In recent years, AI has made its way into every company and every boardroom, most of us can create texts or ask a chatbot for answers. But AI initiatives don’t always get off the ground. To use an aviation metaphor, the reason some AI initiatives fail to fly is not because the AI or engine itself is weak, but because its “wings” (business analytics and decision making) aren’t strong enough to carry it.
Three disciplines working as one
In practice, everything starts with BI, which provides a unified data view, defined metrics, and trust in the data. Only then can AI be deployed to add a future outlook to the view of the past. But real business value only comes when you complement these with DI by creating decision-making rules, adding data-based automation, and assigning accountability.
While BI and AI answer the questions “What happened?” and “What happens next, and why?” respectively, DI answers the question “What are we going to do about all this, and what are we going to do next?” If any of those three links in the chain is weak, the plane will lose its stability or heading.
The three levels of AI capabilities
Whether AI is deployed depends on the business problem, the initial data, and the time horizon.
1) The capabilities of language model-based AI
Language model-based AI is the right choice if your issue is how to explain or structure something, or how to communicate. The solutions at this level include chatbots, text construction, creation of briefs, and personalized communication. As a rule, the output is then texts, answers, or summaries with references. References are needed for trust—they reduce the risk of AI “hallucinations” or confabulation.
2) Predictive and explanatory models based on machine-learning models
Predictive capability is appropriate to use if the question you are asking is “What will happen and why?” Business examples include predicting the likelihood of a customer leaving or signing a contract, detection of fraud, demand forecasting, and pricing models. At this level, the output is scores or forecasts with quantitative explanations, such as a regression equation explaining the impact of different factors on the overall result.
3) Hybrid AI capabilities combining the previous two
Solving a business problem often requires a combination of both: first understanding who or what you are dealing with, then an ability to communicate.
To illustrate this point, let me give the business example of reducing customer turnover.
As mentioned above, first we need BI—or in aviation terms, a runway. We start with a commonly held definition of a key metric, for instance the 60-day rate of customer turnover, then build a reliable data layer and dashboard to track it. This establishes the presumption that all team members will see the same outcome, creating clarity and stability.
In order to fly, you have to take off, and for that we need a predictive AI model that forecasts the likelihood of customers leaving. By comparing data and using the model, we can discover which conditions have the biggest impact on turnover, such as limited use of a service in the preceding 30 days, concerns about a product, or late payments by a customer. We can also use these explanations down the line to automate communication.
Once we are in the air, we need to maintain our heading and communicate—which is where the language model comes in, generating personalized messages. For example, device support is offered in the case of a drop-off in use, pricing advice is provided for price-sensitive customers, and automated messages are implemented when customer support is contacted. These messages are based on explanations and, where needed, refer to articles within the knowledge base.
Lastly, in order to make this action repeatable throughout the customer’s lifetime, we also need to bring in DI—our autopilot. To do this, we can define decision points and rules that trigger actions automatically. Automated agents create the corresponding tasks in the company’s business software, responsibilities are assigned, and outcomes are measured.
In summary, BI maintains clear visibility and ensures data-driven decision-making, predictive AI tells us “Who with and when,” language model-based AI says “What and how,” and finally DI makes it all happen.
Comparison of AI Capabilities
| Layer | Objective | Output | Question |
| BI (left wing) | Understanding + a universal truth | Reports, KPIs, quality rules | What happened? / What is the current status? |
| Predictive AI | Forecast + reason | Scores, forecasts, explanations | Why is it happening and what will happen tomorrow? |
| Language-model AI | Synthesis + communication | Summaries, chatbots, text processing | How to explain/create? |
| DI (right wing) | Decision → Execution | Decision rules, workflows, automation | What are we doing: who and when? |
Get in touch
