Asking your management system questions in plain language
A familiar scene: the owner, in a meeting, needs a number. "How many jobs did we complete for that customer this year? And compared to last year?" The number exists somewhere in the management system. Getting it out requires the right person, the one who knows which screen to open, which filter to apply, how to export the spreadsheet. The answer arrives the next day, after the meeting has ended. Multiply this scene by every question a company asks itself in a week and you have the real condition of data in most mid-sized companies: the data is there, the answers are slow.
The problem
The management system does its job: it records everything. Between the data and the people, however, sits a skill barrier of screens, filters and exports. Access to answers ends up concentrated in two or three people, who become the bottleneck for every question. Since asking has a cost, most questions are simply never asked, and decisions fall back on intuition. The paradox is that the company pays to collect data it rarely interrogates.
Static reports help only in part: they answer the questions of the past. The question born five minutes ago in a meeting has no report waiting for it.
An assistant that queries the real records
In the company where I lead software development, anyone can ask questions in plain language to an AI assistant connected to company data. "How many trips did we run for that customer in June?" "Which files expire this month?" The assistant answers in seconds, with the numbers and the detail of where they come from.
One technical point is worth understanding even without a technical background: the assistant executes queries. When it receives a question, it translates it into interrogations of the actual records, runs them, and builds the answer from the results, the way an expert colleague would, in seconds. When the data holds no answer, it says so.
The second point is confidentiality, and it is a matter of design: the assistant sees only what the person asking is allowed to see. Permissions are the same as in the management system, and access is read-only.
What actually changes
The value lies in access. Questions that used to cost a phone call and a day of waiting now get asked. The bottleneck around the few people who master the system dissolves, and those people return to higher-value work. Decisions shift, gradually, from feeling to numbers, because the number now costs less than the feeling.
Common objections
"What if it gets the numbers wrong?" It answers by executing queries on the records and shows where each result comes from, so it can be verified. That is more than can be said for the answer by intuition, which nobody ever verifies.
"Our data is confidential." It stays where it is, with the permissions it has. The assistant adds an interface on the same data, for the same people, under contracts and configurations that exclude the use of your data for model training.
"Do we need a new ERP first?" No. The assistant connects to the data you have, including a legacy system. It is often a sensible first step precisely because it returns value in weeks while leaving everything that works untouched.
What to take home
A concrete exercise: for one week, write down every question about data that comes up in the company and how long the answer takes. A long list with answers measured in hours or days points to a process worth automating. The starting point is small: the ten most frequent questions.
How many questions go unanswered in your company? Let's talk.