- Choose processes by four tests: volume, clarity of rules, number of exceptions and cost of a mistake.
- Use the simplest kind of automation that works: fixed rules first, an AI step for judgement, an agent only where steps truly vary.
- Put people at the approval points and write down who approves.
- Every automation needs an owner, a log of what it did and a tested way to switch it off.
- Report measured hours given back, including time spent on exceptions.
On this page:Three kinds of automationPick the first processWhere people approveGovern from day oneMeasure the value honestly
Three kinds of automation
Rules-based automation follows fixed steps: when a form arrives, copy it to a sheet and notify a person. It is predictable but breaks when inputs vary. An AI step adds judgement at one point in a fixed flow, such as reading a scanned invoice or drafting a reply. An agent is given a goal, tools and limits, and chooses its own steps. Agents handle more variety and are harder to predict and check. Field Guide 46 compares the main platforms as their makers describe them.
Which recurring process in your organisation would you trust to automate first, and who would own it? Book a free AI Value Briefing and we will look at it with you.
Pick the first process
Ask four questions about each candidate:
- Volume: how often does it run, and for how many people?
- Rules: could a new colleague follow it from a written checklist?
- Exceptions: how often does a case not fit the checklist?
- Cost of error: what happens if one case goes wrong, and would anyone notice in time?
Good first candidates are frequent, rule-bound and low-stakes: sorting incoming requests, extracting fields from standard documents, preparing routine reports for a person to check. Leave rare, judgement-heavy or high-stakes decisions, such as pricing or anything affecting individuals' rights, with people for now. And fix a broken process before automating it, or you will only get the wrong result faster. Field Guide 16 explains how to use what you already have before building.
Where people approve
Automations run without anyone watching each case, so decide in advance where a person must say yes. Typical points are before money moves, before anything reaches a customer or the public, and before records are changed or deleted. Write the approver's role into the design. Give each automation only the data and actions it needs, and prefer read and draft over send and delete.
Govern from day one
Automations multiply quietly when every team builds its own. The NIST AI Risk Management Framework playbook, a US government guide, recommends an inventory of AI systems with a named person or team responsible for maintaining it, clear human roles for oversight, and a deliberate process for retiring systems. In practice, for each automation keep:
- what it does and which data and systems it touches;
- one named owner who reviews it and can switch it off;
- a log of every run, and alerts when a run fails;
- a tested way to pause it at once;
- a review date, at least quarterly.
Silent failures are a common risk: a flow that stops, or produces wrong output, can repeat many times before anyone notices. Sample outputs weekly.
Measure the value honestly
Time a handful of cases before and after, including the time people spend on exceptions and checking. Multiply by how often the process runs to get hours given back, and label each figure as measured or estimated (Field Guide 28). What the freed hours are used for is a leadership decision. Field Guide 32 covers how leaders set this up across teams.
What to do next
- With your leadership team, list ten recurring processes and score each on volume, rules, exceptions and cost of error.
- Choose one frequent, low-stakes process, name its owner and write down where a person approves.
- Start an inventory of every automation already running, with owner, data, logs and a way to switch it off.
Goes deeper in our free AI Field Guide: 16. Choosing the right level: use what you have before you build, 28. Measuring the value: hours given back, 32. Leading AI adoption: what leaders do differently, 46. Workflow automation for leaders: where AI fits in your processes.
Talk it through with us
Your leadership team leaves with a short list of processes to automate first, owners and approval points agreed, and hours given back measured against a baseline, built by your own people with tools and data rules agreed with your IT team.
Book my AI Value BriefingA conversation, not a pitch. We will not ask for confidential data. How we handle your details.How this page was made: researched and drafted with AI assistance from the sources below, each opened and checked on the date shown, and reviewed against our editorial rules.
