Six shifts reshaping enterprise data and AI budgets for 2027

Shift 1Agents are scaling at large organisations and stalling at smaller ones

McKinsey surveyed 1,719 people in 97 countries between 4 May and 8 June 2026. Among respondents from organisations with more than $1 billion in revenue, 40% said they were scaling AI agents, up from 27% a year earlier. Among smaller organisations the figure stayed at 22%.¹

An 18-point gap is large for a technology this young. Agents depend on shared platforms, curated data and an approved route into production, and large organisations usually have more of each. Coding agents follow the same pattern: about two respondents in ten are scaling them, rising to 31% at larger enterprises.¹

Gartner offers the counterweight. In June 2025 it forecast that more than 40% of agentic AI projects will be cancelled by the end of 2027, because of escalating costs, unclear business value or inadequate risk controls. It also estimated that only about 130 of the many vendors selling agentic products offer the real thing.²

Shift 2Spending has run ahead of returns, and each study measures return differently

In McKinsey’s 2026 survey, 37% of respondents attributed at least some EBIT impact to AI, roughly the same share as a year earlier. The group attributing 5% or more of EBIT stayed at about 6%.¹ IBM asked 2,000 CEOs in 33 countries between February and April 2025: 25% of their AI initiatives had delivered the expected return in recent years, and 16% had scaled across the enterprise.³

BCG sorted 1,250 senior executives into three groups. Five per cent were "future-built", 35% were scaling and 60% reported little revenue or cost benefit. The future-built group reported about twice the revenue growth and 40% more cost reduction than the laggards.⁴

The figure most often quoted in board papers, that 95% of generative AI projects return nothing, comes from MIT NANDA. It rests on more than 300 public initiatives, 52 interviews and 153 survey responses, and it counts measurable P&L impact over a short window.⁵ Set beside the others, the four studies are measuring four different things.

None of these can be averaged. Read together they support one statement that survives every definition: most organisations are spending more than they can yet show a return on, and a small group is doing much better than the rest.

Shift 3Top performers redesign the workflow first

McKinsey asked whether respondents had fundamentally redesigned workflows because of AI. Nearly three-quarters of high performers had, up from 55% the year before. About a quarter of other respondents had.¹

The three findings line up. Value collects in the work that makes or sells the product, it needs the process rebuilt around the model, and it suffers when every team buys its own stack. IBM adds that 68% of CEOs see an integrated, enterprise-wide data architecture as critical to getting value from AI.³

Ownership follows from that. A workflow redesign needs someone from the business line with authority over the process, supported by a data and AI team that runs the platform. A central AI team asked to deliver results on its own has the platform and lacks the authority.

Shift 4Data quality is back on top, and shared meaning is becoming infrastructure

BARC’s Trend Monitor for 2026, based on 1,579 responses collected in the summer of 2025, put data quality management back in first place with an importance score of 7.9 out of 10. Agentic AI came 19th of 20, at 4.5.⁶ Salesforce’s 2025 survey found that data and analytics leaders estimate 26% of their organisation’s data is untrustworthy.⁷

Gartner has been making a related argument about meaning. In May 2026 it forecast that organisations prioritising semantics in AI-ready data could improve agent accuracy by up to 80% and cut costs by up to 60% by 2027, and it recommended a context layer as core infrastructure.⁸ In March it predicted that universal semantic layers will be treated as critical infrastructure, alongside data platforms and cybersecurity, by 2030.⁹

Standards work has started. Snowflake announced the Open Semantic Interchange in September 2025, and the first version of the specification was published under an Apache 2 licence in January 2026, with Databricks, Qlik and AtScale among the new members.¹⁰¹¹ On the tool-access side, the Model Context Protocol moved to the Agentic AI Foundation, a directed fund under the Linux Foundation, in December 2025.¹²

The 80% and 60% figures are forecasts carrying the words "up to", and they are best read as a direction. The direction matches what practitioners told BARC, and what Gartner found in 2025: 63% of organisations lack, or are unsure they have, the data management practices AI needs.¹³

Shift 5The regulatory calendar now sets the order of work

Several EU obligations reach data and AI teams between 2025 and 2028, and some dates moved this year. The AI Omnibus, Regulation (EU) 2026/1744, was adopted on 8 July 2026 and entered into force on 27 July.¹⁴ It moved high-risk obligations for standalone (Annex III) systems from 2 August 2026 to 2 December 2027, and for AI in regulated products (Annex I) to 2 August 2028.¹⁵

The December 2027 date gives more time and the same amount of work. Praxikon points out that deciding which systems count as high-risk has not been postponed, and that the final Article 6 guidelines are expected around the end of 2026.¹⁵ The Data Act dates matter for a different reason: from 12 January 2027 a provider can no longer charge a customer for leaving, so an exit plan becomes something a vendor has to support.¹⁶

Shift 6Governance is moving from datasets to decisions

Gartner published its first Magic Quadrant for Decision Intelligence Platforms in March 2026, describing tools that combine business rules, machine learning and generative AI to structure a decision and track its outcome.¹⁹ Its June trends added a figure: explicitly modelled business decisions will be five times more trusted and 80% faster than ungoverned ones by 2029.²⁰

That is a forecast, and the reasoning behind it holds up. A modelled decision has an owner, defined inputs and a recorded outcome, so an agent acting inside it can be audited. Governance applied to a dataset tells you what an agent could see. Governance applied to a decision tells you what it did and who answers for it.

Culture sits underneath all of this. In a Gartner survey of 223 data and analytics leaders in March 2026, cultural resistance outweighed funding as the main reason governance initiatives fail, by 60% to 40%.²¹

ReferenceThe forecasts used in this report, with dates

Forecasts tend to travel without their publisher or date. Here are the ones cited above.

Most of these come from press releases with little method attached. They help size a risk or test a plan, and they say little about any single company.

For the planSix questions for the 2027 budget