Half the mid-market cannot name an AI gain, and only 20.2% call the impact large

Respondents were not asked for a variance analysis. They were asked whether recent technology or AI investments had produced any meaningful efficiency or performance gain, and on that self-graded question 50.5% could not say yes about their own operations ^[3][13]. The share calling the impact highly positive was 20.2% ^[4], which against the 722-person sample is roughly 146 organisations ^[1][16].

The interesting number is the 29.3 points between "any gain" and "large gain" ^[15]. For every leader reporting a highly positive result, about 1.45 report something positive but unremarkable ^[18]. That middle band is where subscription renewals get approved on vibes, because nobody in it has a figure to defend or to kill.

The concern list CLA reports is the tell. Leaders named implementation, governance, data quality, security, compliance, workforce readiness, change management and return on investment ^[10]. Almost none of that is a property of a model. Data quality especially is a pre-existing condition being diagnosed after the purchase order, which is the mechanism behind a 49.5% hit rate ^[3]: the pilot works, the plumbing does not, and the gain never reaches the P&L where someone could measure it.

Note who was surveyed. These are clients of CliftonLarsonAllen, a top 15 accounting firm, and the sample includes businesses, entrepreneurs, nonprofits and civic organisations ^[1][7]. Organisations already paying for accounting and advisory help are, if anything, better instrumented than the average operator. That makes 49.5% the flattering reading.

Meanwhile the spending intent has not moved. Respondents told CLA they plan to invest in technology and automation, enter new markets, pursue acquisitions, retrain talent and improve operations ^[12], all while absorbing rising costs in labour, materials, insurance, health care and financing ^[11]. CLA chief executive Jen Leary frames the change as leaders becoming more disciplined about where they place bets, and says the question is no longer whether AI is a sound strategy but how quickly it will pay off ^[8]. That reframing preserves the budget and relocates the argument to timing, which suits anyone who has already committed. James Watson, CLA's chief solutions officer, puts it more plainly: a year ago leaders asked how fast they could adopt, and now the conversation is practical use cases, preparing people, governance and measuring outcomes ^[9].

There is one internal contradiction worth holding onto. In the same survey, 69.9% say their workforce has the skills needed for future success ^[5], while workforce readiness appears on the list of biggest AI concerns ^[10]. Both cannot be comfortably true. Either the skills question is being answered about the business as it runs today, or the readiness worry is about a workload nobody has scoped yet.

On this evidence, the gap between the 20.2% and everyone else is unlikely to be model selection ^[4]. It is more likely to be whether a baseline existed before deployment. Efficiency claimed after the fact, against no prior measurement, is exactly the sort of answer that lands in the soft middle of this survey.