Margin and headcount: the criterion that cannot be marketed

This is the arithmetic criterion, and it exists because the other six can all be described persuasively by an operation that has changed very little. Capability claims are cheap. A capacity curve is not.

Three ratios carry most of the signal:

• Revenue per employee, compared against MSPs of similar size and service mix. This is blunt, publicly comparable, and the number a buyer or an acquirer will reach for first.
• Endpoints or users under management per delivery engineer. Narrower than revenue per employee and harder to flatter, because it excludes sales and administration and speaks directly to delivery leverage.
• The slope of the two. A one-off improvement is a reorganisation. Leverage that holds as clients are added is a different operating model, which is the thing the criterion is actually testing.

Gross margin on managed services matters too, with a caveat worth stating plainly: automation moves cost rather than deleting it. Labour comes out of cost of delivery and platform, licensing and inference costs go in. An MSP that has genuinely automated will often show a margin improvement smaller than its labour saving, because part of the saving has been paid to the vendors whose agents are doing the work. That is not a failure of the claim. It is the arbitrage described in Automation Arbitrage being visible in the accounts.

Several patterns produce ratios that look AI-native and are not:

• Offshoring. Moving delivery to a lower-cost labour market improves margin materially and has nothing to do with AI. It is a legitimate strategy and it is not this criterion.
• Mix shift. A pivot toward project or licence resale revenue lifts revenue per employee without changing how managed work is delivered.
• Acquisition accounting. In a roll-up, the ratio in the first year after a deal reflects what was bought, not what has been rebuilt. The number worth reading is the one for estates the acquirer has already migrated onto its own platform.
• Under-servicing. Margin can also be bought by answering the phone more slowly. Capacity metrics should be read next to whatever service quality measure the MSP already keeps.

The index is not an auditor, and most submitters are private companies with no obligation to hand over financials. We do not require them, and we do not publish figures a submitter has not agreed to publish.

What we ask for is direction and mechanism: which ratio has moved, over what period, and what specifically changed to move it. An MSP that can name the workflow it automated and connect it to a capacity number is making a checkable claim. An MSP that reports a margin figure with no mechanism behind it is reporting a number.

Where an MSP has published figures, whether through a funding announcement, an acquisition disclosure or its own reporting, we will read them and say what they show. Where nothing is public we say that too, rather than implying verification that did not happen.

Operations reach the capability criteria before they reach this one. Agents get deployed, triage improves, resolution rates climb, and headcount stays where it was because nobody is dismissed on the strength of a pilot. The ratios only move once the organisation is restructured around what the agents now do, which is a management decision rather than a technical one, and it typically lags the technology by several quarters.

That lag is why the index accepts a single criterion for listing. An MSP part-way through this transition is genuinely AI-native in its delivery long before its accounts admit it.