Creating a reliable search model for financial deal sourcing is more than just keyword matching—it requires understanding the context, intent, and nuances behind each search.

We’ve recently improved our search algorithm to tackle complex queries with greater accuracy and speed. While our AI search model was already considered one of the best in the industry, this update pushes the boundaries even further.

Why does this matter? Our database covers over 25 million companies (over 20 million more than other platforms) and all their details. With this upgrade, users can now find the most relevant results even faster.

In this article, Senior Software Engineer Joona Marjakangas explains what’s changed, the technology behind the update, and its impact on our search performance.
‍
‍

These improvements translate directly into better user experience: greater accuracy, faster searches and more confidence when exploring large datasets.

‍

Four Major Upgrades in the New Search Model

Before the update, our AI search model was already industry-leading, but we've taken it even further. The new model features innovations like Generation-Augmented Retrieval (GAR) for scalable LLM use, proprietary AI scoring for superior company similarity and domain-specific AI enhancements driven by extensive industry analysis. While these innovations might sound technical, let's now explore four key improvements in search quality and accuracy and what they mean in practice.

‍

‍
‍1. Better Understanding of Nuanced Queries and Context

The new model demonstrates a significant advancement in interpreting subtle differences between complex queries. For example:

  • “Venture capital funds focusing on B2B SaaS.”
  • “Venture capital funds focusing on early-stage technology startups.”

While these queries may seem similar, they target different types of companies The previous model occasionally overlooked these subtle distinctions. However, the updated algorithm now excels at recognizing context, consistently delivering results that are highly relevant to the user’s needs.

2.  Enhanced Business Model Recognition

The new search algorithm excels at understanding business models. For example, when searching for “manufacturers of industrial equipment,” the system will now ensure that results are limited to manufacturers—not distributors or retailers. This eliminates irrelevant results and improves the specificity of the information delivered.

3. Consistent Accuracy Across Broad and Niche Queries

In the past, the model excelled at niche searches but delivered less consistent results for broader queries. The latest model strikes a better balance:

  • Broad queries, such as “companies in agriculture,” now return a comprehensive set of relevant results.
  • Niche searches, like “noir detective game development companies,” continue to yield highly precise results.

This improvement provides more flexibility when exploring a wide range of opportunities.

4. Improved Long-Tail Result Quality

One of the most challenging aspects of search models is ensuring that result quality remains consistent, even as users move beyond the top-ranked results This is a common challenge in the industry, where top-ranked results are typically highly reliable, but the quality tends to drop as users explore further down the list. With our new algorithm quality holds steady, even when users are exploring beyond the top 500 or 1,000 results. This is crucial in deal sourcing, as the most valuable opportunities often reside deeper in the results.

‍

‍‍For highly specific, multi-criteria queries (e.g., 'Facebook marketing optimization SaaS tools'), accuracy saw a dramatic increase, rising from 41% to 85%, , the highest in the industry.‍‍

‍

Performance Results: Faster, More Accurate, and Reliable Searches

The latest upgrade offers substantial improvements in both speed and accuracy:

  • Top 10% accuracy improved from 70% to 93%, meaning that for highly ranked results, users now see far fewer irrelevant entries.
  • Easier queries, like “orthodontists”, saw accuracy rise from 86% to 97%.
  • More nuanced searches, like “supply chain optimization software”, improved from 67% to 88%.
  • For highly specific, multi-criteria queries (e.g., “Facebook marketing optimization SaaS tools”), accuracy increased dramatically, from 41% to 85%, the highest in the industry.

Speed improvements are equally significant:

  • Company searches are now 3x faster.
  • Investor and deal searches are up to 6x faster.

These improvements translate directly into even better user experience: faster searches, fewer irrelevant results, and more confidence when exploring large datasets.

‍

‍

Raising the Bar in Deal Sourcing Accuracy

With this development, we have set a new standard in the industry by delivering a whole new level of value to our clients by addressing one of the core challenges: accuracy. By improving precision, enhancing versatility across different query types, and ensuring consistent long-tail quality, we’re enabling users to spend less time sifting through irrelevant results and more time identifying the best opportunities.

And while these improvements represent a significant leap forward, we’re not stopping here. Continuous refinement is key to pushing the boundaries of what’s possible in deal sourcing.