In 2021 I was head of IT at Civil Mağazacılık. Hundreds of stores, a data centre, ERP, business intelligence, a cloud migration we built from scratch. Ten years. Asking the suppliers sitting across my desk "why did you do it this way" was part of the job. The same year I founded CX Teknoloji and moved to the other side of that desk. Looking back five years later, what I see most clearly is this: the technology was the easy part.
Having been the customer is an advantage, but not as much as you'd think
I knew what made an IT manager angry: the "extra licence" that appears at installation, the server not in the quote, the feature that works in the demo and not in the field. We built the product around those: no extra hardware, video does not leave the site, the acceptance criterion is in the contract. That was right. But knowing what the IT manager wants was not the same as knowing how the purchasing process works. The decision-maker is the IT manager, the budget-holder is finance, the user is store operations or the safety team. Without convincing all three at once, the project does not move. Learning that took a year.
The product decision: what we will not do
In the first months we said yes to every request. Face recognition, plates, counting, quality inspection, drones. Six months later we had six half-finished things. We sat down and made three decisions: no face recognition, no hardware sales beyond the existing camera, video processed on site. Those three "no"s defined the product. If today there are 23 ready solutions in five categories, they all exist because they stayed inside those three boundaries.
The technopark
The concrete benefit of being at GTÜ Technopark is the proximity to the university: working with graduate students on the computer-vision side, access to GPU infrastructure, regular tracking of R&D incentives. The concrete difficulty is that most customers are in Istanbul and Gebze is "close but far". Two days a week on the road.
Living with an accuracy figure
In the enterprise world a system either works or it does not. In AI, 96% works. Telling a customer "this system will miss 4 people in 100" was hard at first. Then we saw the opposite: a quote that states the accuracy rate openly and puts the measurement method in the contract inspired more trust than the competitor saying "one hundred percent". IT managers recognise exaggeration; so did I.
First customer, first mistake
At our first retail customer we installed people counting and the first week came out at 88%. The model was fine; the camera was looking head-on and staff passages were being counted. The customer was patient and we practically lived in the store for three weeks. Every one of our calibration notes today comes from those three weeks. That customer is still with us and gives references. Lesson: spend a lot of time with the first customer; the product is written there.
The team
In enterprise IT there is a team, there are job descriptions, there is process. In a start-up, for the first two years everyone does everything; I climbed the ladder in the warehouse for installations and adjusted camera angles myself. That is necessary, but at some point it has to end. In year three we separated engineering, field and sales. After separating we grew; before separating we were going in circles.
One habit from the enterprise world paid off: writing process down. Commissioning checklist, acceptance-test protocol, event schema, version management. In start-ups these count as "bureaucracy"; for us they were the thing that saved the day after an update broke 12 stores at once.
Do COBIT, ITIL and PMP help in a start-up?
Directly no, indirectly yes. No customer ever asked about certificates. But once concepts like change management, incident management and scope definition are settled in your head, there is less chaos even in a small team. And we speak the same language as the customer's IT team; when I say "let's open a change request" the other side relaxes.
Awards
The CIO awards are from the enterprise period. On the start-up side references matter more than awards; the number City Lojistik measured in warehouse safety opened more doors than any plaque. But the awards had one use: in the first year, when nobody knew us, "this man ran the IT of a large chain for ten years" got the meeting started.
Five years on
Today we have systems running in plants and stores in Türkiye and Europe. Computer vision is no longer "interesting technology"; it is a line in the safety manager's monthly report, the screen the store manager looks at in the morning. Becoming ordinary is good; when the product becomes ordinary, selling gets easier.
What I think about for the next five years is the camera merging with other sensors. A camera alone is a panel; joined with PLC, POS and sensors it becomes the operation itself. I cover the technical side of that in other articles. This one was for the non-technical side: ten years as a customer helped, but was not enough. The rest was learned in the field.