Your best automation opportunity is between two systems, not inside one

Somebody senior has asked for an AI strategy. There has been a workshop, a whiteboard and a long list of ideas, and most of them are chatbots. A chatbot for HR policy, a chatbot for the product catalogue, a chatbot that summarises the board pack.

None of them touch the thing that actually eats your week: the export that leaves one system every Monday, gets tidied in a spreadsheet, and is typed into another system by somebody who has done it for six years and is the only person who knows how.

That is where the value is. It almost never makes the list.

Ideas come from what people can see

An ideas workshop asks people what annoys them, and people describe the screen in front of them. So the ideas are about single systems: search this better, summarise that faster, answer questions about the other.

The expensive problems are not inside any one system. They are in the gaps between them, where data leaves one piece of software and has to arrive in another. Nobody sees those gaps whole, because every person in the room owns one end of them at most. Finance owns the ledger. Operations owns the booking system. The CSV that goes from one to the other every night belongs to nobody.

You cannot find the opportunity until you can see the connections. Most organisations have never drawn them.

How much of your estate is actually connected?

Less than you would guess. MuleSoft's 2026 Connectivity Benchmark, a survey of 1,050 IT leaders across nine countries including the UK, found the average enterprise now runs 957 applications and only 27% of them are integrated with each other. The other 73% exchange data by export, email, spreadsheet or a person reading one screen and typing into another.

The same survey found 96% of organisations hit barriers using their data for AI, and the single most cited blocker, at 40%, was outdated architecture and data silos. Not model quality. Not cost. The plumbing.

Enterprise applications grew from 897 to 957 in a year, and only 27% of them are connected together. — MuleSoft 2026 Connectivity Benchmark

The estate is also growing faster than anyone is mapping it. Zylo's 2026 SaaS Management Index, drawn from 40 million software licences, found that business units now control 81% of software spend and IT directly manages 15%. Large organisations add around 21 new applications a month. Each one is a new end for a connection that nobody designed.

And the cost of the gaps lands on people. A 2025 survey of 500 US office workers found they spend more than nine hours a week moving data between systems by hand. The survey was commissioned by a company that sells data-extraction software, so treat the exact figure with the scepticism it deserves. The shape is familiar to anyone who has watched a month-end close.

Where to find automation opportunities

How do you find automation opportunities in a business? Map every system you run, every connection between them and who owns each end, then look at the connections rather than the systems. The strongest candidates are the manual and file-based handoffs, which in a typical enterprise is most of them, since only 27% of applications are integrated. Rank those by how often they run, how many people touch them and what breaks when they are late.

Once the map exists, the opportunities tend to fall into a handful of patterns. These are the ones we look for first:

• Handoffs done by hand. An export, a spreadsheet, a re-key. Every one is a candidate, and most need an integration, not a model. We made that argument at length in most of what you want to automate does not need AI.
• The connection everyone complains about. One handoff usually causes most of the pain: the one that fails silently, needs checking every week, or blocks the close. Fixing that one is worth more than automating ten others.
• Retired systems still being fed. A system that was replaced years ago but still receives a nightly file, because one report somewhere reads from it. You are paying to keep it alive and carrying the risk of it failing.
• Two systems doing the same job. Two CRMs after an acquisition, three places that hold the customer's address. Here the opportunity is removal, and the saving is a licence rather than a headcount.
• Systems with no owner. If nobody owns a system, nobody owns the connections into it, and any automation built on top of it will be orphaned the day it goes live.
• Contracts about to renew. A renewal is the cheapest moment to replace, consolidate or insist on an API. Miss it and you have signed up for another three years of the workaround.

What it looks like in practice

The pattern we see most often in financial services is a calculation that matters, fees or commission or reconciliations, sitting in a spreadsheet between a data warehouse and a finance system. The spreadsheet was built as a stopgap. It is now the most critical piece of software in the building, it has no tests, and one person maintains it. That person usually knows it is a problem. Nobody above them does, because it does not appear on any list of systems.

The AI question in that business is not "what could a chatbot do". It is "should that calculation live in the spreadsheet at all", and the answer is often a boring integration that removes a day of manual work each month and a real operational risk with it.

The other pattern is the reverse: an organisation that has bought an AI tool and cannot get it to do anything useful, because the data it needs is spread across four systems that do not talk to each other. The model was never the problem. The data was the project, and the connections were the data.

The uncomfortable part

Most organisations cannot draw their own systems map, and it is rarely because they are careless. The knowledge is spread across people who each hold one piece. The architecture diagram, if there is one, was drawn for a programme that finished years ago.

It follows that any supplier who proposes an automation or AI roadmap without first asking to see how your systems connect is guessing. They may guess well. They are still guessing, and the proposal will be shaped by what they sell rather than by where your money leaks. That is also why so many pilots that looked good in a demo stall at the integration boundary: the connection they depended on was never on anybody's map.

What to do this week

• List the systems. All of them, including the spreadsheets that behave like systems. An afternoon with the right three people gets you most of the way.
• Draw the arrows. For each connection, write down what moves, how (API, file, email, person) and how often.
• Name the owners. Both ends of every arrow. The blanks are your risk register.
• Circle the manual arrows. Then ask the people who run them how long they take and what happens when they go wrong.

That list is worth more than any ideas workshop, because it is built from how the business actually runs rather than what people can see from their desk.

If you would like somebody who has done it before to go through it with you, that is our sanity check: £1,250, a day, and a written answer on where automation would pay off in your business and where it would not. You can see exactly what the write-up contains on what you get.