AI Beyond ChatGPT: Which AI Approach Fits Your Business Problem?

In short

If your question is what will happen next, think forecasting. If it is which case belongs to which category, think classification. If it is what should we suggest next, think recommendation systems. If you want to discover groups without predefined labels, think clustering. If you want to detect unusual behaviour, think anomaly detection. If the problem is mainly about text, documents or language, NLP or generative AI may be relevant.

Why starting with “we need AI” is usually the wrong starting point

“We need AI” is not yet a business problem. It does not tell us what decision should improve, what process should become easier, what customer outcome matters or what data is available. A strong project starts by translating a vague ambition into a question that can be measured.

For example, “we want AI for sales” could mean very different things: forecasting next month’s demand, predicting which leads are most likely to convert, recommending the next best offer, segmenting customers, detecting unusual purchasing behaviour or summarizing sales notes. Each of those questions points to a different modelling approach.

1. Forecasting and time series: what will happen next?

Forecasting is useful when the target changes over time and historical patterns matter. Typical examples include sales, demand, workload, inventory consumption, call volume or other operational quantities. The key requirement is not simply “having data” but having enough relevant historical data at a useful frequency.

2. Classification: which category or outcome is most likely?

Classification is appropriate when the desired output is a defined class or probability. Businesses may use it to estimate churn risk, identify likely converters, route support requests, detect a type of defect or prioritize cases for review. The model learns from examples where the correct outcome is already known.

3. Recommendation systems: what should we suggest next?

Recommendation systems rank products, content, services or actions according to relevance for a user or context. They can use behaviour, similarities, preferences or interactions. The business question is not “can we build a recommender?” but whether personalized ranking would improve a meaningful outcome such as discovery, conversion, retention or user experience.

4. Clustering and segmentation: what groups exist in the data?

Clustering is useful when you do not already know the categories. It can reveal groups of customers, behaviours, products or operational patterns. Unlike classification, there is no predefined label the model is trying to predict. The challenge is interpreting whether the resulting groups are meaningful and actionable for the business.

5. Anomaly detection: is something unusual happening?

Anomaly detection looks for observations or patterns that differ from what is normally expected. It can be useful for transactions, industrial measurements, operational events, quality control or user behaviour. A practical implementation also needs a process for deciding what happens after an anomaly is flagged.

6. NLP and generative AI: what can we do with text and language?

Natural language processing can classify documents, extract entities, route messages, perform semantic search and analyse text. Generative AI can add capabilities such as drafting, summarization or conversational interfaces. But not every text problem requires a large language model. Sometimes a simpler classification or extraction approach is cheaper, easier to evaluate and more reliable.

The model family is only one part of the decision

Even when the modelling family is clear, a useful AI project still depends on data quality, the cost of errors, integration with existing workflows, user adoption, privacy and the ability to measure success. The technically most advanced model is not automatically the best business solution.

This is why Data2Profit uses a business-first approach: clarify the question, assess the data and feasibility, then choose the technology. You can explore more examples on the Business Use Cases page or start with an AI Opportunity Diagnostic if you need to decide which opportunities are worth pursuing.

About Data2Profit

Data2Profit is an applied AI consultancy founded by Shahaf Klein in Grenoble, France. It helps organizations translate business questions into practical AI opportunities, from forecasting and recommendation systems to classification, clustering, anomaly detection, NLP and generative AI. Work is available in English, French and Hebrew, in France and internationally.