Something interesting is happening in enterprise AI right now. Organisations are celebrating how many AI agents they have built. How many pilots are running. How many employees have access to AI tools. How many use cases are in production.
These are the wrong numbers to celebrate.
The activity trap
There is a pattern emerging across enterprise AI deployments that should give every CIO pause.
Organisations are confusing motion with progress.
Building AI agents is not an AI strategy. Deploying AI tools across the enterprise is not an AI strategy. Publishing a list of 100 potential AI use cases is certainly not an AI strategy.
A strategy requires knowing which of those agents, tools and use cases actually matter − and concentrating investment there. Most organisations have not done that work yet.
The data is beginning to tell a clear story. When Prosus − one of the largest operators of enterprise AI agents in the world outside of companies whose core business is selling AI − analysed its deployment of more than 60 000 agents across 40 000 employees, the finding was striking.
There is a pattern emerging across enterprise AI deployments that should give every CIO pause.
Approximately 2% of active agents generated a disproportionate share of measurable business value.
The rest delivered value that was real but diffuse. Personal productivity gains. Time saved on routine tasks. Individual convenience. Useful. But not transformative.
This is not a Prosus problem. It is a pattern. The same dynamic that governs revenue distribution across customers, returns across investment portfolios and performance across sales teams is now showing up in AI deployments.
A small number of use cases create most of the value. The rest fill a spreadsheet.
What the 2% actually looks like
The agents that break through share identifiable characteristics. They solve a specific, recurring operational problem. They serve many employees rather than one. And their value connects directly to revenue, cost reduction, or time savings measured in months rather than minutes.
One Prosus portfolio company deployed AI agents to run a new third-party affiliate marketplace − a business model that was previously uneconomical because each affiliate partner generated too little value to justify human management. The AI made it viable. Projected annual revenue: $83 million.
A food delivery business used an agent to support thousands of small independent restaurants that no account manager could economically serve. Orders from that segment increased by 119%. Retention improved by 73%.
A travel platform built an agent that answered highly-specific property questions by synthesising internal data with external reviews. Customers who engaged converted to bookings at a rate 138% higher than those who did not.
These are not productivity tools. They are business model enablers. That is a fundamentally different category of AI.
The majority of agents are not failures
Precision matters here. The 98% of agents delivering modest individual value are not mistakes. Personal AI assistants that help employees draft documents, manage their schedules and navigate internal systems have genuine utility.
The problem is not that they exist. The problem is when organisations treat them as evidence of AI maturity.
An agent that saves one employee 30 minutes a day and an agent that enables $83 million in new revenue are not equivalent achievements. Both fall under the label of AI deployment. Both appear in the same adoption metrics. Both make the agent count go up.
Only one of them belongs in the boardroom conversation.
The question nobody is asking
For years, business leaders have understood that not all customers create equal value. Not all products contribute equally to growth. Not all investments generate the same returns.
The same reality is now emerging in enterprise AI − and most organisations are not yet asking the question that follows from it.
Which AI capabilities are actually changing how our business creates value?
Not how many agents are running. Not how many employees have access. Not how many use cases are in the pipeline. Which ones are changing the economics of the business itself?
That question reframes the entire conversation.
A practical test worth applying today
The Prosus research offers one diagnostic that requires no data science capability whatsoever. The group calls it the “delete it tonight” test.
Ask the leader of any business unit a single question: what would happen to revenue or costs if this AI agent were permanently removed tonight, with no possibility of recovery?
If the answer is vague, the value is vague.
If the answer is immediate and specific − revenue would drop, a key workflow would collapse, a customer segment would go unserved − you have found something worth investing in further.
Apply this test across your AI portfolio. Most agents will produce an uncomfortable silence. A small number will produce an answer that justifies everything.
The governance gap
There is a structural reason most organisations have not yet done this work. AI governance conversations remain largely focused on risk, compliance and controls. These are important. They are not sufficient.
Governance must also become a mechanism for identifying and concentrating value. That means asking, at a regular executive cadence, which agents are gaining genuine adoption, which are delivering measurable outcomes and which deserve additional investment.
It means being willing to stop funding low-impact initiatives regardless of how much has already been spent on them. And it means resisting the political pressure to treat every department’s AI initiative as equally important simply because every department exists.
Not all AI creates equal value. Governance that pretends otherwise is not governance. It is administration.
The competitive reality
The organisations that will lead in AI over the next three to five years will not necessarily be those with the largest agent portfolios or the biggest AI budgets.
They will be the ones that identify their power law use cases fastest − and scale them before their competitors do.
The technology is increasingly accessible. The discipline to translate it into business value remains scarce. That discipline is the competitive advantage now.
The question every executive team should answer:
Not: how many AI agents have we deployed?
Not: how many employees are using AI?
Not: how many use cases are in production?
But: what are the handful of AI capabilities that are fundamentally changing how we create, deliver and capture value?
Find those. Scale them relentlessly. Everything else is activity.
* Bramley Maetsa writes on enterprise technology strategy and digital transformation for ITWeb.
Share
