For some reason, the whole discussion about corporate AI always boils down to one question: build it yourself, buy ready-made or order development? It's a trap. And while top managers are arguing about this, their projects are quietly dying without ever starting. Because this is the last question, not the first.
The real AI transformation begins much deeper. It starts from the foundation. And if that foundation is not there, then it doesn't matter which strategy you choose. You will simply choose the way in which your budget will be burned.
The first layer of the foundation: Data. There are none or they are scattered
Most companies are confident that they have a lot of data. In fact, they have a lot of information that lies in dozens of different systems, unrelated to each other.
and it is often unsuitable for analysis.
Let's take a typical case: a telecom company wants to predict customer churn. It would seem that the company's call data is stored in billing in the same format. The network parameters are in the monitoring system in another one. The activity in the mobile app is in its analytics. Complaints and requests to technical support are sent to CRM, often in the form of unstructured text. It will take six months just to put it all together, write integrations, clean up duplicates and errors. And then it turns out that location data needs to be difficult and expensive to depersonalize in order not to violate the law on personal data. The month scheduled for data collection has turned into six.
The second layer: Architecture. It will not withstand the new load
Let's say you've dug up and cleaned out your data. Where should I put them? Your current IT infrastructure is most likely not ready for the pressures that AI creates. It is built for scheduled monthly reports, not for real-time data flow analysis. This is called "technical debt."
A classic example is e-commerce. We decided to implement personalized recommendations. But as soon as the model started analyzing every click, every view, every abandoned shopping cart, the old databases just went down. APIs began to "fall off", and operational activity slowed down: updates to stock balances began to arrive with a delay, customers saw outdated information, and cash registers were suspended. Because analytical queries began to conflict with the main business processes. I had to rebuild everything urgently.
The right AI foundation is not just about more powerful servers. It is a flexible architecture that can withstand this new reality without breaking what is already working.
The third layer: Goals. Conflict of interest instead of synergy
Even if you have the perfect technique, the project may fail because of people. Because no one has agreed on what exactly he wants.
Behind the beautiful word "optimization", each department has its own KPI.
The task is to "optimize logistics" at the factory. Three months later, it turns out that everyone has their own understanding of this word. Logisticians looked at the dashboard with the delivery time. Financiers look at an Excel spreadsheet with diesel consumption. Sales — based on Customer satisfaction (SLA). IT specialists — for the cost of integration. These four worlds did not intersect in any way. The result is a conflict of interest, a frozen project, and a demotivated team that wasted a quarter.
The fourth layer: People. You can't find it, you can't grow it for a long time
It is almost impossible to find a ready-made AI specialist on the market who understands the specifics of your business from the start. The experience of the retail star, which worked with clickstream, is useless when optimizing a blast furnace, where data from thousands of sensors must be analyzed.
Therefore, readiness for AI is not about hunting, but about inner growth.
It is much more effective to select your best analysts and developers and invest in their retraining. Attract external consultants not as contractors who will do everything for you, but as mentors.,
who will work together with your team and pass on the expertise to it. This way, at least, knowledge and competencies will remain within the company.
And only now — about the strategy
And when all this invisible, hard work is done, you can return to the question "build or buy?". Now the answer will be informed, not based on fashion trends.
- Build Only if you have truly unique processes and you are ready to maintain an R&D department for years. These are the biggest risks and costs, including the risk that the entire expertise will be based on one or two key employees.
- Order (Custom) You will get a solution for yourself, but you will fall into the "vendor trap". Any change, any revision, is a new account. The cost of owning such a solution may be sky-high in a few years.
- Buy (Buy) Fast, but standard. You get a "boxed" solution that may not take into account all your nuances. Plus hidden integration costs and ongoing royalties.
This whole story is not about choosing software. It's about the company's willingness to make fundamental changes. All these pitfalls of AI are not technical, but managerial problems.
And until you figure out your data, architecture, goals, and people, any strategy for implementing enterprise AI is just an expensive lottery with very low chances of winning.
