Business process automation is software doing the repeatable steps of a process that your people currently do by hand. It moves information between systems, starts the next step when the last one finishes, chases what is overdue, books dates and produces documents. It does not involve robots, and a chatbot on your website is something else entirely. It will not fix a process nobody has agreed, and it does not replace the person who approves consequential decisions. It pays where work is high in volume, follows clear rules and arrives with clear inputs.
We are writing this for the owner or operations lead of an established Irish business. You keep hearing about automation and AI, and you want a straight account of what either would mean for your business.
What it actually does
Most of the work automation takes on is unglamorous. Look at any process that runs every day and you will find the same kinds of step.
Moving information between systems
An enquiry arrives through the website and someone retypes it into the CRM. An order is approved and someone keys it into the accounts package. Every copy takes time and invites a typing error. Automation passes the information across once, in the agreed format, as soon as it exists.
Starting the next step
When a quote is accepted, a job should be created. When a job is complete, an invoice should follow. In many businesses a person notices the change and starts the next step by hand. Software can watch for the change and start that step every time.
Chasing
Someone has to follow up unsigned documents, unpaid invoices and details a customer has not yet sent. Scheduled reminders do this without anyone keeping a list on their desk.
Scheduling
Allocating work to the right person or contractor, booking a date and sending the confirmation usually follow rules that can be written down. Once they are written down, software can apply them.
Producing documents
Quotes, contracts, job sheets and reports are often assembled from information the business already holds. Software can fill a template from the record and pass the result to a person for approval.
What it is not
The word automation carries a lot of baggage. Four misunderstandings come up in almost every first conversation we have about it.
It is not robots
The word suggests machines on a factory floor. In an office-based business, process automation is software connecting the systems you already use, with agreed rules about what happens when.
It is not a chatbot bolted on
A chat window on the website answers questions. It does not move an order through your business. If the process behind it is still manual, the enquiry lands in the same inbox it always did.
It does not fix a process nobody has agreed
If three people handle the same job in three different ways, automating it means choosing one of those ways. That choice has to be made by the people who run the work, before anything is built. Software repeats whatever it is given, so an unclear process simply produces unclear results more quickly.
It does not remove the need for approval
Some steps commit the business to something: a price, a refund, a credit note, a contract. Those keep a named person approving them. The automation prepares the decision and records the outcome, and the person makes the call.
Where it pays, and where it does not
Automation pays where the same steps repeat many times a week, the rules can be written down and the inputs arrive in a predictable form. Enquiry routing, order entry, appointment reminders, invoice chasing and document assembly usually qualify.
It pays less where each case is different or where the volume is low. A process that runs four times a year rarely earns the cost of building and maintaining it. It also pays less where the judgement is the work itself. A negotiation, a diagnosis or a design decision needs a person, and software can only gather the information they need.
Our piece on workflow automation for B2B businesses goes further into choosing which workflows to start with.
Where AI fits
AI is one tool within automation, and it is useful where the input is messy. Reading a supplier invoice that arrives as a PDF is one example. Drafting a reply for someone to check is another, as is sorting incoming emails by what each one needs. Fixed rules struggle with that kind of input. A language model handles it reasonably well, with a person checking the output where it matters.
In most businesses, though, most of the gain comes from plain automation. Connecting two systems and setting a schedule removes a great deal of manual work without any model involved. Start there, and bring AI in where a step genuinely needs to read or write language.
What has to happen first
The process has to be mapped and agreed before anything is built. Our note on process mapping before automation sets out how we do that. In outline, the order of work looks like this.
- Choose one process that runs often. Pick something with real volume and a clear start and finish, such as an enquiry becoming a booked job.
- Map it as it runs today. Follow real pieces of work through the business, including the emails and spreadsheets nobody mentions in a meeting.
- Agree the version you want. The people who run the work settle the disagreements and decide what the standard route is.
- Mark the steps that need a person. Approvals, exceptions and anything that commits the business stay with a named owner.
- Automate the repeatable steps. Build one connection at a time and test it against real work. Keep the manual route available until the new one is proven.
- Measure the result. Compare the new process against how the old one ran, using the same unit of work.
An example from home heating services
One of our published studies covers a home heating service whose enquiries arrive through forms on its website. The commercial objective was to route each enquiry to an installer and support recurring servicing bookings.
We built a lead record that carries each enquiry from the form through to allocation to an installer. The record is connected to a supplier portal, which suppliers reach through a sign-in link known as a magic link. Reminders run on a schedule each year, and transactional email supports the repeat servicing booking from that same continuing record.
This is plain automation, built on a continuing record, a portal and scheduled reminders. The study records 1,800 enquiries allocated to installers each year, from the system's allocation records. You can read the full servicing study on our site.
That is how we approach automation for an established business. The process comes first, the approvals stay with people, and software takes the repeatable steps.
What to ask whoever builds it
Whoever you choose, these questions will tell you quickly whether the work has been thought through.
- Which steps of our process will the software do, and which stay with a person?
- Who agreed the version of the process you are automating?
- Which steps need someone's approval, and how is that approval recorded?
- What happens when a case does not fit the rules?
- Where is AI used, if at all, and who checks what it produces?
- How will we know it worked, and against what baseline?
If you have a process that repeats every day and you want it to run without being retyped, you can start a project with us.
