Business process automation, usually shortened to BPA, means using software to carry out the routine steps of a business process, so that people deal only with the decisions and the exceptions. The software moves information from one step to the next, applies the fixed rules, sends the reminders and produces the documents. It works best on processes that are frequent, follow known rules and are already stable, and the sensible way to begin is with one process, measured before and after.
What business process automation means
A business process is a sequence of steps that starts with a trigger and ends with a result: an order arrives and an invoice goes out, a new member of staff is hired and is ready to work on day one. In most companies a good share of those steps involve no judgement. Somebody copies details from an email into a system, checks a figure against a limit, forwards a form to the right manager or chases a reply.
BPA hands those steps to software. The judgement stays with people: whether to accept an unusual order, what to do about a disputed invoice, how to handle the customer who is the exception to every rule.
Several terms are used alongside it.
- Workflow automation is the routing part: sending a task or an approval to the right person in the right order.
- Robotic process automation (RPA) is software that operates another program through its screens, clicking and typing as a person would. It is used when a system offers no better way in.
- System integration connects two systems so that data entered in one arrives in the other. Much automation depends on it, and we cover it in what is system integration.
- AI adds the ability to read unstructured material, such as an emailed order or a scanned invoice, and turn it into data the next step can use.
Processes that suit automation
These are illustrations of the kind found in small and mid-sized companies.
Approving supplier invoices. An invoice arrives by email. Its details are read, matched to the purchase order and sent to the budget holder, who approves it on a phone. The approved invoice is posted to the accounts package. People see only the invoices that do not match.
From quote to invoice. The details typed into a quote become the order, the job sheet and the invoice without being typed again.
Reminders and renewals. Certificates, contracts and inspections each have an expiry date. The system warns the responsible person in good time and escalates if nothing happens.
Routine reports. A report that someone assembles on the first Monday of each month from three exports and a spreadsheet is produced on a schedule from the same sources.
A process is a good candidate when most of these are true:
- it happens often, so the saving is repeated;
- the rules can be written down, including the rules for the common exceptions;
- the information arrives in a consistent form, or can be made to;
- the process is settled, and is not about to be redesigned;
- you can measure it today, in time taken and mistakes made.
A process is a poor candidate when it is rare, depends mostly on judgement, or does not work well as it stands. Automating a muddled process produces the same muddle faster. Put the process right first.
The ways to automate, simplest first
There is a ladder of options, and it is worth climbing it from the bottom.
| Approach | Suits | Watch for |
|---|---|---|
| Features already in your software | Approval steps, scheduled reports and alerts the product can do but nobody has switched on | Limits on how far the vendor’s rules can be bent to yours |
| Automation services such as Power Automate or Zapier | Simple flows between cloud products, built without programming | Flows that grow complicated and belong to one member of staff |
| Screen automation (RPA) | An older system with no other way to get data in or out | Breakage whenever the screens it drives are changed |
| AI document reading | Orders, invoices and forms arriving as emails or PDFs | Mistakes; a person must check anything uncertain |
| Bespoke software | Your own rules and exceptions, or a process that is central to the business | Higher initial cost, and it needs maintaining |
Microsoft’s Power Automate covers two of these rows: its cloud flows link online services, and its desktop flows are its form of RPA.
The lower rows are not better than the upper ones. A rule switched on in the accounts package you already pay for costs almost nothing and is supported by its vendor. Purpose-built software earns its cost when the process is specific to your business and the simpler tools have been outgrown, which often shows as a tangle of spreadsheets doing a system’s job. Where documents are the bottleneck, AI built into an existing system can do the reading while a person confirms the result.
What goes wrong
Silent failure. An automated step that stops working, with nothing to report the fact, is worse than the manual step it replaced. A person who cannot do their job says so. Every automation needs a log of what it did and an alert to a named person when it cannot proceed.
No owner. Automations built by one enthusiastic member of staff tend to be undocumented and tied to that person’s account. When they leave, the business discovers what it was relying on. Treat an automation as you would any other system: someone is responsible for it, and how it works is written down.
Exceptions nobody mentioned. The people who do the work today know the cases that do not fit: the customer invoiced in euros, the supplier who sends two invoices for one order. If they are not asked, the automation will meet those cases in its first week.
No record of decisions. If the system approves, rejects or prices something, it should record what it decided and on which rule, so that the decision can be explained later.
Personal data. A process that handles information about individuals, and especially one that makes decisions about them without a person involved, raises data protection questions. Check what the law requires of you and take advice before automating it.
Where to start
Pick one process. The best first choice is frequent, dull and well understood, and a mistake in it is cheap to catch.
- Write down what really happens. Follow the process step by step with the people who do it, noting who does what, how long it takes and where it goes wrong. The real process usually differs from the documented one.
- Measure it. Record how many times a week it runs, how long each run takes and how many errors are found. Without these figures you cannot tell afterwards whether the automation was worth it.
- Simplify before automating. Remove steps that exist for reasons nobody can remember.
- Automate the smallest useful part. One hand-off, such as moving approved orders into the accounts, is enough for a first step.
- Run it beside the manual process. For a few weeks, do both and compare the results. Stop the manual step when they agree.
- Compare with the figures from step 2, and decide whether the next process is worth doing.
The measuring can begin this week, without buying anything. Choose the process people complain about most, and for five working days count how often it runs and how long it takes. Those two numbers will tell you whether automation is worth pursuing, and they are the first thing any supplier should ask you for.