AI Product Classification Solution at Scale

Unstructured Data

Over the years, the company has sold millions of technology products, ranging from software licenses and cloud services to hardware, networking equipment, and peripherals. The records of these sales were scattered across multiple systems, from SQL databases to countless Excel spreadsheets.

With thousands of people entering data over the decades, often as free text, in varying formats and levels of detail, consistency was difficult to achieve. The result was around eight million product records, but no reliable, standardised way to analyse or understand them.

Some entries were rich and detailed, with product names, codes, and metadata accurately recorded. Others were little more than shorthand notes, sometimes vague, sometimes incomplete, and occasionally completely unexpected (a “Barbie doll” had somehow made its way into the database).

Key Issues

This inconsistent data meant that the business couldn’t confidently answer some of its most important commercial questions, such as:

• Which products are driving revenue and margin?
• Which lines have been discontinued, and which are growing fastest?
• What product areas should the business focus on

For many companies running off legacy systems this lack of visibility can be a major limitation. The finance and sales teams were relying on gut instinct instead of trusted insight, which created risk and inefficiency.

Oakland was selected for its reputation as a trusted data and AI solutions provider. The team’s role was to bring order to complexity and create a scalable, AI-driven foundation for more informed, data-led decision-making.

“Working with Oakland made a huge difference for us.

“We had millions of product records over many different systems and very little consistency leading to poor data quality, and honestly, it was a mess. Ultimately, we had no structure, which led to no easy way to get insights. The Oakland team jumped in, rolled up their sleeves, and helped us turn all that chaos into a clear, organised AI solution (which was a first for us) that actually made sense for our business users.”

Head of Data

This is where an AI-driven data foundation could make the difference.

To unlock insight, Oakland needed to transform this vast, unstructured product database into a structured, navigable taxonomy.

Our data engineering and AI teams worked together to design a custom classification framework that mapped each product to the company’s five key business areas and drilled-down into more detailed categories beneath them.

A living, growing model rooted in data

Using large language models (LLMs) and a “human-in-the- loop” approach, we trained an AI system to read each product entry and decide what type of product it was, grouping similar items into meaningful, hierarchical categories.

At the highest level, products were grouped into hardware, software, or services. From there, the AI created more specific subcategories such as edge devices, compute, networking hardware, or enterprise applications.

Crucially, the system wasn’t limited to pre-defined labels. If it encountered a new kind of product, for example, 3D printers, it could automatically create a new category and slot it into the taxonomy. This adaptive classification ensured that the structure could evolve along with the business.

The result was a taxonomy that reflected the client’s real-world view of their product universe, not a generic industry template, but a living, growing model rooted in their data and their language.

Human expertise meets machine intelligence

While the AI handled the heavy lifting, human expertise remained vital. Oakland’s data and AI consultants collaborated closely with subject matter experts from across the business, from sales to finance and product management, to review, refine, and validate the AI’s classifications. This human-in-the-loop approach ensured that the final taxonomy not only made technical sense but also made business sense.

Instead of asking people to manually tag thousands of individual records, a process that would have taken years, we gave them a structured, explainable model to review. Adjustments could be made in hours, not months.

Scaling to production

Once the taxonomy was validated, Oakland built a scalable pipeline to classify hundreds of thousands of product records quickly and accurately.

We started with a sample of 20,000 products to train and test the model, then scaled to 300,000 records covering three full financial years, representing the majority of the company’s recent revenue.

The solution was deployed in a secure cloud environment using Azure’s OpenAI service, with custom tooling built by Oakland to manage data flow, compute requirements, and error handling.

This allowed us to process vast amounts of unstructured data in parallel, achieving what would previously have required years of manual effort in a matter of hours.

AI-driven efficiency and ROI

The time and cost savings were eye-opening.

Running the AI model across 300,000 product records costs less than £100. By contrast, a human team classifying the same volume manually would have taken around three years, not to mention the mind-numbing tedium of manually tagging thousands of product lines (precisely the kind of monotonous work AI was made for).

That’s a 1,000x improvement in speed and efficiency.

“Oakland didn’t just throw tech at the problem, instead choosing to work alongside our IT and business teams, making sure the solution was fit for what we really needed. Their mix of AI know-how and practical business sense meant we got results fast. Now, we can actually trust our data, answer important questions, and spot new opportunities in our product data.

“Oakland set us up with a solid foundation for our new LLM, helped us with all the documentation and handover of this new solution, and we’re in a much better place to move forward with confidence in our product data thanks to them.”

Head of Data

Your Turn

Find out how we can develop an AI-driven data foundation to transform your commercial activities by speaking with our friendly team.