The mechanism on offer is organisational distance rather than money. Epicor's Kerrie Jordan argues that firms in the middle bands sit at what she calls a sweet spot of capital, complexity and decision cycles: enough budget to put tools in frontline hands rather than only into back-office dashboards, enough operational complexity for those tools to bite, and a small enough headcount that deployment and training clear without much friction [7]. Against that, she describes enterprise pilots that stall in planning or at the corporate layer and never reach a warehouse, store or factory floor, with the prospect of training thousands of people acting as a further deterrent [8]. If that account holds, what a large employer would need to copy is an approval path, and approval paths are not on any vendor's price list.
How much weight the finding can bear depends on numbers the piece does not print. It reports one total, 1,038 respondents worldwide [2], and states both of its central results as "the majority" of a filtered subgroup without giving the respondent count in any headcount band [13]. The bands named across the two results are 1 to 49, 50 to 249, 250 to 999, and 5,000 or more, which leaves the 1,000 to 4,999 range in neither the accelerating group nor the stalling one [14]. The boundary at 999 is therefore where the reported bands stop, not where a measured advantage was shown to end.
A skeptic would put it bluntly: the report is published under the byline of Epicor's chief marketing officer [1], it asks frontline workers to characterise their employer's investment plans one to three years out [4], and a result that flatters midsize buyers is a convenient result to publish. Half of that objection lands. The execution reading, though, draws support from a figure the article itself borrows: McKinsey found nearly two-thirds of businesses using AI in at least one function are still experimenting or piloting and have not begun to scale across the enterprise [10]. Those companies are already spending. Whatever holds them sits after the budget decision, not before it. The claim that only 8.8% of small businesses use AI deserves less weight, since the article identifies its source only as a recent report from Robert Press [11].
For an operator at the 5,000-plus end [5], the trade-off is better named than discovered. You can manufacture midsize conditions by scoping frontline AI to one business unit with its own budget and its own sign-off, and it will move faster than a group programme. You also give up what centralising was for: a single contract, one data model, one security review. That bill does not arrive this quarter. It arrives when the unit that moved fast has a working stack, the other units want theirs, and consolidation runs at the pace the survey attributes to large organisations [8].
The last thing to keep separate is horizon. A single survey year, resting on stated intentions over one to three years [4], tells you who got tools to the front line first; it does not tell you whether that lead compounds, or whether scale and capital reassert themselves once stalled pilots unstick. On this record we do not know. What the report supports is a reading about sequencing, and the sequencing carries its own consequence: the firms in the middle bands are gathering frontline feedback on live tools now [6], and that feedback is the input to whichever purchase comes next.
