Stop Automating Tasks. Rebuild the Workflow.
Quick answer
BCG and MIT Sloan research shows firms winning with AI agents stopped automating individual steps and redesigned whole workflows around human-AI collaboration instead. Here's what UK SMEs should do about it.
BCG and MIT Sloan research shows firms winning with AI agents stopped automating individual steps and redesigned whole workflows around human-AI collaboration instead. Here's what UK SMEs should do about it.
I see this all the time with UK SMEs. Someone gets excited about AI, picks one painful step — usually chasing invoice approvals or rekeying purchase orders — and automates just that bit. Three months later the step is faster, but the process around it is still a proper nightmare. Bottlenecks just moved down the line.
You've sped up one link in a chain and wondered why the chain didn't get faster. I've seen it dozens of times, and it's not a technology problem. It's a design problem.
What the research actually says
The joint BCG and MIT Sloan Management Review report, The Emerging Agentic Enterprise, makes the point better than I can. Their headline finding is that 76% of executives now view agentic AI more as a coworker than a tool, and 35% of companies have already started putting agents into work that used to be done by people.
But the bit that really got my attention was about work design. The report says the winning organisations have shifted the question they ask. Instead of "where can we automate a step?" they ask "which processes should be rebuilt around human–AI collaboration?"
Stop speeding up broken steps. Design the work around who — or what — does it best.
That's a business analyst's job, not a software purchase. And it maps exactly to what the enterprise data shows: only 31% of organisations have an agent in production, but multi-agent orchestration has grown from 1% to 22% in two years. The firms that get there aren't automating steps — they're redesigning workflows.
Why this matters for smaller firms
In a 60-person manufacturer I worked with, purchase order approvals went through four people, two emails and a spreadsheet. They could have automated the spreadsheet bit and saved ten minutes a week. Instead we mapped the whole flow, found two approval steps that added nothing, and rebuilt it around one person plus an exception agent. Saved them the best part of two days a week across the team.
That's the difference between task automation and workflow redesign. One shaves seconds. The other removes whole chunks of work — including work you didn't realise was optional.
What to actually do
- Map one workflow end to end before touching AI. Not the whole business. One flow — invoicing, PO approvals, stock reconciliation. Get it on paper and mark every handoff, check and wait.
- Kill the steps that add nothing first. The cheapest automation is deleting a step nobody needed. Do that before you buy anything.
- Then design the AI in, around a clean data flow. Agents need reliable inputs. If the ERP data's a mess, sort that first or the agent will just make confident wrong decisions faster.
Notice none of that starts with buying software. It starts with understanding how the work actually flows — which is exactly the bit most firms skip.
If you're eyeing up AI agents but haven't mapped the workflow they'd live in, you're putting a fast engine in a car with square wheels. Sort the wheels first.
References & Further Reading
- BCG / MIT Sloan Management Review, The Emerging Agentic Enterprise: How Leaders Must Navigate a New Age of AI
- BCG, Agentic AI Blurs Line Between Tool and Teammate
- S&P Global Market Intelligence / McKinsey (via Digital Applied), AI Agent Adoption 2026: 120+ Enterprise Data Points
- Anthropic, Building Effective AI Agents
Rethinking a workflow before automating it?
I help UK SMEs map, simplify and rebuild workflows before layering AI on top. Workflow analysis, data readiness, the lot.
Let's have a chatMatty Hatton is the founder of Digital Adaption, an ERP and data consultancy based on the Wirral. He has spent 15 years delivering ERP transformations for manufacturers, including leading the data migration on a £4.5m consolidation of four legacy systems onto a single Infor LN cloud instance for a 220-user group. He holds an MSc in Digital Transformation and IT Strategy from Manchester Metropolitan University and is Microsoft PL-200 certified.
Start with a 30-minute data risk call