast, so a regulatory shift shows up as a scenario before it shows up as a write-down. Boards that fund AI without one are underwriting a model they cannot see.

Sources

Source: AI Business Model Reinvention: Govern It, Fund It, Prove It, IDC | Trusted Tech Intelligence

Recently filed

More on this

Artificial Intelligence In Business
Artificial Intelligence In Business
AI Consulting For Business, Compared
AI Consulting For Business, Compared
AI In Finance UK: What Changes in Practice
AI In Finance UK: What Changes in Practice