AI in business by 2027 will look less like a technology gap and more like an execution gap. Adoption is already near universal, but the financial returns are not, and the data shows the split is caused by how organisations work rather than which models they buy. Companies that redesign a process around AI systems are seeing bottom-line impact; companies that add AI to an unchanged process are seeing individual productivity gains that never reach the P&L. That gap is what the next two years decide.
Key Takeaways
- Adoption is no longer the differentiator. Nearly nine in ten organisations already use AI in at least one function.
- Individual productivity is up sharply, but enterprise financial impact is flat. 80% of people report personal productivity gains while only 37% of organisations report any EBIT impact.
- The organisations seeing real returns redesign workflows. Nearly three-quarters of high performers did so, against one-quarter of everyone else.
- Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, mostly for cost, unclear value, and weak risk controls.
- Build-versus-buy is shifting. Nearly a third of organisations have already declined to purchase software because they could build it with agentic coding tools.
- Operating cost is becoming a real constraint for about one in five organisations, which makes cost modelling part of AI design rather than an afterthought.
🔧Talk to Our Team About Your AI Roadmap
Where AI in Business Actually Stands Today
AI in business means using AI systems to do or assist real work: answering customers, drafting and analysing, writing code, and increasingly taking actions inside company systems. The category now spans three distinct things that get confused with each other, and the distinction drives cost and risk: assistants that advise a person, automation that follows a fixed path, and agents that choose their own path and act.
Any credible 2027 forecast has to start from a measured baseline rather than a vendor claim. McKinsey’s 2026 State of AI survey, fielded across 1,719 respondents in 97 nations between May and June 2026, gives the clearest one.
Adoption is broad. Nearly nine in ten respondents report regular AI use in at least one business function, and 44% now say AI is scaling across the enterprise, up from 38% a year earlier.
Individual impact is real. Eight in ten say AI has improved their own productivity, and about half say it helps them make better decisions.
Enterprise impact is not following. About 37% report AI contributing positively to EBIT, essentially unchanged from the previous year despite more organisations scaling. The share qualifying as AI high performers, meaning those attributing 5% or more of EBIT to AI, sits at about 6% and has not moved.
That combination is the single most important fact for planning. More adoption has not produced more financial return, which means the constraint is not access to AI.
| Measure | Figure | Direction |
| Organisations using AI in at least one function | Nearly 9 in 10 | Rising |
| Scaling AI across the enterprise | 44% | Up from 38% |
| Individuals reporting improved productivity | 80% | High |
| Organisations reporting any EBIT impact | 37% | Unchanged |
| AI high performers (5%+ of EBIT from AI) | About 6% | Unchanged |
| High performers who redesigned workflows | Nearly 75% | Up from 55% |
| Other organisations who redesigned workflows | About 25% | Flat |
| Large organisations scaling agents | 40% | Up from 27% |
| Smaller organisations scaling agents | 22% | Flat |
| Constrained by AI operating costs | About 20% | Emerging |
Source: McKinsey State of AI 2026. The two rows to read together are the 80% and the 37%.
Five Things to Expect by 2027
1. Agent projects will fail at a high rate, and the reasons are known in advance
Gartner forecast in June 2025 that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. Gartner’s own framing is that many current projects are early-stage experiments driven by hype and often misapplied.
One finding in that release deserves more attention than the headline number. Gartner points to widespread “agent washing,” meaning the rebranding of existing assistants, RPA, and chatbots as agents without substantial agentic capability, and estimates that only around 130 of the thousands of agentic AI vendors are real. Its analysts add that many use cases positioned as agentic today do not require an agentic implementation at all.
The useful reading is not that agents do not work. It is that the failure causes are all scoping failures, and all three are visible before a line of code is written.
2. Agents will get more capable and more entangled
Gartner also predicted, in August 2025, that by 2027 one-third of agentic AI implementations will combine agents with different skills to handle complex tasks, moving past single-purpose agents. Capability rises, and so does the difficulty of working out what happened when something goes wrong.
That makes the autonomy question more consequential, not less. The distinction between a system that advises and a system that acts is the one that governs cost, risk, and debuggability, which is covered in our guide to AI agent vs chatbot vs workflow automation.
3. Software development changes shape
Nearly a third of organisations, 32%, already report deciding against buying at least one software product or feature because they could build it in-house with agentic coding tools. Gartner expects the shift to go further, predicting in May 2026 that by 2027 more than 65% of engineering teams using agentic coding will treat the IDE as optional.
For most businesses the practical consequence is that custom software gets cheaper relative to subscriptions. Processes that could not previously justify a build start to qualify, which changes the build-versus-buy line rather than eliminating it.
4. Cost becomes a design constraint
About 20% of organisations already report that AI operating costs, including token costs, are constraining their use. Meanwhile 60% expect to increase AI investment over the next year.
Those two facts together describe a market where spending rises and scrutiny rises with it. By 2027, expect cost per task to be a standard question in AI project approval, in the way cloud spend became one.
5. Workforce change will be slower than the forecasts
This is where predictions have already been tested. In 2025, 32% of respondents expected AI-related headcount declines over the following year. In 2026, only 14% reported that it actually happened. Expectations for the year ahead have risen again, to 39%.
The pattern is worth remembering when reading any 2027 workforce projection, including the ones in this article. Anticipated disruption has consistently run about double the realised disruption.
💡Get an AI Build Cost & Timeline Estimate
What Separates the 6% From Everyone Else
The most actionable finding in the McKinsey data is the profile of the high performers, because it describes behaviour rather than technology.
They redesign workflows instead of inserting AI into existing ones: Nearly three-quarters report fundamentally redesigning workflows because of AI, up from 55% the previous year. Among everyone else, just one-quarter report doing so. This is the largest single behavioural gap in the survey.
They aim past efficiency: Roughly 80% of both groups pursue efficiency gains, but most high performers also pursue growth or innovation, and they are 3.3 times more likely to intend to use AI to fundamentally transform the business within three years.
They measure: They are twice as likely to report both senior leadership commitment and defined processes for measuring the impact of AI initiatives.
They manage risk deliberately: They are much more likely to be actively mitigating risks such as unauthorised or unintended actions, which matters more as agents gain autonomy.
Note what is absent from that list. Nothing about model choice, vendor, or being early. The differentiator is organisational.
What Workflow Redesign Looks Like in Practice
The redesign finding is abstract until you see it applied, so here is the shape it takes.
AB Ark’s Eventas AI build is the clearest example in our own work. The starting point was manual coordination across disconnected steps, and the output was a self-operating ecosystem built around an AI Command Center rather than an assistant bolted onto the existing process. That distinction is the whole finding: the coordination work was not made faster, it was restructured so that most of it no longer needed a person.
The AI Receptionist Platform is the same principle in a narrower scope. Calls previously handled by whoever was available are answered continuously by a voice agent that books and resolves rather than drafting a response for someone else to send.
Both are examples of the pattern McKinsey’s high performers report and most other organisations do not: the process changed shape, rather than the same process running with AI assistance inside it.
How to Prepare Now
Five steps, ordered, each drawn from what the data says actually separates outcomes.
- Pick one process and redesign it, rather than adding AI to many: Workflow redesign is the strongest correlate of financial impact in the survey. Inserting AI into an unchanged process reliably produces the pattern most organisations are stuck in, where individuals feel faster and nothing shows up in the numbers.
- Name the metric before you build: Hours, error rate, resolution time, or cost per transaction. High performers are twice as likely to have defined measurement processes, and without a baseline you cannot tell a successful deployment from an expensive one.
- Model the cost per task, not the licence: With one in five organisations already constrained by operating costs, cost per completed task belongs in the design phase. Consumption-based AI pricing means a successful deployment costs more, which is the opposite of how software budgets usually behave.
- Choose the least autonomous system that solves the problem: Gartner’s cancellation forecast is driven by cost, unclear value, and weak risk controls, and over-scoping to an agent when a deterministic workflow would do creates all three at once.
- Prove it on a bounded scope first: A time-boxed AI POC with a defined success metric is how you find out whether the value is real before committing to a programme, and it is considerably cheaper than discovering it after cancellation.
Where This Leaves Smaller Organisations
One divide in the data is worth naming directly. Among organisations above $1 billion in revenue, the share scaling AI agents rose from 27% to 40% in a year. Among smaller organisations it stayed flat at 22%.
That gap is often read as a resource problem, but the high-performer profile suggests otherwise. Redesigning a workflow, defining a metric, and committing leadership attention are not capital-intensive activities, and a smaller organisation can change a process far faster than a large one can. The advantage available to smaller businesses is speed of change, not scale of spend, and falling build costs make bespoke systems reachable for processes that previously could not justify one. Our guide to bespoke software examples covers what that looks like in practice.
Frequently Asked Questions
How is AI being used in business right now?
Nearly nine in ten organisations report regular use in at least one function, most commonly chatbots, with 47% scaling those across the enterprise. Agent use is narrower, and concentrated in IT, knowledge management, and software engineering. About 44% of organisations now report scaling AI across the enterprise rather than in isolated pockets.
Is AI actually delivering return on investment?
Unevenly. About 80% of individuals report improved personal productivity, but only 37% of organisations report any EBIT contribution from AI, a figure unchanged year over year. Roughly 6% qualify as high performers attributing 5% or more of EBIT to AI, and that share has also stayed flat.
Why do AI projects fail?
Gartner attributes its forecast that over 40% of agentic AI projects will be cancelled by the end of 2027 to escalating costs, unclear business value, and inadequate risk controls. The McKinsey data points at a related cause: organisations that insert AI into existing workflows rather than redesigning them rarely see financial impact.
What should a business do to prepare for AI by 2027?
Redesign one process rather than adding AI across many, define the metric it should move before building, model cost per task rather than licence cost, choose the least autonomous system that solves the problem, and prove the value on a bounded scope before committing to a programme.
Will AI reduce headcount by 2027?
Expectations have consistently outrun reality. In 2025, 32% of respondents expected AI-related workforce declines within a year, while only 14% reported them actually occurring. For the year ahead, 39% expect declines and 43% expect little or no change, so treat workforce projections as directional rather than scheduled.
The Gap Is Organisational, Not Technical
By 2027 almost every business will have AI. The ones seeing it in their financials will be those that changed how work happens rather than which tools people use to do the same work.
If you can name one process, what it costs today, and what would count as success, that is enough to scope the first build properly rather than joining the 40% that get cancelled.
📞Schedule a Free Consultation Call
CEO At AB Ark Solutions
