When a business decides to automate, the conversation usually starts with tools: which platform, which AI model, which integration. Those decisions come later. The first question is what to automate, and answering it takes more work than most teams expect.
You cannot decide what to automate, how to automate it or when until that work is done, and most of what you need to know is not written down anywhere.
The Process Lives in Someone’s Head
Every repeatable task has an official version and a working version. The official version might be an SOP, a training guide or a flowchart from the last system rollout. The working version is what the employee actually does on a Tuesday morning when a customer file is missing a field and the person who approves it is out.
Ask how much of the SOP is really followed, and the honest answer is usually “parts of it.” People add checks after a past mistake, skip steps that stopped mattering and keep workarounds for gaps in the system. None of that is in the document.
Automate the document, and you automate a process nobody runs.
Interview More Than One Person
Talk to the people who do the work, and talk to more than one of them for the same task. Two employees processing the same invoice often follow different paths, and both believe theirs is the standard. Those differences are where you learn the most: the exception one person learned to catch, the shortcut another relies on, the approval a third never asks for.
Walk the process end to end with each of them, from the moment the work arrives to the moment it leaves their hands. Then write it down, step by step, in their words.
Verify Before You Build
A written process is still a draft. Cross-check it with other employees who do the same work, and with the business owner or manager accountable for the outcome. You need both views before the process is solid enough to automate.
Owners often know why a step exists. Employees know whether it still happens.
This is the same discipline we described in Before You Automate, Observe: understand the current state and its history before deciding what the future should look like.
Measure the Work
Once the steps are agreed, measure them:
- How long does the task take from start to finish, and how much of that time is waiting?
- How many times does it go back and forth for internal approvals?
- How much time goes into checking and re-checking data before it can be processed?
- How often does an exception send the work back to the start?
These numbers show where the time goes. Sometimes the slow part is the data entry. More often it is the waiting, the approvals and the rework around it.
Match Each Step to the Right Kind of Solution
A mapped and measured process usually breaks into three kinds of work.
Rules-based steps such as moving data between systems, formatting, standard calculations and routing can be automated with code. They are predictable, and once built they are inexpensive to run.
Judgment steps such as reading an unstructured document, classifying a request, drafting a response or choosing which exception path applies are where AI fits. It handles variation that fixed rules cannot, as long as someone defines what a good decision looks like and checks the results.
Accountable steps such as approving, signing off and sending the final deliverable to an internal or external customer stay with a person. This is the human in the loop: the system prepares the work, and a person makes the call and owns the outcome.
Most good designs use all three. Fully automating judgment and accountability usually creates risk the business never intended to take on.
Be Clear About What the Automation Is For
Cost narrows the options quickly. Building, licensing and maintaining automation costs money, and a solution that saves ten minutes a week will rarely pay for itself.
Cost savings are only one reason to automate, though. In some cases the goal is time: getting a quote, an approval or a deliverable to the customer faster. In others it is accuracy: removing the re-keying errors that create rework downstream. Decide which one you are buying, because it changes which steps are worth automating and how you will measure success.
What It Gives Back
Automation is challenging work, and done well it raises productivity, accuracy and profitability. The larger return is time. The people who understand your business best spend less of their day on data checks and approval chasing, and more on the work that needs them: building relationships, meeting clients and presenting outcomes and deliverables.
At Strategy67, every automation engagement starts by interviewing the people who do the work, writing the real process down, verifying it and measuring it, before anyone chooses a tool.