Workflow automation for leaders: where AI fits in your processes

What it is

Workflow automation is software carrying out the steps of a business process, such as moving an invoice from inbox to approval to payment. Three kinds, often mixed:

Rules-based automation follows fixed steps ("when a form arrives, copy it to this sheet"): predictable and cheap, but it breaks when inputs vary.

AI steps add judgement at one point in a fixed flow: sort an email, read a scanned form, draft a reply.

Agents get a goal, tools and limits, and choose their own steps (see guide 17): more flexible, harder to predict and check.

Think of a kitchen: a recipe card, a cook judging the fish, a cook told "make lunch for twelve".

Main platforms, as their makers describe them (as of November 2026; check the maker's page):

All of them need the same governance (owner, approvals, logs); none removes the need for a human check.

Why it matters

• Whole processes are where hours given back add up (see guide 28).
• A poorly chosen process automates the wrong thing faster.
• Nobody watches each run, so a failure can repeat unnoticed; each automation also holds access to data and systems.

How to do it

1. Score candidate processes on volume, clarity of rules, number of exceptions and cost of a mistake.
2. Start where volume is high, rules clear and errors cheap to catch; leave rare or high-stakes work with people.
3. Use the simplest kind that works: rules, then an AI step, an agent only where steps truly vary (see guide 16).
4. Place people at approval points: before money moves, anything reaches a customer or records change.
5. Name one owner per automation who reviews it and can switch it off.
6. Measure time per case before and after, including exceptions and checking.
7. Govern: an inventory entry, run logs, a tested kill switch; review each quarter.

Example

A leadership team scores three processes. Supplier invoices (high volume, clear rules): rules route them, an AI step reads fields, a clerk approves mismatches and every payment. Complaints (many exceptions): AI drafts, a person sends. Pricing changes (rare, costly errors): not automated. The invoice inventory entry:

Automation: [supplier invoice intake], rules + AI step
Owner: [name, role]
Data and systems: [inbox, finance system, read and draft only]
Human approval: [role] approves mismatches and all payments
Logs: [where]; kill switch: [how, who]; review: [date]
Measured: [minutes per invoice before / after]

Common mistakes

• Silent failures. Set alerts for failed runs and sample outputs weekly.
• Too much access. Prefer read and draft over send and delete (see guide 26).
• Sprawl. No inventory entry, no go-live.
• Automating a broken process. Fix the steps first; report measured hours.

Try it this week

1. With two colleagues, list five weekly processes (5 minutes).
2. Score each on volume, rules, exceptions and cost of error (5 minutes).
3. Pick one; write where a person approves and who owns it (5 minutes).

Going further

• Ask IT for automations already running, including those built outside IT.
• The US NIST AI Risk Management Framework playbook (voluntary) suggests an inventory of AI systems, someone responsible for it, and a way to retire systems safely.
• Tools close (OpenAI's Agent Builder shuts on 30 November 2026): write processes and approvals down outside any tool.
• Re-run a small test set whenever a model or flow changes (see guide 27).

Key words

• Rules-based automation: software following fixed steps set in advance.
• AI step: one point in a flow where AI reads, sorts or drafts.
• Agent: AI given a goal, tools and limits that chooses its steps.
• Approval point: where a named person must say yes before the flow goes on.
• Inventory: a list of every automation with owner, data, approvals and logs.
• Kill switch: a tested way to pause an automation at once.

Sources

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (284 KB), or the complete AI Field Guide (2.5 MB).