AI adoption for SMEs starts with process, not tools

← Back to all articles

5 February 2026 · Nathan Jones

AI is most useful when it supports how work already gets done.

So rather than beginning with "Which AI tool should we buy?", a more reliable starting point is:

Pick one workflow. Define the outcome. Then choose where AI supports it.

That business-first approach is echoed in management research. HBR discusses the risks of "AI-first" thinking and why strategy begins with the problem, not the tool: Is Your AI-First Strategy Causing More Problems Than It's Solving?

Plain-English definitions

Workflow

The steps that turn an input into an output (enquiry → proposal → delivery → invoice → support).

Friction

Where the workflow slows down, needs rework, or creates risk.

Outcome

What "better" looks like in measurable terms (faster turnaround, fewer errors, higher conversion, lower cost, lower risk).

AI adoption

A change in how work happens, so the business gets a repeatable benefit.

McKinsey's guidance on genAI and change management leans on this idea: adoption is shaped by outcomes and ways of working, not simply tool access: Reconfiguring work: Change management in the age of gen AI

A simple first-principles view

What AI is good for in day-to-day SME work

AI tends to help most when work includes:

• searching and synthesising information
• drafting and refining text
• summarising long threads and meetings
• classifying, tagging, extracting
• producing a strong first draft that a person reviews

McKinsey's research on generative AI frames value in terms of work activities and functions: The economic potential of generative AI

The levers SMEs can pull quickly

1. Make the workflow visible (one page is enough)
2. Break it into tasks (repeatable vs judgement)
3. Add guardrails (data rules + review habits)
4. Build adoption into the week (simple routines)

MIT Sloan's guidance on selecting genAI use cases is aligned with this: break down workflows into tasks, consider costs, then launch pilots: How to find the right business use cases for generative AI

The Process-First Start method (run this in under an hour)

Step 1: Choose one workflow that matters

Pick something frequent and meaningful. For many SMEs, good candidates include:

• handling inbound enquiries and creating proposals
• customer support triage and responses
• onboarding new clients
• monthly reporting packs
• recruitment screening and shortlisting

Step 2: Map it in 12 minutes

On one page, list the steps and handoffs.

Step 3: Price the friction (quick baseline)

For each step, estimate:

• minutes per item
• volume per week
• rework rate (rough %)
• cost of error (low / medium / high)

Step 4: Spot "AI-shaped tasks"

Look for tasks that are:

• text-heavy
• repeatable
• dependent on finding and combining information
• easy to review quickly

Step 5: Write the outcome in one sentence

Keep it human and measurable.

McKinsey recommends crafting a North Star based on outcomes when approaching genAI-enabled change: Reconfiguring work: Change management in the age of gen AI

A practical operating rule

A simple rule that works well early on:

AI drafts. People decide.

HBR on adoption also points towards product-minded habits (define value, test, measure, iterate): To Drive AI Adoption, Build Your Team's Product Management Skills

The Elansio buckets you'll see throughout this series

A note on pace (and why small starts compound)

MIT Sloan explores how payoffs build through complementary changes: skills, processes, supporting technology, and infrastructure: Artificial intelligence pays off when businesses go all in

For SMEs, the sweet spot is:

• small enough to move quickly
• meaningful enough to keep using
• measurable enough to learn from

Copy/paste template: Workflow Inventory

WORKFLOW INVENTORY (Process-first AI adoption)

Workflow name:
Owner:
Teams involved:
Volume (per week):
Current cycle time:
Primary outcome to improve (choose 1): Speed / Quality / Cost / Risk / Growth

STEPS (today)
1)
2)
3)
...

FRICTION POINTS (where work gets sticky)
- Step #:
- What happens:
- Why it happens (missing info / handoff / unclear rules / rework / approvals):
- Impact (time / cost / risk):
- Frequency (% of cases):

TASK BREAKDOWN (mark each step)
For each step, mark:
- R = Repeatable (rules-based)
- J = Judgement (needs human decision)
- S = Sensitive (customer/confidential)

AI OPPORTUNITIES (draft → review)
Opportunity:
- Task(s) it supports:
- Expected benefit:
- Review required? (Yes/No)
- Data allowed (Public / Internal / Confidential):
- Quality checks to apply:

MEASURES (2-3 only)
- Time saved per item:
- Error/rework rate:
- Customer impact metric (CSAT / response time / win rate):

NEXT ACTION (10 minutes)
What we will test this week:
Who owns it:
When:

The 10-minute action

Today, pick one workflow and write:

• the steps
• the two biggest friction points
• one outcome sentence