Many AI projects begin in the wrong place.
They start with a tool, a trend, a demonstration, or a general desire to “use AI”.
That is understandable. AI tools are visible, impressive, and easy to experiment with. But for SMEs, the strongest opportunities usually do not begin with the tool.
They begin with friction.
The best AI opportunities are often found where manual work, repeated decisions, and disconnected systems already create drag.
That is where AI has a clearer job to do.
Friction shows where value is leaking
Operational friction is the resistance inside a business.
It appears in the places where work slows down, information gets copied, staff repeat themselves, reports take too long, customers wait, decisions are delayed, and systems fail to connect.
This friction can be easy to ignore because businesses often adapt around it.
A spreadsheet appears.
A manual checklist is created.
A team member becomes the person who “knows how it works”.
A report is rebuilt every month.
A process depends on someone remembering to follow up.
A customer update waits while staff check three different systems.
These workarounds keep the business moving, but they also reveal where value is leaking.
AI should be applied where that leakage is real.
Starting with tools creates scattered experiments
A tool-first approach often produces disconnected activity.
One team tests AI for writing. Another experiments with summaries. Someone tries a plugin. A manager asks whether a chatbot could help. A dashboard tool adds AI commentary. A few staff become enthusiastic while others ignore it.
None of this is necessarily bad.
But without a clear connection to operational friction, it rarely becomes a business capability.
The company may feel like it is experimenting with AI, but the underlying problems remain:
- reporting is still manual
- systems are still disconnected
- decisions are still delayed
- knowledge is still hard to find
- customers still wait
- staff still duplicate work
- leadership still lacks visibility
AI activity is not the same as AI value.
Friction gives AI a purpose
When a project starts with friction, the role of AI becomes clearer.
Instead of asking:
“What can AI do?”
the business asks:
“What is slowing us down, and how could intelligence improve the flow?”
That changes the quality of the project.
For example:
- If staff spend hours summarising documents, AI may help extract and summarise information.
- If reporting takes days to prepare, AI may help generate commentary from structured dashboards.
- If customer enquiries are slow to triage, AI may help classify, route, and draft responses.
- If internal knowledge is hard to find, AI may help create a searchable assistant over approved content.
- If systems do not connect, AI may only become useful after integration or custom software creates the right operating layer.
The value is clearer because the problem is clearer.
Look for repeated decisions
AI is often useful where people make repeated decisions using similar information.
This does not mean the decision should always be automated. Often, the best approach is decision support.
Repeated decisions might include:
- classifying customer enquiries
- prioritising support tickets
- assessing document completeness
- deciding what follow-up is needed
- identifying exceptions in reports
- routing work to the right team
- comparing cases against criteria
- checking whether information is missing
- summarising what a manager needs to review
These decisions may still need human oversight, but AI can help prepare, organise, summarise, or recommend the next step.
The return is strongest when the decision happens frequently and currently consumes time, attention, or consistency.
Look for manual information movement
Another strong signal is manual information movement.
This happens when staff move data between tools because systems do not connect.
Examples include:
- copying customer details from emails into a CRM
- exporting data to build reports
- updating job status in multiple places
- transferring invoice information between systems
- manually creating tasks from form submissions
- re-keying information from documents
- building spreadsheets because no system shows the full view
This is often not an AI problem by itself. It is a systems problem.
But AI may become valuable once the workflow is redesigned. It may help extract information, classify it, summarise it, check it, or trigger the next step.
The key is not to add AI on top of broken movement. The key is to improve the movement itself.
Look for reporting drag
Reporting is one of the clearest places to find operational friction.
In many SMEs, reports are created through manual exports, spreadsheet manipulation, copy-paste commentary, and individual knowledge.
Leadership may receive reports, but the process is slow and the interpretation depends on the person who prepared them.
This creates several problems:
- delayed visibility
- inconsistent definitions
- limited trust in the numbers
- too much manual effort
- weak accountability
- slow decision-making
AI can help with reporting, but only when the reporting structure is sound.
The strongest approach is usually:
- define the KPIs
- connect the data sources
- build the dashboard or reporting model
- automate refresh where possible
- use AI to summarise changes, flag exceptions, and support interpretation
That is very different from simply asking AI to explain a spreadsheet.
Look for knowledge bottlenecks
Growing businesses often rely on knowledge that is difficult to access.
Important information lives in:
- shared drives
- old emails
- policy documents
- project notes
- customer histories
- staff memory
- spreadsheets
- technical documents
- previous proposals
- supplier information
When staff cannot find what they need, they ask around, search manually, recreate work, or make decisions with incomplete context.
AI can create strong value here by helping people retrieve, summarise, and use internal knowledge more effectively.
But the content needs structure.
“AI activity is not the same as AI value.”
A useful knowledge assistant depends on approved sources, clear boundaries, and an understanding of what information should be trusted.
Again, the AI opportunity is found by looking at the friction first.
Friction does not always mean automation
One mistake is assuming every friction point should be automated.
Some friction exists because the workflow needs redesign.
Some exists because systems do not connect.
Some exists because reporting definitions are unclear.
Some exists because roles and ownership are ambiguous.
Some exists because the business lacks the right internal tool.
Some exists because judgement is required.
AI may be part of the answer, but not always the whole answer.
The right solution could be:
- automation
- reporting improvement
- system integration
- custom software
- AI assistance
- process redesign
- better data structure
- a new internal workflow
- a combination of these
This is why starting with friction is more useful than starting with AI. It keeps the solution open until the problem is understood.
How to map operational friction
A simple way to begin is to examine the business across five areas.
- Workflows
Where does work slow down, repeat, or rely on manual handoffs? - Systems
Where do tools operate in isolation or require duplicate entry? - Data
Where is information inconsistent, hard to access, or difficult to trust? - Reporting
Where does leadership lack timely, clear, decision-ready visibility? - Knowledge
Where do staff waste time searching, asking, or recreating information?
For each area, ask:
- How often does this happen?
- How much time does it consume?
- What errors or delays does it create?
- Who is affected?
- What would improve if this friction was reduced?
- Is the issue worth solving?
- Is AI, automation, reporting, integration, or custom software the right response?
This turns AI adoption from vague exploration into structured opportunity identification.
Prioritise by value, not excitement
The most valuable AI project may not be the most impressive demonstration.
It may be something practical:
- reducing manual report preparation
- automating document handling
- improving enquiry response
- connecting systems for better visibility
- helping staff find information faster
- generating consistent first drafts
- flagging exceptions in operational data
- supporting managers with decision-ready summaries
These may not sound futuristic, but they can produce clearer return because they sit close to real work.
The best AI projects usually have three qualities:
- the problem is frequent
- the value is visible
- the solution fits the operating environment
That is a better test than whether the technology feels exciting.
Final thought
AI projects should not start with the question:
“What should we do with AI?”
They should start with:
“Where is work harder, slower, or less visible than it needs to be?”
That is where the useful opportunities are.
For SMEs, the clearest AI value is often found in the friction the business already feels: repeated decisions, manual work, fragmented systems, slow reporting, and knowledge bottlenecks.
Start there, and AI becomes much easier to apply.
Not as a trend.
Not as a disconnected tool.
But as a way to improve how the business actually operates.