Almost every growing business has one. The person who knows which export to run, which columns to fix, which supplier codes never match, and what has to happen on the third working day of the month before anyone else can close the books. They are usually one of the most reliable people in the building.
That is exactly the problem. When a process only works because one person remembers how it works, the business is carrying a risk it can’t see on any report. We call this human middleware: people doing the job that software should be doing between your systems.
This isn’t a criticism of the person. Most of the time they built the workaround because nobody else would, and it has quietly kept the business running for years. The aim is to take the weight off them, not to take them out of the picture.
What human middleware looks like
It rarely looks like a problem from the outside. It looks like diligence. Common examples:
- Exporting stock levels from the inventory system, pasting them into a spreadsheet, fixing product codes and re-keying totals into Xero.
- A weekly management report that someone builds by hand every Monday morning from three different sources.
- A customer price list updated manually whenever supplier costs change.
- Timesheet hours checked line by line against payroll before every pay run.
- A reconciliation spreadsheet with tabs named FINAL, FINAL_v2 and FINAL_use_this_one.
Each of these works. Each also depends on one person’s memory of the steps, the exceptions and the fixes.
Why it’s riskier than it looks
The obvious risk is absence. When that person is sick, on leave or resigns, month-end waits. Reports don’t go out. Suppliers don’t get paid on time, or get paid twice. The person often can’t take a proper holiday because everyone knows what happens when they do.
The less obvious risks are just as costly:
- Errors stay hidden. If only one person understands the spreadsheet, only one person can check it. Mistakes in formulas or copied ranges can run for months.
- Growth adds work. Every new customer, product line or entity adds rows to the manual process. The business scales, the workaround doesn’t.
- Senior time goes to copying. The person doing this is often experienced and well paid. Hours spent re-keying are hours not spent on analysis or decisions.
- The logic is invisible. Business rules end up buried in cell formulas, macros and habits rather than written down anywhere the business can see them.
How to find the processes that live in one person’s head
You can usually map most of your human middleware in a couple of weeks. Work through these steps:
- Ask the three-week question. For each key person, ask: “What stops or goes wrong if you’re away for three weeks?” Write down every answer, however small.
- Walk the month-end calendar. List every task between the last day of the month and the day reports go to the owners or board. Note who does each one and which system it touches.
- List every hand-off between systems. Anywhere data leaves one system and enters another by export, import, copy and paste or re-keying is a candidate.
- Look at the file names. Spreadsheets with version numbers, initials or “do not touch” in the name are usually doing a job a system should do.
- Sit beside the person for a week. Watch the real process, not the described one. The exceptions and manual checks are where the knowledge lives.
- Note the checks. Every time someone “just eyeballs” a number before it goes out, record what they are looking for. That is a business rule nobody has written down.
What it really costs
To put a rough number on it, estimate the hours each process takes per week, multiply by a realistic hourly cost for the person doing it, and annualise it. Then add the things that are harder to count: the cost of errors that get through, the delay to decisions while reports are late, and what a two or three week outage would cost if that person left suddenly.
There is also an opportunity cost. When data sits in separate systems held together by a spreadsheet, nobody can easily ask bigger questions of it. When we brought one distributor’s inventory and accounting data into a single model, the analysis surfaced $850K of dead or overstocked inventory and $691K of overdue receivables. That kind of finding needs the data in one place first.
How to take the risk out
There are three moves, and the order matters.
Document it first. Write the process down as it actually runs, including the exceptions and the checks. A simple process map with swim lanes for each person and system is often enough. Don’t automate a process nobody fully understands yet, or you will automate the mistakes along with it.
Automate the repeatable parts. Move exports, imports and re-keying into integrations that run on a schedule or when something changes. Make sure each one logs what it did, so the business can see whether it worked without asking anyone. Keep a person approving the steps that genuinely need judgement.
Own the data model. The long-term fix is one model of your customers, stock, jobs, people and cash that the business owns, sitting in your own Microsoft or Azure tenant. Business rules live in documented, version-controlled logic rather than in cell formulas. New reports and questions then come from that model instead of from another spreadsheet.
Throughout all of this, keep the person involved. They know the process better than anyone, and their role shifts from operating the workaround to owning and reviewing the process. That is a better use of their experience, and it lets them take a holiday.
A quick self-check
- Could your month-end close happen on time if your most relied-on finance or ops person was away?
- Is there a spreadsheet that would be hard to rebuild if it were deleted today?
- Do any reports depend on someone copying data from one system to another?
- Are your business rules written down anywhere outside people’s heads and cell formulas?
- Does every integration you rely on tell you when it fails?
If you answered no, yes, yes, no and no, you have human middleware worth addressing.
Where to start
Pick the one process that would hurt most if its owner left tomorrow, and document it this month. If you want help mapping it, automating it or building a data model your business owns, see how we approach integration and automation and fractional CTO support, or get in touch for a conversation about where your risk sits.