Agents Earn Real Revenue Now. Data Is the Bottleneck

Two numbers landed within a day of each other this week, and they belong in the same sentence.

Salesforce reported that Agentforce, its AI agent product, reached US$1.2 billion in annual recurring revenue, up 205 per cent year on year, with more than half of bookings coming from existing customers and 3.8 billion "agentic work units" delivered to date. Whatever you think of the branding, that is a genuine product category with a revenue line, growing at triple digits, bought mostly by companies that already tried it and came back for more.

The day before, Accenture published research across 2,000 companies in 15 countries finding that only 7 per cent qualify as "data reinventors" with AI-ready data at scale. The same study found 64 per cent of companies have pushed AI past pilots into production, and that the 7 per cent with their data in order achieve up to 1.6 times the profit-margin uplift of their peers.

Put together: the agents are real, the money is real, and roughly nine in ten businesses are trying to run them on foundations that cannot support them.

The proof, close to home

If the Salesforce number feels abstract, the local evidence is not. Commonwealth Bank disclosed this week that an AI agent now joins its engineers on 2am incident calls, investigating outages from the moment a PagerDuty alert fires, targeting a 30 to 50 minute reduction in time-to-root-cause. The same underlying agent already handles about half of internal support requests at AWS, according to the same report.

Meanwhile the channel serving Australian mid-market businesses is straining to keep up. Kaseya's 2026 State of the MSP research, reported this week, found 48 per cent of managed service providers say AI is now their clients' top need, while only 13 per cent have turned it into a meaningful revenue stream. Demand from businesses like yours is outrunning the supply of people who can deliver it. That gap will not close this year, which means the businesses that prepare their own foundations will move first.

Why data is the constraint that bites

An agent is only as good as what it can see and touch. A quoting agent needs your price list, your job history and your margin rules in a form software can read. A service agent needs tickets, warranty terms and customer records that agree with each other. When those things live in seven systems, four spreadsheets and one long-serving employee's head, no model on earth ships a reliable agent on top of them.

That is what Accenture's 7 per cent figure is measuring, and it explains a pattern I see repeatedly in mid-market businesses: the pilot dazzles, then production stalls, and the post-mortem blames the tool. The tool was fine. It was pointed at data that was incomplete, inconsistent or locked in formats nothing else could use.

The margin finding is the incentive. The companies that fixed this are not marginally better off. They are running at up to 1.6 times the margin uplift of peers, because every subsequent AI initiative gets cheaper and faster once the foundations exist.

A readiness test you can run this week

You do not need a data strategy document. You need honest answers to five questions about the one process you would most like an agent to run:

1. Where does the data for this process live, and how many systems is that?

2. Could you export it today without asking a vendor for help?

3. Would two different people pulling the same report get the same numbers?

4. Is the knowledge needed to handle exceptions written down anywhere, or does it live in someone's head?

5. Who is allowed to see what, and is that enforced by the system or by habit?

Five clean answers and you are closer to the 7 per cent than you think; go trial something. Two or more ugly answers and your next AI dollar should go to fixing those, not to licences. Unglamorous, cheap relative to a failed deployment, and it compounds.

The agents have stopped being a demo. The revenue lines prove it. The only question left is whether your business will be ready to employ one when it shows up for work.

Where does your business stand? The free AI Scorecard takes three minutes and shows you. If you want a straight steer from a person, book an AI Opportunity Call.