Know-how
Ready for automation? Twelve questions about your data
Automation rarely fails because of the technology, but because of the data underneath. Twelve questions show whether a process is ready and where to start.
When a process is meant to run by itself, a program has to make the decisions a person makes in passing today: which customer is meant? Is this price still valid? Who asked for the change? A person fills gaps with experience; a program cannot. It works only as well as the data it is built on.
That is why an honest look at the foundation pays off before any automation. The twelve questions below help with that. They are grouped by four qualities that make a good data foundation. Answer each with yes or no, ideally for one particular process, such as "from order to invoice".
1. One source for every piece of information
Every piece of information should have exactly one place where it is maintained. Everywhere else reads it from there. This is often called a single source of truth. Where there are two places, sooner or later there are two truths, and a program cannot know which one applies.
- Is there exactly one list each for customers, products and prices that everyone uses? Or does the office keep one spreadsheet, the field staff another and accounting a third?
- Is a change entered only once? Does a new address apply everywhere at once, or does someone have to remember to change it in three places?
- Does everyone know which version is current? Or is there a "price_list_new_final_2.xlsx" somewhere?
2. Traceable
When work moves from people to programs, you need to be able to tell afterwards what happened. That is not bureaucracy. It is what makes it possible to trust an automation.
- Can you tell who last changed an order or a price, and when?
- Can you follow the path of a case back? From the inquiry through quote and order to the invoice, without searching mailboxes.
- Is it clear what is closed and may no longer be changed? An invoice that has been sent, an approved plan, a completed check. In my own accounting system, for example, a posted line can no longer be altered, only corrected by a reversing entry.
3. Clean data
Clean does not mean perfect. Clean means unambiguous, complete enough and in a form a program can check.
- Does every customer, order and product have a unique number? A name is not enough; there are three called "Miller".
- Are the same things written the same way? Dates, units, spellings. "pcs", "pieces" and "pc" are the same to a person and three different values to a program.
- Are entries checked when they are made? Or can the "quantity" field also say "about 5, see e-mail"?
4. Maintainable
An automation is never simply finished. Processes change, programs get updates, staff move on. A good solution copes with that.
- Does the process still run when the one person who knows it is away? Is it written down anywhere?
- Does something depend on a file nobody dares to touch? The spreadsheet with nested formulas, the macro written by a former colleague.
- Do you notice when something has gone wrong? A silent automation that stopped three weeks ago is worse than none at all.
What your answers mean
- Mostly yes: a good foundation. The process is probably suited to automation. Think about which step should run by itself first.
- Mixed: the most common case. Usually what is missing is not an automation but one place where the data comes together cleanly. Creating that place is often the first and most rewarding step. The automation then builds on it.
- Mostly no: then automation would come too early today. It would only carry the gaps along faster. First map the process cleanly in digital form. That alone saves time and questions.
Why this foundation comes before any automation, and certainly before any AI, I have described in more detail in the post Data first, then AI.
How I turn this into software
My applications start from exactly these four points. The data lives in one place, entries are checked when they are made, important steps are recorded in a traceable way, and the rules of the process sit in one central place instead of being spread across spreadsheets and heads. Automations work on this foundation instead of running alongside the system. If something fails, it gets noticed.
Not sure about a process? Describe it in the initial assessment. I will tell you honestly whether it is ready or what is missing first.