Business process automation with AI, inside your software.

A business process is a series of steps that produces a result for the company: an order entered, an invoice paid, a quote sent. Automating it means handing some or all of those steps to software that runs them the same way every time, with no retyping.

You’ll also hear BPA (business process automation), automated workflow, or simply task automation. The principle is always the same: a trigger (an email received, a date, a form), a series of actions in your software, and a traceable result.

The goal isn’t to remove people but to take away the low-value work: retyping, comparing, filing, chasing. These tasks often get dismissed as unimportant, until the mistake that costs you a credit note or a customer. Your team keeps the decisions and the conversations that matter.

Traditional automation (rules, connectors, tools like Make or n8n) links software together when the data is already clean and structured: a form that creates a contact, an order that generates an invoice.

RPA, or robotic process automation, mimics a person’s clicks on a screen. It helps when software has no API, but it breaks as soon as the screen changes.

AI automation adds reading and understanding: a PDF purchase order that doesn’t look like the last customer’s, an email in the sender’s own words, an ambiguous request. It extracts the information, compares it with your rules, and prepares the action. A well-designed project combines all three: rules wherever they’re enough, AI only at the steps that require reading or interpreting. When the task also means choosing the right next step among several, you move up to an AI agent.

The best candidates meet four criteria: they’re frequent (daily or weekly), repetitive (they follow known rules), measurable (you know how long they take), and easy to check (a person can verify the result at a glance).

In practice, in wholesale and distribution, manufacturing, and business services, I almost always start with one of these five processes:

• entering orders received by email;
• pricing quotes;
• processing vendor invoices;
• sorting incoming emails;
• following up with customers and suppliers.

They’re visible, their gains show up in hours within the first few weeks, and they set up the connections the next ones will need. If you’re torn between several candidates, an AI audit ranks them by gain and difficulty.

The cost depends on three things: the number of tools to connect, the variety of documents to read, and the level of control you want. A simple automation between two tools is nothing like a tool that reads orders from fifty customers, each in a different format.

At DigitalCX, each of the three steps (audit, prototype on your data, go-live) has its own fixed price, in writing before any work starts. If the project continues, the audit fee is credited toward the prototype.

Two recurring costs remain, paid on accounts opened in your name, with no commission for me: usage of the AI model, billed by its provider based on volume, and hosting. Monthly support is optional. Before comparing offers, always ask what happens when your volume doubles: that’s where opaque flat fees get expensive.

The timeline fits into three steps, each with its own date.

• The audit, ten business days: it measures the time lost, checks access to the data, and ranks processes by gain and difficulty.
• The prototype, five days: it runs on your own data, and your team tests it on real documents.
• Go-live, on the date written in the proposal: the tool works inside your software, with human approval at the start.

Your team spends a few hours on it during the audit, to show how the task is done today, then a short trial of the prototype. Quick wins come first: a first task can run while the next ones are being built, reusing the same connections to the ERP and CRM.

Three safeguards make automation reliable.

• Rules: the tool only enters what meets your thresholds (prices, discounts, quantities, customer terms) and sets the rest aside.
• The log: every action is recorded with the original document, so you can see in an instant what was done, when, and why.
• Human review settings: task by task, you choose between approval by a person and fully automated operation.

An AI model can make mistakes: that’s why a new task starts in approval mode. The goal is fewer errors, not a promise of perfection. Once a task has been confirmed many times without corrections, you can let it run on its own, and switch back to manual in one click if anything looks off.