Business process problems: 9 signs you need a system

Nobody calls to ask for a system. They call because the same order was entered twice, because a client complained about something no one had logged, or because the team spends every Friday assembling a report that is already out of date by Monday.

Those are symptoms, and each one points to a different underlying problem. The gap between a project that pays for itself and one that does not is usually decided before a single line of code is written: in whether the right problem was identified.

Below are the nine signs that come up most often, how each one looks from the inside, and what kind of intervention actually fixes it. You do not need all nine. Two or three is already a case.

1. The same data gets entered more than once

How it shows up: someone copies from an email into a spreadsheet, from the spreadsheet into the management system, and from the system into a report. Or the salesperson enters the order in the CRM and finance re-enters it in the ERP.

What is actually happening: this is not an effort problem, it is an integration problem. Two systems that should be talking to each other are not, and a person is acting as the bridge. That human bridge is slow and — more importantly — it is where discrepancies are born: by the second keystroke the data is no longer the same.

What fixes it: an integration between the two ends. It is the least impressive work you can commission and the fastest to pay for itself, because re-entry errors are expensive and silent.

2. Finding out the status of something means asking a person

How it shows up: a client asks about their order and the answer takes a while because someone has to go and find out. Internally, knowing whether something got done means messaging the person who does it.

What is actually happening: the state of the process does not live anywhere. It lives in the memory of whoever ran it and in a chain of messages. At low volume that works; as volume grows, half the day goes into finding things out.

What fixes it: every step leaving a record as it happens, not afterwards. This does not require a large system. It requires the status to be answerable without interrupting anyone.

3. The process stalls when one person is away

How it shows up: some tasks only one person knows how to do. When they take holiday, work piles up. When they left, the process had to be reconstructed by asking around.

What is actually happening: the process was never written down. It exists as a practice, not as a definition. That is not a failing of the person — usually they are the one who does it best — it is operational risk concentrated in a single point.

What fixes it: writing the process down is half the work, and it can be done this week without buying anything. Automating it afterwards is what guarantees the written version is the one that actually runs.

4. The client finds the errors

How it shows up: the wrong invoice surfaces when it gets rejected. The stock shortage, once something has already been sold. The mistyped record, when someone complains.

What is actually happening: there is no validation at the point of entry. The error is made at minute zero and caught weeks later, by which point it has propagated to three places and costs ten times more to unwind.

What fixes it: moving the check to the moment of entry. A flow that validates on the way in does not eliminate errors — nothing does — but it catches them while they are still cheap.

5. Every department has its own spreadsheet

How it shows up: sales keeps its own client list, finance keeps another, and the warehouse runs a third with their own product codes. When they are compared, the numbers disagree and nobody knows which one to trust.

What is actually happening: there is no single source of truth. Each spreadsheet was born to solve a real need the central system did not cover, and over time became a parallel system that nobody maintains.

What fixes it: deciding who owns each master record and making everything else read from there. It is an organisational decision more than a technical one, and it is the one that unlocks everything else — including any artificial intelligence project, as I explain in AI and ERP.

6. Growing means hiring

How it shows up: handling twice the orders takes twice the people. The maths is linear and at some point it stops adding up.

What is actually happening: capacity is tied to person-hours. That is not wrong in itself, but it means margin does not improve with scale, and commercial decisions end up constrained by operations rather than by the market.

What fixes it: automating the repetitive stretch of the process, which is almost always most of the time even when it is the least of the judgement. It is the most quoted benefit and it is not the only one: there are four others that change the business more.

7. Requests arrive through five different places

How it shows up: they come in by WhatsApp, Instagram, phone, email and the website form. A different person handles each channel and none of them can see what the others did. The same client writes through two routes and gets two different answers.

What is actually happening: the channel is defining the process when it should be the other way round. There is no single queue, so there is no way to prioritise, no way to measure response time, and no way to know how many enquiries were lost.

What fixes it: merging the intake into one inbox, with the full history per client regardless of which channel they used. Only from there does automating the reply make any sense.

8. Nobody can answer “how many” without building a report

How it shows up: a simple question — how many orders came in this month, what did we invoice per channel, which product is slowing down — requires someone to export, cross-reference and assemble. And by the time it is ready, the question has moved on.

What is actually happening: the data exists but it is not queryable. It sits in different places in different formats, and it takes someone with specific knowledge to bring it together. That bottleneck means decisions get made on instinct — not for lack of information, but because asking for it costs too much.

What fixes it: a query layer over the systems you already have. This is where artificial intelligence pays off fastest, provided the underlying data is in order.

9. Replying takes longer than solving

How it shows up: the work itself takes ten minutes, but the client waited two days. The time went into getting the enquiry to the right person.

What is actually happening: this is a triage problem, not a capacity problem. Nothing is classified or assigned on the way in, so everything waits in the same pile until someone reviews it.

What fixes it: classifying and routing at the moment the enquiry arrives. It is one of the simplest things to automate and one of the most visible from the outside.

Which one to tackle first

If several of these apply, do not rank them by how much they annoy you. Rank them by two variables:

How often it happens. A boring daily problem pays off more than a dramatic monthly one, because the saving compounds and because you will have enough data to know whether it improved.

What an error costs. Not only in money: how long it takes to detect and how much work it generates to correct.

The first project should be high frequency, low error cost. That is where the system gets tested at real volume without a failure turning into a problem with a client. Expensive-error cases come later, once the team trusts the flow and there is enough history to tune the rules.

Starting the other way round — with the most serious problem — is what produces the projects that present beautifully and never quite reach production.

The sign you picked the right first process is that nobody finds it impressive and everybody is using it a month later.

Three things a system will not fix

A process nobody agrees on. If two departments disagree about who approves what, automating it freezes the argument in code. Define first, automate second.

A rule that changes by exception every week. If ninety per cent of cases are resolved “depending on the situation”, there is no rule to encode yet. Better to measure for a couple of months until the real pattern surfaces.

A demand problem dressed up as a capacity problem. Sometimes operations cannot keep up because the company is selling something it should not be selling. No automated flow fixes that, and automating it makes it harder to see.

One warning that applies to all nine signs: automating a bad process makes it fail faster and at greater scale. Before moving anything, it is worth asking which steps exist only because they always have. Typically between a third and a half of what was going to be automated can simply be removed. That part costs nothing and returns the most.

The next step

Pick one sign — the one that recurs most — and write the process down as it works today: who does what, with what information, and where it gets recorded. No tools, no vendors, one sheet of paper.

That exercise usually reveals one of two things. That the problem was smaller than it looked, or that it was a different problem entirely. Both conclusions save a lot of money.

With the process written down, what comes next is deciding how to get it into production without it becoming a permanent pilot — which is where most of them stall, and what I cover in how to implement AI on the systems you already have.