Operations teams can use AI to find and fix bottlenecks by working in four steps: describe the process as it really runs, ask AI to point out repeated steps, hand-offs and waiting points, draft the improved procedure, and only then decide what to automate. AI is good at organising messy descriptions and spotting patterns in text. It is not a replacement for walking the floor or asking the people who do the work.
Start by mapping the process as it actually runs
Most bottlenecks hide in the gap between the documented process and the real one. Before you open any AI tool, collect the raw material: the existing SOP if there is one, a few real examples (an order, a purchase request, a claim), and short notes from the people who do each step.
Then ask an assistant such as Claude to turn those notes into a numbered step list, with the owner of each step and the point where work is handed over. Ask it to mark anything it had to guess. Those guesses are exactly the questions to take back to your team.
A useful prompt structure: say what the process is, who is involved, what triggers it and what "done" looks like, then paste the notes. Our guide on writing a good prompt for Claude covers this structure in more detail.
Ask AI to spot repeated steps, hand-offs and waiting time
Once the steps are written down, ask the assistant to review them against a short checklist:
- Repeated work. Is the same information typed, checked or copied more than once?
- Hand-offs. Where does work move between people or departments, and what can get lost there?
- Waiting. Which steps depend on an approval or reply that has no deadline?
- Rework. Which steps exist only to fix errors made earlier?
- Unclear ownership. Which step has no single named owner?
Treat the answer as a list of hypotheses, not findings. AI works only from what you gave it, so it cannot tell you how long a step truly takes. Check each hypothesis against real timestamps, tickets or a short observation exercise before changing anything.
Draft the improved procedure and the SOP together
When you agree on a change, use AI to produce the first draft of the new procedure: purpose, steps, owners, exceptions and a short checklist. Writing the SOP while you design the change keeps documentation from becoming an afterthought. Our article on using AI to draft SOPs walks through this, including how to have the people who do the work review it.
Keep a human reviewer on every draft. Ask them to run the procedure on one real example and note where the draft does not match reality.
Decide what to automate first
Not every bottleneck should be automated. A simple filter helps you choose where to start:
- Frequent and rule-based. Steps that repeat often and follow clear rules are the best candidates.
- Low consequence if wrong. Start where an error is easy to catch and cheap to fix.
- Clean inputs. If the incoming information is inconsistent, fix that first.
- Clear owner. Someone must be accountable for the result, whatever tool produced it.
Steps involving judgement, sensitive personal data or decisions that affect people deserve more caution. Read which tasks you should not automate with AI before you commit, and keep a human in the loop wherever the cost of a mistake is high.
Run it as a small pilot, then measure
Pick one process, one team and a short trial period. Agree beforehand what you will compare: time from start to finish, number of corrections, or number of follow-up questions. Use the numbers from your own baseline, not generic figures from elsewhere. If the pilot helps, document the new way of working and extend it. If it does not, you have learned something cheaply. Our guide to running an AI pilot sets out the steps.
Protect company data while you do this
Process notes often contain customer details, supplier terms or staff information. Remove or mask what the task does not need, and follow your company's AI usage policy. Check what you may share with our article on customer data and Claude, and confirm anything uncertain with your IT or compliance team.
Build the skill across the operations team
The method works best when several people in the team can do it, not just one enthusiast. Agmo Studio, a Select Partner of the Anthropic Claude Partner Network, runs hands-on workshops through Agmo Academy, and the Claude Certified Associate workshop is a practical starting point for non-technical staff. For team-wide programmes, see AI training in Malaysia. HRD Corp claimability is subject to eligibility and HRD Corp approval.