One back-office process, automated with AI agents so that most of it runs without your team.
Invoice processing, supplier bills, credit control, collecting documents from clients: work that follows the same steps hundreds of times a month. I can map and automate it for you, or advise your own team while they do it.
- Firms of 20 to 100 people
- Built for you, or with your team
- Inside the tools you already use
- Remove it: nobody needs this step
- Plain software: the rule never changes
- An AI agent: it takes reading and judgment
- A person: the decision carries real risk
I am Federico Toscano, an AI automation consultant for small and medium businesses in the UK. I take one back-office process at a time, such as invoice processing, credit control or client onboarding, and automate it with plain software and AI agents inside the tools the firm already uses. I can do the work, or advise your own team while they do it. The first call is free, an audit call costs from £450, and the work starts from £2,000.
Why back-office work in a small business stays manual
Most firms of 20 to 100 people run their back office on an accounting package such as Xero or QuickBooks, a shared inbox and a set of spreadsheets. A handful of people who know how it all fits hold those together. They are usually among the most capable people in the firm, and a large part of their week goes on the same steps repeated: copying figures from one place to another, checking them, asking someone for the missing piece, and waiting for the reply.
You see the cost in several places. Invoices go out days or weeks after the work was finished, so the money arrives later than it should. Supplier invoices pile up before each payment run. Credit control happens when somebody finds the time, and bank reconciliation is left until month end. When the one person who knows a process is away, it slows down or stops, and when the firm grows, the only way to keep up is to hire another person for the same work.
Ordinary software has never solved this, because it only handles the standard case. It can move a number from one system to another when the number arrives in the expected format. An invoice that comes in as a photo, or an email that answers only half the question, still lands on a person's desk as data entry and checking. In most back offices those exceptions are a large share of the work.
A second difficulty makes the first one harder to fix. The process that people describe is rarely the process as it runs. The description leaves out the loops: the invoice sent back because a reference was missing, or the approval that sits in an inbox for a week. A lot of the delay sits in those loops, and nobody has them written down.
How to automate a business process with AI agents
Two things make this fixable now. The first is the AI agent: software that can read documents and messages the way a person would, choose between a small set of actions it has been allowed to take, and carry one out. The exceptions that ordinary software left to your staff can now be handled too. The second is cost. The tools to build with have become affordable for a firm of your size, where a few years ago this was a project for large companies.
The first step is to write down how the process really runs, before any software is chosen. That means talking to the people who do the work and reading the history in your systems, because the two rarely tell the same story. The result is a map of the process with a few numbers attached: how long it takes from start to end, how often work goes back for correction, and what it costs you each month.
Every step in the map then gets one of four answers.
- Remove it. Some steps exist because of a problem that went away years ago. Taking them out costs nothing, and it can save more time than any software.
- Plain software. Where the rule never changes, ordinary code does the step the same way every time.
- An AI agent. Where something has to be read and understood, such as matching an invoice to an order or drafting a payment reminder, an AI agent does it and passes the result on.
- A person. Where a decision carries real risk, such as approving a payment, it stays with someone on your team.
Which business processes can be automated
The method is the same whichever process you start with. These are the usual candidates in a small or medium business.
- Invoice processing and accounts payable. An AI agent reads supplier invoices as they arrive, checks them against orders and codes them, and a person approves the payment run.
- Invoicing and accounts receivable. Timesheets, job records or delivery notes become invoices on the day the work is finished.
- Credit control. Payments arriving in the bank are matched against what was invoiced, and reminders for what is late are drafted for your team to send.
- Bank reconciliation and month end. Transactions are matched as they come in, so the close becomes a review of the items that did not match.
- Data entry between systems. Information in emails, forms and documents is moved into your accounting package or your CRM without retyping.
- Client onboarding. Documents are requested from the client and checked when they come back, and whatever is missing is asked for again.
How it is built, and how it stays safe
The new process is built inside the tools you already use, so nobody has to learn a new system. Approvals reach your team by email or in the chat tool they already have open.
Because this work touches money, the design follows one rule: an AI agent proposes, fixed rules do the calculation, and a person approves. When the automation is unsure about an item, it stops and passes the item to a person. Every action is recorded, so you and your accountant can see what was done and on what basis.
Once it is running, the same numbers from the map are measured again. You compare before and after on the same measures, and decide from that whether the next process is worth doing.
As your AI automation consultant, I can do the work or advise your team
Much of this work no longer needs an engineer. Someone on your team who knows the process can map it and build a good part of the automation. An engineer's judgment still matters where an automation acts unattended on money or client data.
I do the work
I map the process, build the automation and keep it running afterwards. This suits a firm where nobody has the time to take it on, or a process that crosses several systems.
Your team does the work, and I advise
Someone on your team maps and builds. I review the plan before they build, check the result before it is allowed to act unattended, and stay available for questions while it runs. This suits a firm that wants to keep the skill in-house.
Either way it starts with a free 30-minute call about the problem. If it looks worth going further, the next step is an audit call, from £450, where we go through the process step by step with the people who run it. The work itself starts from £2,000, at a price agreed before anything begins.
I do not guarantee how long the work will take, because that depends on your process and on how much time your team can give it. As a guide, expect an automation I build to be running one to two months after the audit call.
I am based in London and work with small and medium businesses across the UK, mostly remotely, in English or Italian. You can read more about how I work.
Questions about AI automation in a small business
What does an AI automation consultant do?
What business processes can be automated with AI?
Can AI process invoices?
Can AI do bookkeeping?
How much does AI automation cost for a small business?
Is AI automation worth it for a small business?
How long does it take to automate a business process?
Do we need to replace Xero, QuickBooks or Sage?
Start with one process
Book a free 30-minute call. Tell me which process takes too much of your team's time, and I will tell you whether it looks worth automating.