Choose a process with a visible beginning and end
“Improve operations” is too broad to investigate. “Reduce the time between receiving an enquiry and assigning an owner” is specific enough to observe. The same applies to campaign approvals, product listing updates and customer onboarding.
Pick a process the team performs regularly. Define its trigger, its final output and the person responsible for that output. Then collect a small sample of recent cases. Include ordinary work and the cases that became difficult.
My preference is to begin with evidence from the work itself: timestamps, task histories, correction requests and conversations with the people doing it. A polished process document can describe how the business hopes to operate rather than how it actually operates.
Map the handoffs, including the waiting
Write down every stage. Who receives the information? Where do they put it? What decision do they make? Which person or system receives it next? Mark each place where somebody copies information, waits for an answer or asks a question already answered elsewhere.
Separate working time from elapsed time. A proposal might take 40 minutes to prepare but remain unfinished for three days because pricing approval is unclear. Generating its first draft faster would not resolve that delay.
A simple audit table can use six columns: step, owner, input, tool, working time and waiting time. Add a note for the common exception. The objective is to make the bottleneck visible enough that the team agrees on what to change.
Decide what to remove, simplify or automate
Not every repeated task deserves software. A weekly report that nobody uses may need to stop. A request form that repeatedly collects incomplete information may need clearer questions. An approval chain may need a named decision-maker.
Once those issues are addressed, consider automation. Moving a validated enquiry into a CRM is a straightforward connection. Interpreting a long, unstructured request may benefit from AI. Deciding whether to accept a strategically important client may remain a human responsibility.
Assess each opportunity against frequency, effort, error cost, data readiness and ease of reversal. A modest improvement that happens every day can be a stronger first project than an impressive demonstration that rarely matches real work.
Check the information the workflow will depend on
Ask where the authoritative record lives. If a customer has three different names across three systems, connecting them automatically could spread the confusion. Establish identifiers, required fields and clear rules for updates before extending the process.
For AI tasks, collect examples of acceptable and unacceptable outputs. Determine what the model is allowed to read and who checks uncertain results. NIST’s AI Risk Management Framework provides a voluntary foundation for considering AI risks throughout design and use. See the NIST framework.
The practical implication I draw is to include quality and oversight in the original project brief. They should be part of the workflow’s design and workload estimate.
Turn the audit into a short implementation roadmap
For each shortlisted project, record the problem, proposed change, owner, dependencies and success measure. Estimate the effort to implement it and the ongoing time needed to maintain it. Mark assumptions that still need testing.
A useful first phase might standardise the enquiry form and ownership rules. The next phase could connect the form to the CRM. A later phase might add AI summaries after the business has enough examples to evaluate them.
Finish with a decision meeting, not another dashboard. Agree which project will proceed, which will wait and what evidence would change that order. A strong business process audit creates a shared picture of the work and a credible first step towards improving it.
What would this look like in your business?
Explore Analyse or tell me what you would like to improve.