Key takeaways
- Automate a process that is repetitive, measurable and based on available data.
- Organise exceptions and responsibilities first, then choose the tool.
- AI supports classification, search and content creation, but requires human oversight.
- The implementation should have an owner, a test and a success criterion.
How do you choose a suitable process?
A good candidate is a task performed frequently, according to repeatable rules, which consumes the team's time. Measure the number of operations, the time involved, the number of errors and the cost of manual handling. Do not start with the tool – start with the problem.
Automation
Rules and workflows for repetitive activities.
Integration
Transferring data between systems and APIs.
AI
Working with text, documents and classification when the rules are less clear-cut.
Available solution options
A simple workflow in Microsoft 365 or Power Automate works well for documents and notifications. API integration is better when data needs to move between ERP, e-commerce and accounting systems. AI is worth using where content needs to be understood, but the output should be verifiable.
Risks and security
Restrict the automation's access to the data it needs, log its actions and prepare error handling. Do not provide confidential data to a model without assessing the supplier and its data retention rules. Retain human approval for decisions that affect customers.
Automation will not fix a poor process
If data is incomplete or responsibility is unclear, a workflow will simply replicate the problem faster.
What determines the cost?
Relevant factors include the number of systems, data quality, the number of exceptions, licences, integrations, maintenance and the testing approach. Include the cost of monitoring and of changes required after updates to source applications.
Action plan
- Describe the process. Record the input, steps, exceptions and outcome.
- Measure the current state. Time, errors and number of cases.
- Choose a small pilot. Limit the scope and data.
- Establish controls. Logs, an owner and the ability to take over manually.
- Measure the result. Compare time, quality and cost.
- Document maintenance. Record dependencies and the response procedure.
Checklist
- The process has an owner.
- We know the inputs and exceptions.
- A success measure has been agreed.
- Access is restricted.
- Errors generate an alert.
- It is possible to return to manual handling.
Frequently asked questions
Is AI always necessary?
No. If a process has clear rules, a simple workflow may be cheaper and easier to control.
Does automation replace an employee?
It most often takes over repetitive activities, leaving decisions and exception handling to people.
How can AI be tested safely?
On a limited dataset, with anonymisation, output logging and manual approval.
Do you have a process that takes up too much of your team's time?
We can assess the opportunities for automation and recommend a safe first step.
Let's talk about your IT