Where should businesses actually start with AI?
Focus on repetitive decisions, information-heavy processes and areas where employees spend time searching instead of acting.
Read article →Many organisations are asking the same question: where should we use AI? The better question is often simpler — where does work currently require too much searching, reading, checking, repeating or waiting?
Start with the work, not with AI.
It is easy for an AI initiative to begin with technology.
Which model should we use? Should we build a chatbot? Do we need an AI agent? Should we introduce Copilot?
These are useful questions later.
They should rarely be the first questions.
The starting point should be understanding how work is actually performed today.
Look at where employees repeatedly search for information, read large volumes of documents, copy information between systems, prepare similar responses, classify requests, review exceptions or make routine decisions.
Good AI opportunities usually become visible when you study the friction in the process before studying the technology.
This changes AI from an abstract innovation programme into a practical business-improvement exercise.
Look for repetitive decisions and information-heavy work.
Not every repetitive task requires AI.
If a process simply moves structured information from one system to another, conventional automation may be the better solution.
AI becomes especially useful when the process contains information that needs to be interpreted.
Finding information
Employees spend time searching policies, procedures, contracts, knowledge bases or historical records.
Understanding documents
Teams repeatedly read emails, forms, proposals, invoices or support requests before deciding what to do.
Repetitive decisions
People apply similar judgement to classify, prioritise, route or recommend the next action.
Repetitive content
Teams prepare similar summaries, responses, reports or communications from existing business information.
If employees spend more time finding and interpreting information than taking action, there may be a useful AI opportunity.
Decide what role AI should actually play.
AI does not need to replace an entire process.
In many useful implementations, AI handles one specific part of the workflow while people and conventional automation continue to manage the rest.
A useful way to think about AI is to give it a clearly defined role.
Retrieve
Find the relevant information from approved business knowledge or systems.
Understand
Interpret unstructured documents, messages, requests or other business information.
Recommend
Suggest a classification, next action, response or decision for a user to review.
Create
Generate a draft summary, response, report or structured output.
Act
Trigger an approved business action through workflow automation or connected systems.
The more clearly this role is defined, the easier the solution is to design, test and govern.
Keep people in control where judgement matters.
One of the most practical ways to introduce AI is to use it as a decision-support layer rather than immediately giving it complete autonomy.
AI can prepare the work. People can make the final decision.
Support request classification
Instead of asking AI to resolve every incoming support ticket automatically, it could first read the request, identify the likely category, suggest priority, locate relevant knowledge and prepare a recommended response.
The support employee then reviews the recommendation and takes the appropriate action.
This type of implementation can reduce repetitive work while preserving accountability and business judgement.
Automation should remove unnecessary work. It should not remove necessary judgement.
Build the foundations before increasing autonomy.
AI is only as useful as the business environment around it.
A powerful model cannot compensate for information that is outdated, inaccessible, poorly structured or inconsistently managed.
Before increasing the role of AI, organisations should understand a few fundamentals.
- Which information is authoritative?
- Who is allowed to access it?
- How current and reliable is the information?
- Which systems should AI be permitted to use?
- Which actions require human approval?
- How will AI-generated outcomes be reviewed?
- What should happen when confidence is low?
These questions are not barriers to AI.
They are what make AI useful in a real business environment.
Pilot narrowly. Measure clearly. Scale intentionally.
The first AI project does not need to transform the entire organisation.
A focused use case can often provide more learning than a large programme with an unclear outcome.
Choose one workflow
Select a process with a clear owner, measurable friction and enough repetition to evaluate the impact.
Establish the baseline
Understand how long the process takes today, where effort is spent and where errors occur.
Introduce AI narrowly
Give AI a specific role rather than attempting to automate the entire process immediately.
Measure the outcome
Compare time saved, quality, adoption, accuracy and business outcomes against the original baseline.
Expand what works
Improve the solution, strengthen controls and extend it to adjacent processes only when the value is demonstrated.
The best place to start with AI may already be visible.
It may be the employee who opens five systems before answering a customer question.
It may be the manager who reads dozens of updates to understand what needs attention.
It may be the operations team manually classifying requests before routing them.
Or it may be the salesperson spending more time searching CRM notes, emails and documents than speaking with customers.
These are often better starting points than asking the organisation to “find an AI use case.”
Find where information slows people down. Give AI a clear role. Keep the right controls. Measure whether work actually improves.
AI becomes valuable when it is connected to a real business process, trusted information and a measurable outcome.
That is where businesses should start.
Find the right business process before choosing the AI solution.
Epicarp helps organisations identify practical AI and automation opportunities, connect them to existing business processes and move from experimentation to measurable operational value.