Before automating a process, you need to know where each piece of data it uses comes from and how reliable that data is. When that isn't settled, automation ends up reproducing the same errors faster and with fewer people checking them.

This plays out often enough to be predictable. A tool gets evaluated for months, it gets implemented, and six months later the team is still reconciling by hand exactly what it reconciled before. When you look at what happened, the software rarely failed. What went into it did.

Where does information live in a company of 10 to 300 employees?

Almost always in more places than the org chart suggests. There is an ERP or an administrative system that was bought complete and in practice gets used partially, with entire modules whose fields stayed empty because nobody ever defined who fills them or at what point in the day.

Alongside it sit local spreadsheets that tend to be the real operating system. They live on one person's computer, they have versions with improvised names, and when that person is away the process stops or someone improvises a copy that no longer matches the original.

Then there is WhatsApp, which in many companies is where authorization actually happens. An order gets released there, a credit exception gets accepted there, and that decision is not recorded in any system. Months later nobody can reconstruct the reasoning behind it.

There is also whatever gets written on paper and captured the next day, or on Friday: a warehouse slip, a service sheet, a visit note. And there is the knowledge that isn't written down anywhere. Someone on the team knows which supplier delivers late, which customer always asks for an extension, which filing gets stuck if it doesn't go out before noon. That judgment is real and it is valuable, but it is not information the company can use.

A process spread across those five sources is not in shape to be automated. What it needs first is order.

Why do operational records stop matching?

For fairly ordinary reasons. The most common is that the same data gets captured twice, in two different systems, by two people, at different points in the day. As long as they agree, nobody notices. Once they diverge, there is no way to tell which one is right.

Close behind is the absence of an owner. A field can stay wrong for months because everyone consults it and no specific person maintains it. When you ask who owns that data, the answer is usually a department rather than a name, and departments don't correct records.

Exceptions carry weight too. The documented process covers the normal case, while the unusual case gets resolved over the phone and leaves no trace. In most operations we've seen, those exceptions are exactly the ones consuming the director's attention, and they are also the ones no automation will handle if they aren't described somewhere.

One contrast is worth pointing out. Data with external validation tends to be the most reliable in the company: invoices filed with the tax authority, payroll reported to social security, anything that moved through a bank. It carries a date, a reference number, and a third party that reviewed it. When a process can be anchored to that kind of record, the starting point is considerably better than when it depends only on internal entries.

How do you know whether a process is ready to automate?

This table works as a quick filter. It isn't a verdict, it's a way to order the conversation before evaluating tools.

A process that falls on the right in two or more rows should rarely be first on the list, however much it hurts. There are exceptions, particularly when volume is high enough that ordering it by hand is no longer viable, but those are worth treating as exceptions rather than as the general case.

When is it better not to automate yet?

There are four situations where waiting costs less than moving.

When the process changes every quarter, because you'd be automating a version that is about to stop existing. When nobody can describe the decision rule, since if the criterion lives in one person's judgment and can't be written down, what exists is an individual skill rather than a process. When the stated goal is headcount reduction, because those projects sabotage themselves: the people who know the process are the same ones who have to explain it. And when there is nobody to maintain it afterward, given that an automation without an owner degrades within months and ends up as one more system nobody uses.

Where should you start?

Before comparing platforms, there is a duller and more useful exercise: mapping where the time of whoever runs the operation actually goes. Not the team's time, the leadership's. The approvals, the reconciliations, the follow-ups, and the same questions answered every morning. That map usually reveals that the most expensive process for the company is not the highest-volume one.

With that list in hand you can locate, process by process, where the data lives and who answers for it. The table above works as the filter. What doesn't pass isn't discarded, it gets ordered, and ordering it often frees up time on its own.

What follows is automating in order of capacity freed rather than in order of technical difficulty. A well-chosen process can free 2 to 5 times the execution capacity in that specific process. The conversation that matters afterward isn't about how much was saved, it's about what the company is going to do with the hours that opened up.

In most companies this size the obstacle isn't the available technology. It's that the person who could be growing the business is the same one running it every day. Putting the data in order is what makes changing that possible.

About Kirana Labs

Kirana Labs is a consulting firm based in San Pedro Garza García, Nuevo León, working with Mexican companies of 10 to 300 employees. We help owners and operations directors get operations off their desk: we identify the processes consuming their attention, put in order whatever needs it, and implement automation and agents where they free the most execution capacity.

If you want to review which of your processes are ready and which aren't schedule a call.