On 12 November 2026 Microsoft published a blog post by Alysa Taylor, its chief marketing officer for Commercial Cloud & AI. It previews IDC research that Microsoft commissioned, the fourth year it has done so. IDC surveyed 5,204 senior IT and business leaders across 28 industries in July 2026. It ranked their organisations on an AI maturity scale and compared the top 10%, which it calls Frontier Firms, with the bottom 30%, which it calls slow adopters. The post's main claim is that value and control are not a trade-off: the firms reporting the highest returns are also the ones most likely to have formal agent governance and company-wide AI training.
This matters to builders because the figures are likely to be quoted in procurement and budget discussions, especially around Microsoft Ignite on 18 November. They are survey data that a vendor paid for: returns are self-estimated, the maturity scale partly includes outcomes, and the full methodology document has not been released.
What we checked
- What concrete findings, examples, and recommendations does the IDC study report about scaling AI value while retaining control? partly confirmed. Microsoft's 12 November post lists concrete figures: return multiples by maturity group, agent-governance and skilling rates, a gap between seeing results and scaling them, four customer examples, and five recommended practices. The underlying IDC InfoBrief is listed as still to come. We could not read the text of the linked IDC infographic PDF, so we checked every figure against Microsoft's post only. [1][2]
- Are the headline return figures audited, causal or comparable with earlier editions of the study? refuted. Microsoft says the returns are reported by the leaders themselves and show association rather than cause. The 2025 edition used a different sample of more than 4,000 leaders and a different headline framing ('three times higher' returns), so the year-on-year multiples should not be read as a trend. [1][3]
What we found
- IDC put Frontier Firms in the top 10% of its AI maturity scale and slow adopters in the bottom 30%. According to Microsoft's post, the scale itself weighs spending discipline, breadth of use, agent adoption and skills, among other factors. So the groups are partly defined by the same behaviours the study later presents as their distinguishing traits, which makes some of the gap built in. [1]
- Frontier Firms report an average 4x return on AI versus 2.7x for slow adopters, and 39% of them report returns of 5x or more versus 4% of slow adopters. For agents specifically, the reported return is 3.8x versus 3.0x. Microsoft adds that these returns are leader-reported and that the research shows association, not cause. The post doesn't say whether the multiple is gross or net, or over what time period it is measured. [1]
- Across the whole sample, the problem the post describes is scaling. 66% of leaders say AI has delivered substantial or transformative results, but only 48% are scaling that value across several business units, and 36% are still piloting agents. Only 61% are highly confident they can measure AI's value and return, against 93% of Frontier Firms. The agent gap between the groups (3.8x vs 3.0x) is smaller than the overall AI gap (4x vs 2.7x), so agents are not where most of the difference comes from. [1]
- On control, 65% of organisations have an established process for governing agents, but only 60% are highly confident they can handle AI security and governance risks. The biggest differences between the groups are in governance and skilling. 87% of Frontier Firms govern and oversee agents versus 51% of slow adopters, and 80% have a company-wide AI skilling strategy versus 3%. [1]
- The source note cites 'Business Opportunity of AI Study, IDC Survey, July 2026, N=5204', and the post says the full InfoBrief will follow. The linked IDC infographic is a PDF whose text we could not extract, so we could not check the figures against IDC's own document. The 2025 edition surveyed a different group of more than 4,000 leaders and framed its headline as 'three times higher' returns, so the two years' multiples are not a like-for-like series. We searched once for an IDC methodology document for 2026 and found none published. [1][2][3]
What it means
Use the study as a list of practices, not as a forecast of returns. The practical signal is in the gaps between the groups: agent governance at 87% vs 51%, company-wide skilling at 80% vs 3%, and confidence in measuring returns at 93% vs 61%. These line up with what platform teams can build. Track every agent's identity and permissions, log what agents decide and do rather than only what models output, and measure processes before and after AI so the return is something you calculate rather than estimate. Do not put the 4x or 3.8x multiples into a business case. They are leaders' own estimates, the maturity scale partly includes the outcomes being compared, Microsoft itself calls the link associational, and Microsoft sells the platforms the study recommends.
For teams that already have pilots running, the 66% vs 48% figures describe the usual failure: value shows up in one unit and doesn't spread. Whatever the study's causal limits, it points to governance and skills as the bottleneck rather than model quality. Plan accordingly, and check against the full IDC InfoBrief when it is published.
Competing readings
- The study gives useful, directional evidence that governance and skilling go together with scaling AI value. It should be read as practice guidance, not as a measured return (likely, about 65%). What would tell it apart: The biggest differences between the groups are in governance and skilling, not in agent returns [1]; Microsoft itself says the returns are self-reported and associational [1].
- The gap between the groups is mostly built in: the maturity scale ranks firms on the behaviours later reported as their traits, so the findings mainly restate the scoring and support Microsoft's marketing (unlikely, about 30%). What would tell it apart: The maturity scale includes agent adoption and skills, which are the same traits the study highlights [1]; The headline framing changed between the 2025 and 2026 editions [3].
- The reported multiples are reliable estimates of the return companies will actually get from AI (almost no chance, about 5%). What would tell it apart: The returns are leaders' own estimates, with no audit or definition of how they were calculated [1].
Base rate
Of 48 earlier research stories about companies or organisations in the New Runtime archive over the last 12 months that are at least 3 months old, 40 (83%) were followed within 3 months by a release story on the same company or product line. Of 48 earlier research stories about companies, 40 were followed by a release story within 3 months. For Microsoft, which has Ignite on 18 November, a follow-on release is close to certain and says little about whether this study's findings hold up.
What would change this view
- The full IDC InfoBrief publishes a return definition based on measured financial data and maturity scores that exclude outcomes, and the gap between the groups remains (signal): The association would be stronger than a survey of perceptions, and the gap would not be built into the scoring.
- An independent survey not funded by a vendor finds similar governance and skilling gaps between high-return and low-return firms (signal): The governance–value link would not depend on who paid for the research.
Still open
- How exactly does IDC define and calculate 'return': gross or net, over what period, and from what survey question?
- How are the maturity-scale inputs weighted, and does excluding outcomes from the scale shrink the gap between the groups?
- How is the sample split by country and company size, and how is it weighted? None of this has been published.
- Will the forthcoming IDC InfoBrief match the figures in Microsoft's post?