Business process automation

Why it matters

A growing business tends to accumulate processes that made sense at five staff and creak at fifteen: re-keying orders into the accounting system, chasing approvals by text message, copying details from one tool into another. Each is small; together they consume a surprising share of payroll and generate the errors that annoy customers. Automation done well turns that administrative drag into capacity for growth without adding headcount. Done badly, it hard-wires bad processes, hides errors inside systems nobody watches, and leaves the business dependent on automations nobody understands.

Find the processes that are worth it

Not every process deserves automation. The best candidates are frequent, rule-bound, time-consuming and error-prone, and they usually involve moving the same information between people or systems. Classic small-business examples include quote-to-invoice, new customer onboarding, purchase order to payment, staff rostering and leave, and the recurring reports an owner wants every Monday.

A simple two-by-two grid helps. Plot each candidate by how much value automating it would release (hours, errors avoided, faster cash) against how much effort it would take (number of systems, exceptions, data quality). Start in the high-value, low-effort corner. The high-value, high-effort corner is where custom builds and AI justify their cost later. Low-value processes are left alone, however tempting the tool demo.

Fix the process before you automate it

The most expensive automation mistake is to automate what people do today without asking whether it should be done at all. Processes grow steps the way sheds grow clutter: an approval added after one incident years ago, a report nobody reads, a form that duplicates a system record. Automating these preserves them forever and makes them harder to remove.

Map the process as it actually happens, not as the manual says. Walk through it with the people who do it. For each step ask: who needs this, what happens if we skip it, and does the information already exist somewhere? Remove or merge steps first. Standardise the inputs, because automation that receives orders by email, text, phone and a web form in six different formats will fail on half of them. Only then decide what to automate.

• Map the real process with the people who run it.
• Remove steps whose only justification is "we have always done it".
• Standardise inputs: one form, one format, one entry point where possible.
• Define the exceptions explicitly, because they are what automation struggles with.

Rules, AI, or both

Most process automation is rules-based and should stay that way. When a new customer is created, send the welcome pack. When an invoice is paid, update the job status. When stock falls below a level, raise a purchase order. These are predictable, auditable and cheap, and they are the fixed steps of nearly every process. Robotic process automation (RPA) tools extend this by mimicking a person clicking through screens of older software that has no proper integration.

AI earns its place at the points where input is unstructured or judgement is needed: reading the attachment on an emailed order and extracting the line items, classifying an incoming message as a complaint, quote request or spam, summarising a long customer history for the person who picks up the phone. A well-designed automated process uses rules for the backbone and AI at these specific joints, with exceptions and anything high-stakes routed to a human. Using AI for a step that rules could handle adds cost and unpredictability for no benefit.

Own it, monitor it, keep it current

Automation is infrastructure. It needs an owner who understands what it does, a way to see when it fails, and a routine for updating it when the business changes. Silent failure is the specific danger: a broken automation does not complain, it simply stops sending the welcome emails or reconciling the payments, and nobody notices for a month.

Build in alerts for failures and for volumes that look wrong. Keep a plain-language description of each automated process, including the exceptions it routes to people. Review quarterly. For businesses running many automations and AI components, this operational discipline is what the industry calls AIOps, and it is where SMBH Consulting’s operations practice spends most of its time: not building the first automation, but keeping a growing set of them reliable.

How to automate a business process

1. List candidate processes Frequent, rule-bound, time-consuming, error-prone. Ask the team where the re-keying happens.
2. Place each on the value-effort grid Start with high value and low effort.
3. Map the real process Every step, hand-off, system and exception, as it actually happens.
4. Simplify and standardise Remove unneeded steps and settle on one input format.
5. Assign rules or AI to each step Rules for fixed logic; AI for unstructured input; humans for exceptions and high stakes.
6. Build, test with real cases, then go live Run a parallel period where the automation and the person both do the work.
7. Monitor and review Alerts on failures, a named owner, a quarterly review.

Real-world example

A Perth wholesale bakery automates order-to-invoice

Swan Valley Bakehouse supplies bread to about ninety cafés and grocers. Orders arrive by text, email, phone and a web form, and two staff spend each afternoon keying them into the production sheet and then into the accounting system, with regular errors in quantities and delivery days. The owner, Hamid, maps the process and finds eleven steps, three of which exist only because of a dispute two years ago.

He first standardises intake: customers are moved to the web form or a simple ordering app, with phone orders keyed by staff into the same form. That alone removes most errors. Rules-based automation then creates the production sheet and draft invoices from the form data each evening. For the dozen customers who insist on emailing free-text orders, an AI extraction step reads the email, proposes line items and flags anything ambiguous for a person to confirm. After two months, the afternoon data-entry shift has gone, invoice errors have dropped sharply, and the two staff now handle customer calls and new-account onboarding instead.

Common mistakes

• Automating the process as-is Unnecessary steps become permanent and invisible. Simplify first.
• Using AI where rules would do AI adds cost and unpredictability; reserve it for unstructured input and judgement.
• Ignoring exceptions The unusual cases are what break automation. Define them and route them to people on purpose.
• No failure alerts A silent automation can stop for weeks before anyone notices. Monitor volumes and errors.
• No owner Automations outlive the person who built them. Document each one and name a current owner.

KPIs to track

• Process cycle time Elapsed time from trigger to completion, such as order to invoice. Should fall sharply after automation.
• Manual touches per transaction How many times a person intervenes. Falling means the process is clean; rising means exceptions are growing.
• Error rate Mistakes per hundred transactions, before and after. Automation should reduce it, not just speed up the errors.
• Hours redeployed Staff time released and what it is now spent on. Automation that frees hours nobody uses well has not paid off.
• Automation failures caught by alerts Failures detected by monitoring rather than by a customer. Target all of them.

Frequently asked questions

Do I need AI to automate a process?

Usually not. Most process automation is rules-based and uses the automation features in your existing software or a connector tool. AI is needed only where a step involves unstructured input or judgement.

What is RPA and does a small business need it?

Robotic process automation uses software "bots" to click through application screens the way a person would, which helps when older systems have no proper integration. Most small businesses on modern cloud software can integrate directly and do not need it.

How do I know a process is ready to automate?

It is mapped, simplified, has a standard input, and fewer than roughly one in five cases needs human intervention. If you cannot describe the exceptions, you are not ready.

Will automation cost jobs in my business?

In most small businesses it removes tasks rather than roles, and the time goes to customers, sales and quality. Involve staff in mapping and redesign; they know where the real friction is and will adopt the result faster.

Key takeaways

• Automate processes that are frequent, rule-bound, time-consuming and error-prone, starting with high value and low effort.
• Map and simplify first; automating a messy process preserves the mess.
• Rules for the fixed backbone, AI only at the unstructured joints, humans for exceptions and high stakes.
• Standardise inputs: most automation failures are really intake problems.
• Automation is infrastructure: give it an owner, alerts and a quarterly review.