Sinequa


Founded 2001 | HQ Paris, France | 100 employees (approx.) | $25M revenue (est.)

The use of deep learning to enhance and improve complex search environments is impressive, and outside of major vendors such as Amazon, Google, and Microsoft, it’s rare to see. Sinequa has taken significant steps forward, and we will watch with interest as more customers begin to explore and use deep learning within their search platforms and applications.


The Company

Sinequa is a software vendor that focuses solely on enterprise search and specialized business applications powered by search. The firm was founded in 2001 and is headquartered in Paris, France. Alexandre Bilger has served as the company’s CEO since 2001. Sinequa has received a total of $28.3 million in funding over two rounds; the latest was for $23 million in 2019, led by Jolt Capital. We estimate the company’s revenues to be around $25 million and employee count to be around 100+. In addition to its headquarters in France, Sinequa has offices in the US, the UK, and Germany. This report focuses on the Sinequa Insight platform and, in particular, on its use of deep learning technology and the introduction of its starter/skills apps.

The Technology

The Sinequa Insight platform is an enterprise search platform or, as the firm prefers to call it, “Intelligent Search.” The intelligent element comes from the platform’s extensive and increasing use of artificial intelligence (AI) and machine learning (ML). Though most, if not all, enterprise search engines utilize some form of AI or ML, Sinequa significantly differentiates itself through its use of deep learning (artificial neural networks), and how it uniquely applies multiple deep learning models to provide more accurate search results.

What this means, in essence, is that the Sinequa Insight platform provides extremely advanced federated search capabilities (see Figure 1). The platform is used almost exclusively in high-end, complex, geographically dispersed organizations where, in theory, dozens of search engines can be consolidated into one. In October 2021, Sinequa announced the availability of starter/skills apps, pre-configured modules for specific business use cases to speed up and reduce the complexity of implementations for such activities as expertise location and research.

The roots of the technology (and the company) are in natural language processing (NLP), and Sinequa has been developing its NLP technology since its inception. Now, however, the company is rolling out the use of deep learning to enhance content classification and improve its understanding of the intent of queries. Sinequa expects to add passage-ranking and question-answering capabilities in the coming year. It should be noted that these deep learning features/modules are augmentations to the existing platform features rather than replacements.

Sinequa has been around for a while, and over that time has been built and tested at scale to run in complex environments. To that end, it provides a library of over 200 pre-built and managed connectors to different data sources. All of the connectors are non-intrusive, meaning that they do not change or modify the content or data you want to index. This library is an important differentiator, as – unlike some other enterprise search vendors – Sinequa provides the connectors all bundled together as part of the platform.

Though rooted in NLP, the Sinequa platform uses multiple text mining techniques in both the indexing and search query stages to provide a multilingual “semantic” search. In other words, it has an understanding of the meaning contained in the text rather than merely recognizing key words. As with any such multilingual system, the index’s quality and results will vary based on the language used. For example, the training data and activity in Portuguese may be significantly less than in English.

Most notable to us in our analysis is the use of deep learning in the Sinequa platform. Sinequa moves away from Apache Spark Classification (machine learning) and replaces it with Bidirectional Encoder Representations from Transformers (BERT) for neural networks. The BERT transfer method was first used for computer vision work, and its use to enhance text and language processing only started to be explored relatively recently. Using BERT, assuming it is used correctly, should produce higher result accuracy. Perhaps more importantly, it will require less data and time to train and, once trained, will detect much more complex language patterns than previously.

This year, the firm launched a module that aims to understand a query’s intent by understanding its context. It does this by factoring in such data as your location, previous activities, time, etc. In other words, though many searches are generic in structure (for example, “operating procedure”), by taking into account what it knows about the context for your search query, the system can try to understand your actual intention or goal and deliver a more refined result. Together, the improved and automated classification of data and better understanding of query intent should provide a noticeable and high-value improvement to the Sinequa search platform.

In short, Sinequa is a bit of a goliath in terms of search functionality and power, which in turn means it can be a bit of a beast to implement, hence the recent launch of the starter/skills apps. This first iteration has three such apps: basic search, analytics, and a usage monitoring dashboard. As the name suggests, these apps provide the ability to get up and running relatively quickly as they are pre-configured to meet many common needs. We expect more starter/skills apps to be added to the portfolio over the coming year; either way, they go a long way toward making such a formidable platform customizable and easy to use.

Named entity recognition and text mining analytics have been used in the Sinequa platform for some time. Over the years, it has extended to include two dozen instances of personally identifiable information (PII) to help meet legislation requirements such as Europe’s General Data Protection Regulation (GDPR), the California Consumer Privacy Act (CCPA), and the California Privacy Rights Act (CPRA). But more interesting to us is the release this year of its product called NLP Skills, a relatively simple mechanism to perform entity recognition and to detect relationships between multiple entities.

We regularly point out to our clients that enterprise search is highly complex, with many moving parts, and a successful large enterprise search deployment can be difficult to pull off. Hence, Sinequa works with a partner ecosystem to train and certify specialist third-party consultants such as Infosys, CapGemini, Accenture, and Cognizant. This is an important point, for though the Sinequa technology is impressive, it will require well-trained experts to build out and deploy an effective solution. It should be noted, though, that this is the same for any high-end search platform.

Figure 1
Image legend here

Our Opinion

The use of deep learning to enhance and improve complex search environments is impressive, and outside of major vendors such as Amazon, Google, and Microsoft, it’s rare to see. Though deep learning holds significant promise to improve the performance, accuracy, and relevance of enterprise search, it is early. Indeed, Sinequa is one of the first independent firms to embrace deep learning. They rolled out new modules such as named entity recognition and question answering this year and have plans to add neural search for natural language understanding in early 2022. Sinequa has taken some significant steps forward, and we will watch with interest as more customers begin to explore and use deep learning within their search platforms and applications.

Advice to Buyers

Sinequa is not a fit for everyone; this is a big, complex, and powerful search/insight platform. But for those in a large, geographically dispersed organization with critical search requirements, it should be on your shortlist. Moreover, your shortlist is likely to be very short: though there are around 60 enterprise search vendors in the market, few can compete at the high end in the way Sinequa can. Though we applaud Sinequa’s use of AI, you will naturally need to approach any engagement or purchase in conjunction with a highly skilled, and ideally Sinequa-certified, system integration team, and potentially make use of the new starter/skills apps to speed up the process.


SOAR Analysis

Strengths

  • Advanced federated search capabilities
  • Well ahead of the curve in using AI to enhance enterprise search

Aspirations

  • Dominate the high end of enterprise search
  • Accelerate international growth

Opportunities

  • Differentiate by using deep learning
  • Natively support multi-language user interfaces

Results

  • Well established in high-end search
  • Delivering cutting-edge deep learning for search and insights