What we built this week

Your bot now speaks the language of each system it works in

Every business has the same small headache: one system calls a supplier "ACME (PTY) LTD", the next one wants "ACM001". Someone keeps that translation in their head, or in a spreadsheet nobody else can find.

This week we gave Automize lookup tables. You list what a value looks like in one place and what it should become in another. The bot looks it up as it works.

The part we like most is what happens when the bot meets something new. It doesn't guess. It writes the value down as a "miss" and carries on with a safe default you choose, like REVIEW. A person fills in the answer once, and the bot knows it from then on.

Every table has its own page now. You can search it, upload it from Excel or CSV, see which automations use it, and see who changed what.

Small tables like these are where a lot of automation projects quietly fail. We'd rather they were easy.

An automation supervisor that learns, and knows when to ask

Automations break in boring ways. A screen loads a second slower than it used to. A button moves. A pop-up appears that nobody has seen before.

This week our supervisor learned to do something about it, within rules each company sets.

When a step keeps failing or wasting time, it works out a small change, such as waiting a little longer or running at a quieter hour. It tries the change on the next few runs and compares them with the four weeks before. It keeps the change only if things got better. Otherwise it puts everything back.

If a button's usual address stops working but a backup one does, it can switch them round. If the bot meets a pop-up it doesn't recognise, it stops and asks the owner which button to press. It never guesses.

Every change is logged, can be undone in one click, and the process owner is told. Steps you mark "can't be undone", like making a payment, are never retried automatically.

It's off until a company switches it on. We think that's how AI in operations should work: useful, measured, and on a short lead.

When a system goes down, you should know which automations it takes with it

If your accounting system has an outage tonight, which of your automations will fail? Most teams find out the next morning, one failure at a time.

Automize now keeps track of which system every step of a process works in. Open an application such as Sage or SAP and you see every automation that uses it, down to the individual steps. Open a process and each step shows the system it's working in.

That opens up two useful things.

When several different automations fail against the same system at around the same time, Automize now says "this system looks like it's down". That's very different from "your automation is broken". A person still makes the call.

You can also tell a process when it's allowed to run, for example only between 17:00 and 21:00, by painting the hours on a weekly grid. The rule holds however the run starts: on a schedule, from a queue, by hand, or from a watched folder on the bot's own machine.

Our aim is simple: fewer surprises, and fewer runs landing in the middle of month-end.

Every invoice now adds to what we know about the supplier

Reading an invoice is only half the job. The other half is asking whether it looks right for this supplier. Is the bank account the one they always use? Is the amount in line with their usual invoices?

This week, the steps that read invoices, credit notes and statements can record what they read in a supplier history as they go. You only need to switch it on. There's no separate step to build.

We also gave suppliers a proper home in the library. Each supplier gets a card showing its VAT and registration numbers, how many documents it has sent, and anything that needs a decision. Critical findings and new bank accounts waiting for approval are listed first. You can also find any supplier from the main search bar by name, VAT number or registration number.

For a finance team, that's the difference between capturing invoices and actually checking them.

The work you don't see: a faster, steadier Automize

Not every week is about new features. This one also went into making everything already there work better.

We worked through 20 performance items and the first findings from watching our live database. Many pages now ask the database fewer questions, and the bots' regular check-in carries only what the bot needs.

The app now stops a page's background work the moment you leave it. Before this, a page you'd left could keep updating and show its information on the next page you opened.

Every list in the app now looks and behaves the same way, so the screens are calmer and easier to scan.

On the bot side, a copy of Excel that stops responding can no longer hang a run forever. Finished processes are reported as finished on Windows, and bots sign themselves back in more reliably.

We also wrote automated tests for another 3,800 lines of code, which lifts the share of our code checked by tests from 90.4% to 92.9%. It's not glamorous work, but it's why you can trust the automations running on it.