Intelligent automation in 2027 will be shaped less by new technology and more by proof. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027, while UK rules on automated decisions loosened in February 2026, so the firms that win will be the ones that can show a measured return and a clear audit trail.
This is our 2027 edition. The 2026 edition covered the shift from scripted bots to agents and the rise of low-code tools. Those trends have not gone away. What has changed is the evidence: we now have a year of adoption data, a revised EU timetable and new UK law. Every figure below comes from a source we checked on 6 November 2026, listed at the end.
| Trend | The evidence | What it means for a UK mid-sized firm |
|---|---|---|
| 1. Fewer, more accountable agent projects | Gartner: over 40% of agentic AI projects cancelled by end of 2027 | Start with one agent on one process, with a named owner |
| 2. Agent identity becomes the blocker | Gartner: through 2028, over 50% of AI initiatives halt over unresolved agent identity | Decide what each agent may access before you build it |
| 3. The mid-sized gap | McKinsey: 40% of large firms scale agents, smaller firms flat at 22% | Room to move ahead of peers with focused projects |
| 4. Admin work goes first | ONS: 38% of AI-using firms (10+ staff) say admin and clerical roles are most affected | Document intake, data entry and reporting are the first targets |
| 5. UK allows more automated decisions | Data (Use and Access) Act 2025, in force from 5 February 2026 | Automated decisions are possible, with four safeguards |
| 6. EU deadlines move, not vanish | AI Act high-risk rules from 2 December 2027 (Annex III) and 2 August 2028 (Annex I) | Firms selling into the EU get time, not an exemption |
| 7. Proof of return as the gate | McKinsey: 37% attribute at least some EBIT impact to AI | Measure before you build, or the project will not survive review |
1. Fewer agent projects, each with an owner
In 2026 almost every software vendor added the word agent to its product. In 2027 the projects built on that promise will be reviewed. Gartner predicts that by the end of 2027, over 40% of agentic AI projects will be cancelled because of escalating costs, unclear business value or inadequate risk controls.
That is not a reason to avoid agents. Gartner also expects at least 15% of day-to-day work decisions to be made autonomously by agentic AI by 2028, up from 0% in 2024, and 33% of enterprise software applications to include agentic AI by then. The direction is clear. The lesson is about scope. Projects that survive tend to have one process, one owner and a measured baseline. If you are unsure what counts as an agent, our guide to AI agents in the enterprise sets out the difference from scripted automation.
2. Agent identity and permissions become the bottleneck
An agent that reads your inbox, updates your CRM and drafts client letters needs access to all three. Most firms have not decided how that access is granted, logged and taken away. Gartner predicts that through 2028, more than 50% of AI initiatives will halt because of unresolved agentic identity challenges.
For a mid-sized firm this is practical, not abstract. Before building an agent, write down which systems it can read, which it can change, whose account it acts under, and who can switch it off. If your IT provider cannot answer those questions, that is the first piece of work, and it is far cheaper to do before the build than after an incident.
3. The gap between large and mid-sized firms
McKinsey's State of AI survey, published in August 2026, found that 40% of respondents from large organisations (over $1 billion in revenue) report scaling AI agents, up from 27% a year earlier. Among smaller organisations the share stayed flat at 22%.
UK official figures point the same way. The ONS Business Insights and Conditions Survey found that 30% of UK businesses used at least one AI technology in September 2026, rising to 38% of businesses with 10 or more employees. Adoption has consistently been higher in larger firms. For a firm with 50 to 250 staff, that gap is an opening. A focused project on one costly process can put you ahead of most peers without an enterprise budget.
4. Admin and clerical work is automated first
When the ONS asked businesses with 10 or more employees that use AI which roles were most affected, 38% named administrative and clerical roles in September 2026, up 9 percentage points since the question was first asked in December 2025. Creative and design roles came next at 24%.
The same survey found that 6% of these firms reported reduced headcount as a result of AI, a figure that has been broadly stable over the past year. The pattern is that admin work changes before jobs do: fewer hours retyping documents, chasing missing information and assembling reports. That is where intelligent document processing earns its keep, and where most mid-sized firms will find their first measurable saving in 2027.
5. UK law now allows more automated decisions, with safeguards
The Data (Use and Access) Act 2025 brought its main data protection changes into force on 5 February 2026. One of the most important for automation is the change to automated decision-making. The general ban on solely automated significant decisions has been lifted, except where special category data such as health data is involved.
Four safeguards still apply to every such decision. You must tell people an automated process is being used, give them a chance to make representations, offer a route to meaningful human intervention, and let them contest the decision. In practice this means the human review step does not disappear. It moves to where a person asks for it, and it has to be designed in from the start, with records of what the system decided and why. Our guide to using Claude with client data under UK GDPR covers the data side of the same question.
For firms regulated by the FCA, the position has not changed. The FCA states that it does not plan to introduce extra regulations for AI and will rely on existing frameworks instead. The firm stays accountable for the outcome, whatever tool produced it.
6. EU AI Act deadlines move, but do not disappear
Many UK firms sell into the EU, so the AI Act still matters. Following the Digital Omnibus changes, the European Commission's AI Act Service Desk now lists 2 December 2027 for the rules on stand-alone high-risk systems in Annex III, such as AI used in credit scoring or recruitment, and 2 August 2028 for high-risk AI built into products covered by Annex I. The transparency rules in Article 50 have applied since 2 August 2026.
The delay gives firms time to prepare, not a reason to wait. Documentation, risk classification and human oversight take months to build into a live system. A firm planning an automation project in 2027 that touches EU customers should design for those rules now, so it does not have to rebuild in 2028.
7. Proof of return becomes the gate for new projects
Belief in AI is high. Gartner reports that more than 90% of IT leaders believe AI agents will deliver significant productivity benefits. Evidence is thinner. In McKinsey's 2026 survey, nearly nine in ten respondents report regular use of AI in at least one business function, but only 37% attribute at least some EBIT impact to it, about the same share as the year before.
That gap is why boards will ask harder questions in 2027. The answer is simple and often skipped: before anything changes, record four numbers for the process. How many cases a month, how long each takes, how often work is redone, and how long a case takes from start to finish. Measure the same four after launch, and count the full cost, including the time someone spends checking the output. Our AI barriers guide explains why projects without this baseline stall after the pilot.
What to do in the first half of 2027
Pick one process. Choose the highest-volume admin task where the cost is visible, such as document intake, onboarding checks or monthly reporting.
Write down the baseline. Volume, time per case, rework rate and turnaround, before any tool is bought.
Settle access and oversight. Decide what the system can see and change, who reviews its exceptions, and how a client can ask for a person.
Check your rules. UK GDPR safeguards for automated decisions, FCA expectations if you are regulated, and the AI Act dates if you serve EU customers.
Build, measure, then decide. Run the automation on real work for a month and compare the four numbers. Scale only what the numbers support. If you want to see how a fixed-scope first project is priced, our guide to AI consultancy costs in the UK sets out typical ranges.
How Kelriva can help
We are a two-person London consultancy that builds document and workflow automation for UK firms, with the human review step and audit trail designed in from the start. If you want to know which of your processes is ready for automation in 2027, our free AI Readiness Score takes about two minutes and gives you a tier and a specific next step.
Sources
Checked on 6 November 2026: Gartner, AI agent success hinges on reliability, not autonomy (40% of agentic AI projects cancelled by end of 2027, over 50% of AI initiatives halting over agent identity, 90% of IT leaders); Gartner press release, 25 June 2025 (15% of day-to-day decisions and 33% of enterprise apps by 2028); McKinsey, The state of AI in 2026 (published 25 August 2026); ONS, Business insights and impact on the UK economy, 8 October 2026; The Data (Use and Access) Act 2025 (Commencement No. 6) Regulations 2026 and Travers Smith on the automated decision-making changes; European Commission AI Act Service Desk timeline; FCA, AI and the FCA: our approach. This article is general guidance, not legal advice.