Automating the boring parts of invoice processing
We recently helped a transport company near Ely get their Tuesday mornings back. They were dealing with 156 invoices a month that someone had to type into a ledger by hand. Our simple bot now reads those documents and sorts the data in under 4 seconds per page.
The Tuesday morning backlog in Ely
At a small transport yard just outside Ely, the start of every week used to be a bit of a nightmare for the office manager. She wasn't managing the fleet or talking to drivers; she was stuck staring at a pile of 156 paper and PDF invoices that had arrived over the previous seven days. Each one had to be opened, read, and then manually typed into a spreadsheet that tracked their monthly spending. This took her about 14.5 hours every month, usually spread across three or four long Tuesday mornings. Because the work was so repetitive and dull, she would often make small mistakes, like flipping a 6 and a 9 or missing a decimal point in the VAT column.
These tiny errors do not seem like a big deal until you realize they lead to real money leaks. In November 2024, one typo on a fuel invoice meant the company overpaid a supplier by £382, and nobody noticed for nearly three months. For a business with a team of 9 people, that is the kind of leak that adds up fast over a year. We sat down with them to see if we could stop the manual typing entirely. The goal was to use simple bots for boring tasks so the staff could focus on the 47 active accounts that actually needed their attention. Honestly, most small offices have a 'Sarah' who is drowning in this kind of admin without realizing there is a better way.
We found that the office manager was spending 14.5 hours a month just typing numbers from paper into a screen.
Why manual data entry costs £3,800 a year
When we looked at the numbers for this logistics firm, the cost of manual entry was higher than they thought. If you pay someone £18 an hour and they spend 14.5 hours a month on one task, that is £261 a month just on typing. Over a year, that is £3,132. But then you have to add the cost of the errors. We audited their previous 83 invoices and found 4 mistakes that had cost them an extra £670 in overpayments and late fees. Tech that actually does its job should pay for itself by stopping these silly mistakes. In this case, the total waste was roughly £3,802 per year, which is a lot of money for a small firm to just throw away.
Most people think they need a massive IT department to fix this, but you don't. The problem is that humans are not built to look at 156 different layouts and extract 5-digit numbers perfectly every time. We get tired, we get distracted by the phone, or we just get bored. A bot does not get bored. It looks for the same markers every time—things like the 'Invoice Number' label or the 'Total Due' box. By moving this work to a bot, the firm saved roughly 3.6 hours every single week. That is time they now spend on growing their regional delivery routes instead of staring at a flickering monitor.
Training the bot on 18 different layouts
The biggest hurdle for this logistics firm was that their suppliers all sent different-looking invoices. Some were modern PDFs from big fuel companies, but others were hand-written notes or messy scans from local mechanics. We counted 18 different formats that they received regularly. A basic software might struggle with this variety, but our approach uses specific markers to find the data regardless of where it sits on the page. We don't care if the VAT is at the top left or the bottom right. The bot is trained to look for keywords and the number patterns that follow them.
During the first week of October 2024, we ran a test batch of 47 invoices. The bot correctly identified the date, the supplier name, and the total amount on 46 of them. The one it missed was a scan that was so blurry even a human had trouble reading it. (Heads-up: If a human can't read it, a bot probably can't either.) We showed the team how to handle these rare exceptions in about 10 minutes. Instead of typing everything, they now only have to look at the 1.4% of files that the bot flags for a quick manual check. It turned a whole morning of work into a 5-minute task on a Monday afternoon.
The bot correctly handled 46 out of 47 invoices in the first live test batch.
Getting the data into the ledger
Extracting the data is only half the battle; you also have to put it somewhere useful. For this client, they used a standard accounting software that needed a CSV file upload. Before we arrived, the manager was typing the data into the software one by one. Now, the bot gathers all the info from the 156 monthly invoices and creates one clean file. This file lists the supplier, the date, the net amount, the VAT, and the total. It even maps the expense to the right category, like 'Vehicle Maintenance' or 'Office Supplies', based on the supplier's name.
We set up the system to run every night at 11:00 PM. This means when the team walks into the office on Innovation Way the next morning, the previous day's invoices are already processed and waiting in a draft folder. There is no more 'backlog' because the work happens while everyone is at home. Since we started this in late 2024, the company has not had a single late payment fee. They used to get hit with at least one £40 fee every two months just because an invoice got buried under a coffee mug or lost in an inbox. That simply doesn't happen anymore.
Handling the 1.4% of errors
No bot is perfect, and we are upfront about that. We don't promise 99.8% accuracy because that is not realistic in the real world. Our bot currently hits about 98.6% accuracy for this specific client. When it finds an invoice it's not sure about—maybe because of a coffee stain on the paper or a very unusual font—it doesn't guess. Instead, it moves that file to a 'Needs Review' folder and sends a short email to Sarah. This happens about twice a month now, compared to the 156 files she used to have to touch every single time.
This 'human-in-the-loop' style is why the system works so well for a small business. You don't lose control of your finances; you just stop doing the grunt work. In November, the bot flagged an invoice where the total didn't match the sum of the line items. It turned out the supplier had made a mistake on their own math. Because the bot checks the math on every page in 3.2 seconds, it caught an error that a tired human would have definitely missed. The firm ended up saving £112 on that one bill alone just by asking the supplier to correct the total.
Starting your first batch on Monday
If you are still typing data from invoices, you are likely wasting at least a full working day every month. You don't need to change your whole business or hire a developer to fix this. At TTM Bot, we focus on simple bots for boring tasks that can be set up in a few days. For the logistics firm in Ely, the whole setup took about 6 hours of our time and about 2 hours of theirs. We didn't need to install anything on their old computers; the bot just watches a specific email folder and handles the rest.
By the time January 2025 rolled around, the office felt completely different. There was no more stress on Tuesday mornings, and the desk was actually clear of paper. If you have 30 or more invoices a month, this is a change that pays for itself almost immediately. Made in Cambridge, our tools are built for local businesses that want to get back to real work. You can start with a small batch of 10 invoices just to see how it looks. Once you see the bot extract the data in real-time, you'll wonder why you ever spent your life typing in the first place.
