The Data Dilemma: Two Worlds, One Challenge

A store manager opens her laptop. She needs last week’s sales figures, but she also wants to know why checkout times are lagging. The numbers sit neatly in a database, ready to be queried. The answers to the “why” are buried in PDFs of operational guidelines and customer feedback.

This is the reality for many businesses. Structured and unstructured data live in separate worlds. One gives you precise metrics, the other provides context. On their own, they only tell half the story. Combined, they unlock insights that drive smarter and faster decisions.

At Riverflex, we have seen this challenge repeatedly. From retrieval-based assistants for retailers to natural language interfaces for structured databases, our experience shows that the future lies in hybrid AI solutions that bring both worlds together.

Structured Data: The World of Numbers

Structured data is the orderly side of business information: customer records, sales figures, inventory levels, and KPIs. Traditionally, accessing this data required technical expertise or constant reliance on data teams.

New tools such as Databricks Genie are changing that. Instead of writing code, a business user can simply ask: “How many new customers signed up last quarter in Germany?” Genie translates the question into a query and delivers the answer.

As Mohammed Sarfarazuddin explained, “Databricks is offering something known as Genie, which sits on top of your structured tables. You can basically ask Genie a question, and under the hood, it’s nothing but a text-to-SQL solution.”

This approach empowers employees, accelerates insights, and reduces bottlenecks. But numbers alone never tell the whole story.

Unstructured Data: The World of Knowledge

On the other side lies unstructured data: policies, training manuals, customer feedback, reports, and all the other documents that capture business knowledge. These are the notes in the margins of an organisation. Historically, they have been difficult to access and even harder to use at scale.

This is where retrieval-augmented generation (RAG) comes in. In simple terms, RAG acts like a smart assistant that can read through documents, pull out relevant information, and provide answers with references back to the source.

Maja Podrug described a proof-of-concept project for a global retailer: “They had this knowledge base with all their retail expertise. We built a little chatbot where a user could ask any question, and it would pull answers straight from those documents.”

In just two weeks, thousands of PDFs were turned into a conversational assistant. Store managers could ask questions such as “What is the recommended layout for checkout counters?” and receive sourced guidance instantly. As Maja put it, “It was just a simple, textbook RAG implementation, but it opened the door to something bigger.”

The impact was immediate. Internal teams later developed the idea into a full production system, embedding it in day-to-day operations. What began as a quick experiment grew into a business tool shaping everyday decisions.

Why Hybrid Matters

Numbers tell you what happened. Documents and feedback explain why. Only by bringing them together can leaders see the full picture.

As Maja noted, “What’s interesting is the bridge between the two. That’s where it’s most powerful, when you use LLMs to structure unstructured data, and then combine both approaches.”

The potential is clear:

• A sales dashboard highlights declining performance, while a knowledge assistant surfaces staff feedback that points to a shortage of training.
• A compliance officer checks risk scores and can instantly compare them with policy requirements.
• An HR leader reviews headcount data alongside onboarding feedback to improve employee experience.

As Mohammed put it, “It completely depends on the use case. If an organisation has a lot of structured data, text-to-SQL is the way to go. But if they have a lot of documents, then RAG is the right approach. The future is combining both pipelines into one.”

Making Hybrid AI Work

Adopting hybrid AI is not just a technical exercise, it is a strategic decision. A few principles help ensure success:

• Integration: Systems need to know whether to answer with numbers, documents, or both.
• Trust: People will only rely on these tools if they can see why an answer was generated, whether through a query or a source document.
• Compliance: Enterprises often need in-cloud or in-tenant solutions to keep data secure and sovereign.
• Adoption: Embedding assistants in existing tools such as Teams or CRM systems ensures they become part of daily workflows.

The Future: Hybrid AI as the New Normal

The boundary between structured and unstructured data is disappearing. The most effective AI systems of tomorrow will not choose one or the other but integrate both. Structured data provides the facts, unstructured data provides the context, and together they deliver richer, faster, and more actionable insights.

This shift is not just technical but strategic. Leaders who embrace hybrid AI will empower their people, eliminate silo,s and move faster than competitors still working in one-dimensional data worlds.

Closing Thought

The question is no longer whether to invest in AI. The real question is whether you are making full use of both structured and unstructured data. Structured tools like Genie can tell you what happened. RAG assistants can explain why. The real opportunity lies in connecting them into a single experience that mirrors how leaders think: facts and context, numbers and narrative.

At Riverflex, we help organisations experiment quickly, scale responsibly, and design AI solutions that people actually use. If your business is sitting on mountains of structured reports and unstructured documents, it is time to ask: are you combining them, or leaving half of your knowledge untapped?

About the Contributors

Maja Podrug
Maja Podrug is a software and AI engineer with hands-on experience building practical applications of large language models. She has developed AI features for both internal platforms and client projects, including retrieval-based assistants that make unstructured knowledge accessible and useful in day-to-day operations.

Mohammed Sarfarazuddin
Mohammed Sarfarazuddin is a data and AI consultant with deep expertise in Platform Engineering. He specialises in creating robust platforms for data and AI engineering that enable organisations to scale their solutions effectively. His work focuses on unlocking value from structured data, making complex datasets accessible through natural language interfaces such as text-to-SQL systems, and designing hybrid approaches that balance speed, compliance, and business impact.