Anna Bondarenko
SAP CX and SAP BTP Consultant
Don't chase trends: how to understand if your business needs AI
Artificial intelligence (AI) is everywhere today. On one hand, you shouldn't implement it just for the sake of it or to follow the crowd. On the other hand, it's time for businesses to start experimenting. Ignoring AI can make you lose your competitive edge, especially in industries where data processing speed, automation, and data management are critical for success.
Let's discuss how to strike this balance and move from hype to real business value.
Step 1. Assess business and process readiness
Before buying or building any AI solutions, evaluate your company across four key areas:
If a process is inefficient or poorly documented, AI will only automate the existing problems rather than fix them.
Step 2. Evaluate the need for AI
Innovations make sense when you are dealing with routine tasks that demand speed, consistency and scalability. The most successful AI projects usually start with a single, specific task and scale up only after proving their business value To identify these opportunities, ask yourself: where do employees spend too much time on routine tasks or where do manual data processing errors lead to delays or penalties?
Of course, ROI (Return on Investment) should always be calculated. Modern AI solutions, from large language models to agentic systems, come at a cost. These expenses include not just tokens or subscriptions, but also integration, maintenance, data preparation and change management. However, you shouldn't compare the solution's cost solely to an employee's salary. You must also consider scalability, execution speed, output quality and error reduction.
Be careful with AI if:• Standard automation or process optimization can solve the problem better. • The task requires human judgment or a deep understanding of complex context. • The costs of implementation, tokens or waiting times far outweigh the potential benefits.
Step 3. Choose the right technology
AI is not a one-size-fits-all tool. Depending on your specific goals, different approaches work best.
If a process has strict rules and consistent input data, classic automation should be your first choice. AI is worth adding when you need to handle text, images, uncertainty, or complex exceptions. For instance, processing a customer complaint requires a standard automated check against shipping databases using classic algorithms. However, you can still use AI tools (like Claude or ChatGPT) to design the automation concept, build the solution architecture or write the code.
If you need to build AI directly into your product, here are the main options available today (keep in mind that technologies evolve rapidly):
Specialized business models (ready-to-use AI within enterprise platforms like SAP, Microsoft, Salesforce, etc.):
● When to use: when you have plenty of structured historical data in tables.● What for: classification or forecasting (predicting customer churn, analyzing late payments or generating recommendations).
Classic Machine Learning (ML):
● When to use: when you need absolute control over the model and lightning-fast response times. Large companies usually have ML already in place.● What for: predicting demand and production volumes, identifying sensor anomalies in manufacturing or ranking items.
Large Language Models (LLM) and Generative AI:
● When to use: when dealing with unstructured data (texts, documents, conversations). This also includes RAG (Retrieval-Augmented Generation), an approach that lets AI use corporate documents as a knowledge base, greatly reducing the risk of made-up facts.● What for: language understanding and generation. It's great for summarizing meetings, extracting data from scans, enabling semantic search or writing code.
AI Agents (Agentic AI):
● When to use: when highly complex tasks require multi-step reasoning and interactions across various tools or systems.● What for: coordinating multiple steps on behalf of users and connecting different systems. However, they need rigorous testing and control before deployment in critical business workflows.
Step 4. Estimate the expected impact
Define your target metrics before the project begins:
If you can't answer these questions, implementing AI is too early.
Before a full-scale rollout, run a pilot project on a specific process or department. This helps verify expected results, assess risks and prevent costly mistakes.
Step 5. Ensure data security and build trust in results
What to verify before using AI:
Watch out for hallucinations. Keep in mind that generative AI and agents can confidently provide incorrect answers. This is known as "model hallucination" — when AI generates believable but factually wrong information. That's why you must always verify the outputs or include a human validator in the loop, especially for finance, legal documents, healthcare or crucial business decisions.
Ultimately, not every business process requires artificial intelligence. The ideal solution is often a blend of AI and traditional automation. The secret to success lies in finding a real business problem, picking the most effective tool, calculating its ROI and rolling out the solution gradually and under strict control.
Let technology work for your business growth, not the other way around.