The Predictability Problem how 50 businesses buy software, judge value and think about AI in 2026

Asked what matters most when evaluating a tool, respondents put ease of use and onboarding first at 26%. Price and value follows at 24%, features and security are tied at 18%, and support and documentation trails at 10%. Those gaps sit inside the margin we said we would not over-read, so treat this as a cluster rather than a ranking.

The reason usability leads becomes clear one question later. Only 16% say their teams are very satisfied with the usability of the tools they already own, and 32% call the experience mixed: some tools easy, others frustrating. So usability tops the buying criteria because buyers keep being let down by it. They are not expressing a preference. They are describing a scar.

The barriers tell the same story from another angle. Integration and data fit is now the biggest obstacle to adopting a new tool at 26%, ahead of subscription cost (24%), resistance to change (20%), learning curve and training (18%) and security approvals (10%).

Again the spread is tight, and we are not claiming integration has decisively beaten the rest. The shape is what matters. Five real barriers, none dominant, and the two that vendors spend the most effort on, training and onboarding, sitting at the bottom rather than the top. The hard part of adoption is no longer teaching people the software. It is making the software fit the estate it lands in.

Now the result that surprised us most. Asked which factor usually decides between two SaaS products, respondents put better usability first at 28%, integrations and ecosystem fit at 24%, and a stronger feature set at 20%. Lower price wins 14% of the time, second from bottom of five, ahead only of vendor reputation.

Sixty-four per cent call themselves price-conscious. Fourteen per cent let price decide. Any vendor who hears the first number and responds with a discount is answering a question nobody asked.

The two findings stop fighting once you separate the jobs price does. Price is a qualifier: it decides whether a product gets into the conversation at all, and a price that cannot be predicted or defended keeps it out. Price is rarely the decider. Once two products are both affordable, something else settles it.

Sixty-six per cent are comfortable or very comfortable with AI being involved in their KPI tracking, against 12% who are uncomfortable. For a subject that generates this much heat, that looks like a settled majority.

The make-up matters more than the total. The biggest single group, 40%, chose "comfortable as long as it is transparent and controllable". Only 26% chose the unconditional option of actively seeking AI features out. Most of the comfort in this market comes with strings, and the strings are specific: show me what the system did, and let me stop it. Those are product requirements, not reassurances.

Intent tracks a little below comfort. Sixty-two per cent would be likely or very likely to adopt new AI features, and within that, 38% chose "likely, if they add clear value" against the 24% who would roll them out quickly. The gap between 66% comfortable and 62% likely to adopt is small enough that we would not make much of it alone. What is not small is the third of this audience sitting in neutral, unsure or unlikely across both questions. Remember who these people are: an engaged, analytics-literate sample that had already chosen to use a data product, and a third of it was still not ready to switch AI on. Anyone sizing an adoption plan should size it against 62%, not against the absence of hostility.

Asked which AI capabilities would be most useful, respondents chose automated insights and explanations of trends (28%), intelligent alerts for unusual KPI behaviour (26%) and predictive forecasting (22%). Natural-language querying, the interface pattern that has dominated product marketing in this category, takes 16%. Automated report writing finishes last at 6%.

That last figure inverts the industry's current pitch. The capability promoted hardest across analytics products, generating written summaries, is the one this audience wants least. The three leading answers share a different character. They are all forms of interpretation: tell me what changed, tell me whether it matters, tell me what happens next. Producing more text about data the user can already see solves a problem they are not reporting.

Only 10% of respondents report no significant concerns about AI in business analytics. Accuracy and reliability of outputs leads at 32%, data privacy and security follows at 28%, loss of human judgement and context sits at 14% and regulatory or compliance issues at 12%.

Accuracy outranking privacy is the notable result. Privacy concerns get answered with policy, certification and contracts, which is familiar ground for any enterprise vendor. Accuracy concerns can only be answered in the product: show the working, cite the figures underneath, make it quick to check a conclusion instead of taking it on trust. Put that next to the 40% who made transparency and control the explicit condition of their comfort and the requirement is consistent and quite specific. This audience will accept AI that explains itself and asks permission. It is much less interested in AI that just writes confidently.

Half of respondents expect analytics and BI tools to become more central to their business over the next two to three years, used daily across more teams. Twenty-two per cent expect things to stay roughly as they are, 12% expect consolidation into fewer platforms and 10% expect analytics to become more specialised and handled by a few experts.

The two largest answers point in compatible directions. Whether a company ends up with more analytics or fewer platforms, both paths reward the same thing: a tool that earns its place in the working day rather than one that waits to be visited.