Why “Repetitive” Doesn’t Mean “Automatable”
Repetitive is not enough. The system behind the task has to expose its data.
Sequel to “Is Your Business AI Ready? 6 Essential Foundations Every Australian Business Must Check”
The most common thing we hear from a business owner considering AI is some version of “we want to automate our repetitive tasks.” That’s a reasonable instinct. It’s also the wrong question, and answering it the wrong way is why so many AI projects stall before they deliver anything.
In our piece on AI readiness, we named connected processes as one of six foundations a business needs before AI investment pays off. This is the deep dive on that foundation, because the mistake underneath it is almost always the same one: assuming repetitive means automatable.
The scale of the problem
It isn’t. Repetitive tells you a task happens often. It tells you nothing about whether the system running that task can actually talk to anything else.
Salesforce’s MuleSoft runs an annual Connectivity Benchmark Report surveying IT leaders on exactly this question, and the gap it finds is stark. In the 2025 edition, the average organisation was running 897 applications, with 45 percent of respondents managing 1,000 or more. Only 2 percent had successfully integrated more than half of them. The 2026 edition found the same shape of problem at a larger scale: organisations that consider themselves furthest along in adopting AI manage an average of 1,057 applications, and even among that group, only 32 percent of those applications are actually connected to each other.
Read that again. The businesses doing the most with AI are still only talking to a third of their own systems. Everyone else is lower than that.
Asana’s Anatomy of Work research puts a human cost on the same gap. The average knowledge worker spends roughly 209 hours a year on duplicative work, tasks that already happened somewhere else in the business and have to be redone or re-entered because nothing carried the result forward automatically. That’s more than five working weeks a year, per person, spent re-doing work instead of doing it once.
Why repetitive tasks don’t automate themselves
A task can happen fifty times a day and still not be automatable, because automation isn’t about frequency. It’s about whether the system behind the task can expose its data to anything else, automatically, without a person copying, pasting, exporting or re-keying it somewhere along the way.
Ask the question that actually changes the conversation: does the system behind this task expose its data, or does someone move it from one place to another by hand? If the honest answer is manual extraction, you don’t have an AI opportunity yet. You have a systems integration problem, and it needs solving first, because layering AI on top of a manual handoff just makes the handoff happen faster without making it disappear.
What a real example looks like
We worked with a childcare group running 18 centres where compliance reporting was completely manual. Every centre manager spent around four hours a week filling in an Excel sheet and dropping it into a shared folder. Across the business, that was close to 80 hours a week once the general manager’s team had collated everything and checked it for errors, every single week, all year.
The task was about as repetitive as it gets. It was also entirely unautomatable in its original form, because the “system” behind it was a spreadsheet with no shared structure, filled in slightly differently by every centre. There was nothing for an automated process to reliably read.
We didn’t start with AI. We built a standardised digital form with error-checking built in at the point of entry, so every centre manager was feeding data into the same structure, correctly, the first time. Once that was in place, the collation work at head-office level became close to automatic. Centre manager reporting time dropped by more than half. The general manager’s team’s collation time dropped by more than 75 percent. None of that came from artificial intelligence. It came from making the process capable of feeding a system instead of a spreadsheet.
How to tell if a process is actually ready
Pick your three most repetitive processes and ask the same question of each one: does the system this runs on expose its data automatically to something else, or does a person have to move it by hand? Every “by hand” answer goes on your priority list, not because it’s urgent, but because it’s the thing quietly capping how much value any tool you buy later can actually deliver.
This costs nothing but time to work out, and it tells you exactly where to start before a single dollar goes toward automation or AI.
Where this fits with the rest of AI readiness
A process that’s genuinely connected still won’t help if the data flowing through it is inaccessible to the people who need it, or if the team running the process was never consulted about where it actually breaks down. Repetitive, connected processes are the raw material AI needs to work with. They’re not the whole picture, and they’re the piece almost every business gets wrong first, because it looks the simplest from the outside.
If you’re not sure whether your business’s processes are genuinely connected or just genuinely repetitive, our free health check gives you a plain-English picture in under five minutes.
Take the free health check: linkingintegrating.com/health-check
If the results point to something worth a proper conversation, the next step is a direct call about what’s fixable and what the right sequence looks like for your business.
Book a 30-minute call with Mark Cooray
Frequently asked questions
What’s the difference between a repetitive task and an automatable one?
A repetitive task just happens often. An automatable one runs on a system that can share its data with other systems without a person moving it by hand. A task can be both, or it can be repetitive and completely stuck, which is more common than most businesses expect.
Do we need new software to fix this, or can we fix the process first?
Usually the process first. The childcare group’s fix wasn’t a new platform, it was a standardised way of capturing the same information consistently, so the data was finally in a shape any system, AI or otherwise, could actually use.
How do we know if a system “exposes its data”?
Ask whether another system or tool can pull information out of it automatically, through something like an API or a scheduled export, without a person opening it and manually copying anything. If the only way data leaves that system is a person typing it somewhere else, it doesn’t expose its data yet.
Is this only a problem for larger organisations with lots of systems?
No. MuleSoft’s research is based on large IT estates, but the same failure shows up at much smaller scale, usually as one or two spreadsheets standing in for a proper shared system. The childcare group had 18 centres, not 18,000 employees.
Where should we start if several processes fail this test?
Start with whichever one costs the most people-hours, not whichever feels most urgent. That’s usually the process everyone already complains about, and fixing it first proves the value of doing the rest properly.
Mark Cooray is the Founding Director of Linking Integrating, a Melbourne-based digital transformation and systems integration consultancy helping growing businesses across Victoria and Australia connect the systems behind their everyday processes, before recommending any tool, AI included.