When a user lands on an ecommerce portal with hundreds of thousands of product references and types into the search bar, they have only a few seconds of patience. If the search engine doesn't return what they're looking for quickly and relevantly, they leave. And that abandonment costs real sales.
This is the problem BigMat brought to us. And the solution was BigFinder, the semantic search engine that TBE developed in-house and that today handles millions of queries per month on their digital portals in under 100 milliseconds on average.
The challenge: massive catalogues and a search engine that wasn't up to the task
BigMat is one of the largest construction and DIY materials distribution cooperatives in Spain. Their ecommerce portals — BigMat La Plataforma and BigMat Digital — bring together catalogues with hundreds of thousands of references: from fasteners to machinery, including finishing materials, plumbing, tools and much more.
The problem wasn't the volume of products. The problem was the search.
The platform's native search engine operated on exact string-matching logic. If a user searched for "mason's rule" and the catalogue had the product listed as "aluminium rule for construction", it wouldn't be found. If they typed with a typo, the result was empty. If they used a colloquial trade term, nothing.
The impact was direct: high abandonment rate after searching, users who couldn't find what they needed even though the product existed in the catalogue, and a measurable and frustrating loss of conversion.
SaaS Finder solutions on the market were evaluated as an alternative. The problem: their per-search or per-indexed-catalogue-volume billing model generated a cost that grew in direct proportion to usage volume. At BigMat's scale, the numbers didn't add up.
The solution: BigFinder, a proprietary engine with semantic intelligence
TBE made a clear decision: develop a proprietary search engine. It wasn't the fastest option, but it was the smartest one in the medium term.
The result was BigFinder, a solution built on three technological pillars:
- Elasticsearch as the indexing and information retrieval engine
- Docker for containerisation, deployment and horizontal scalability
- A proprietary semantic intelligence layer to understand search intent beyond exact words
Integration with BigMat's portals was carried out via REST API, without needing to migrate the existing ecommerce platform. BigFinder acts as a transparent search layer: it receives the query, processes it semantically and returns results sorted by relevance in milliseconds.
How the technology behind BigFinder works
Elasticsearch: indexing at real scale
Elasticsearch is the world's most widely used distributed search and analytics engine for use cases that require speed and scale. Its architecture allows indexing hundreds of thousands of documents (in this case, product references with their attributes, descriptions, synonyms and technical metadata) and performing full-text searches in near-instantaneous time.
For BigMat, the index was built with the attributes that matter most in the sector: manufacturer reference, EAN, trade name, technical description, category, sector-specific synonyms and common alternative terms used in the language of construction professionals.
Elasticsearch also allows hot reindexing: when new products are added or existing attributes are modified, the index is updated without interrupting the search service.
Docker: scalability without friction
BigFinder's infrastructure is deployed in Docker containers. This solves several problems at once:
- Reproducible environments: the same stack works identically in development, staging and production, eliminating the classic "it works on my machine" issue.
- Horizontal scaling: during peak traffic moments (campaigns, industry events, mass access), additional nodes can be spun up in minutes without touching the system configuration.
- Zero-downtime updates: the containerised deployment model allows rolling updates that do not affect service availability.
For a portal of BigMat's volume, where traffic peaks can multiply the usual load several times over, this elastic scaling capability is not a luxury: it's a requirement.
The semantic layer: understanding what the user means
This is the part that turns a fast search engine into an intelligent one.
Traditional search compares text strings. Semantic search compares meanings. BigFinder uses embedding models that transform both the user's query and the catalogue documents into high-dimensional vectors representing their semantic content.
The practical result: if a BigMat user searches for "tile adhesive paste", BigFinder understands they are looking for ceramic adhesive, and returns the correct results even if no catalogue reference uses those exact words. If they type "angle grinder" with a typo, the engine understands it. If a professional uses the trade's technical term instead of the product's commercial name, it finds it all the same.
The semantic model was trained and fine-tuned specifically for the vocabulary of the construction and DIY sector, which substantially improves precision compared to generic models.
The results
After deploying BigFinder on BigMat's portals, the numbers speak for themselves:
The improvement in result relevance had a direct impact on the conversion rate: users who search and find, buy. Users who search and don't find, leave.
In addition, the Docker architecture allowed BigMat's infrastructure team to manage seasonal campaign traffic peaks without permanently oversizing resources. Infrastructure costs were optimised significantly.
Why custom development over a SaaS solution
The legitimate question is: why not use an already-built SaaS Finder?
For specific use cases — a small store with a limited catalogue and no large traffic volumes — SaaS solutions make sense. The setup is fast and the initial cost is low.
But the billing model of these solutions has a structural problem: the cost grows with usage. They charge per search, per indexed product, or both. In a portal like BigMat, with hundreds of thousands of references and millions of monthly searches, that model generates a recurring bill that far exceeds the cost of a custom development amortised in just a few months.
With BigFinder, the scheme is the opposite:
- A fixed implementation project with a known cost upfront
- A flat monthly maintenance fee, independent of volume
- No per-search royalties, no limits on indexed products, no surprise costs
- Full control over the code, the semantic model and the product's evolution
The greater the volume, the more obvious the economic advantage of custom development over SaaS.
At BigMat's scale, BigFinder paid for itself in less than a year compared to the cost of an equivalent SaaS solution. And from that point on, the savings are monthly and cumulative.
Conclusion
BigFinder is the result of a real problem solved with proprietary technology. Elasticsearch provides the indexing and retrieval power. Docker guarantees scalability and reliability in production. And the semantic layer converts queries in human language into precise results, even when the words don't match exactly.
For BigMat, the impact has been real and measurable: more searches handled, in less time, with greater relevance and at a cost radically lower than market alternatives.
If you have an ecommerce site, a marketplace or a distribution portal where search is a critical point of the user experience, BigFinder may be the solution.