Business automation / Field guide

When is an AI workflow worth automating?

Good automation candidates have frequent, measurable tasks and reliable access to the systems involved.

Start with the current workflow

Choose a task performed repeatedly: inventory checks, sales reports, document intake, support routing, invoice reconciliation or appointment follow-up. Document who performs it, which systems are used, how long it takes and what happens when it goes wrong.

Separate rules from judgment

Many workflows do not need AI. A stable rule-based system is often faster, cheaper and easier to test. Consider a language model when the work genuinely involves interpreting varied text, understanding requests, summarizing information or making constrained recommendations.

For high-impact actions, keep decision authority with verified rules or people even if the model prepares a recommendation.

Calculate a conservative business case

A useful initial calculation is: monthly hours avoided × realistic loaded hourly cost, minus software costs, maintenance and required human review. Then account for quality changes, integration effort and operational risks. Do not count every theoretical hour saved as cash savings.

For example, if a recurring workflow requires 20 staff-hours a month, the project has a possible productivity value. Whether that value justifies development depends on labor cost, actual time recovered, error handling and ongoing support.

Verify your data access first

Some platforms provide official APIs; others restrict access or offer only exports. Confirm the POS, CRM or ERP vendor's supported options before committing to architecture or price. Unauthorized scraping and brittle workarounds create avoidable operational risk.

Make the pilot intentionally small

• Choose one process and one accountable owner.
• Measure the baseline before automating.
• Check integrations and data quality early.
• Set measurable success and safety criteria.
• Run the pilot alongside the current workflow.
• Expand only after the economics and reliability are demonstrated.

A focused automation pilot should demonstrate reliable operation and measurable value before expansion.