In this guide 6 sections
A process is worth automating when it repeats often, follows clear rules, passes through several hands and costs something when it goes wrong. Score each candidate on those four points, then check that its data can be reached. Start with the top scorer. Leave judgement-heavy work with people.
List the work before you judge it
Most teams pick a process because it annoys someone. That is a fair starting point, but it is not a method. Begin with a plain list of recurring work instead.
Ask each team lead to name what their people do every day, week and month. Include the small jobs. Copying a figure from one sheet to another counts. So does chasing a signature.
Keep the list short and factual. Write the name of the task, who does it and where it starts. Do not argue about value yet.
Score each process on four questions
Give every task on the list a score from one to three on each question below. A rough score is enough. The aim is to compare tasks with one another, not to measure them exactly.
A task that scores high on all four is a strong candidate. A task that scores high on only one is usually better left as it is, or fixed by a simple change of habit.
- Frequency: how often does the task repeat?
- Rules: can the steps be written down without saying it depends?
- Handoffs: how many people touch it before it ends?
- Cost of error: what happens if a step is missed or mistyped?
Check that the data can be reached
A task can score well and still be a poor candidate. Automation needs to read and write somewhere. If the inputs arrive as photographs of handwritten notes, the first job is capture, not automation.
Look at where each input lives. A system with an export or an interface is easy to connect. A paper register is not. A shared sheet that three people edit differently is somewhere in between.
Note the owner of each source too. Someone must agree to the change, and that person is often not the one doing the task.
A worked example with two candidates
Take a trading firm with two candidates. The first is monthly expense claims. Staff email photographs of bills, a manager approves, and accounts keys the amounts into the ledger. The second is a weekly stock count of a few hundred items.
Expense claims repeat monthly, follow clear policy rules, pass through three people and cause friction when a claim is lost. The data arrives by email, which can be read. It scores well.
The stock count repeats weekly and has clear rules, but one person does it alone and the cost of an error is small. It scores lower. The firm starts with expense claims and revisits the count later.
What to leave alone for now
Some work should stay with people. Tasks that need judgement, negotiation or a customer’s trust do not become better when a machine handles them. Automate the preparation around them instead.
Also leave out any process that is about to change. Automating a policy that will be rewritten next quarter only fixes the old way in place. Settle the process first, then automate it.
GullySystem’s Business Process Automation service starts with this kind of review, so the first build is a task that earns its place.
- Work that depends on judgement more than rules
- Processes the business plans to redesign soon
- Tasks done a few times a year
- Steps whose inputs are still on paper
Automation candidate scoring sheet
A single page listing tasks down the side. Columns cover owner, frequency, rules, handoffs, cost of error and data source. Each team lead fills in a score of one to three for every column. Add the scores, sort the tasks and discuss the top three together.
Open a blank worksheet to printQuestions owners ask
How many processes should we automate first?
One or two. A small first build lets the team learn how the work behaves in practice. Further processes can follow once the first one is running steadily.
Is a process with many exceptions a bad candidate?
Not always. If the exceptions can be listed, the rules can handle them or route them to a person. If nobody can describe them, map the process before deciding.
Who should do the scoring?
The people who do the work. A manager’s view of a task often differs from the daily reality. A short joint session usually settles the scores.
Do we need new software to start?
Not for the scoring. The review needs a list and a conversation. Software choices come after you know which task, which data and which owner.