Bringing Deep Research to Enterprise

Introduction#

Recently, the AI landscape was reshaped again when OpenAI released Deep Research.

This was closely followed by similar releases from Perplexity, xAI, Manus, and others. At its core, Deep Research is a search agent that can take a user query and conduct complex, multi-step research on the internet for up to 30 minutes, sometimes using over 100 sources. It outputs a comprehensive research report in minutes—work that would typically take a human researcher several hours to complete.

We think it’s one of the most significant AI releases since GPT-3.

In the conversations that have followed with our clients, one question stands out: “When will we be able to apply this kind of deep research to our own data?”

Many of these clients have spent years - or even decades - meticulously crafting some of the world’s largest proprietary knowledge stores. This data is invaluable and fuels their competitive edge in their respective industries.

Recognising this need, we set out to develop an approach that enables every business to build deep research capabilities tailored to their specific market, data, and context.

We call this solution Enterprise Deep Research.

Bringing Deep Research to Enterprise#

In Enterprise Deep Research, we integrate many of the learnings we talked about in our deep dive on RAG approaches.

• Iteratively building a better user query is even more critical in deep research tasks than in classical RAG
• Context management is still king. It is critical that the Research Agent has a good understanding of what it has explored so far and which research directions were more fruitful. By managing this state intelligently, the reasoning in the model can work most effectively

Research Agent#

The Research Agent is the engine that powers Enterprise Deep Research.

The Research Agent has visibility of the user question, queries that have been executed before, and what useful information they retrieved. It then uses reasoning to translate its curated data feed into actions.

The Research Agent is given access to many tools to help it accomplish its goal. This can include web-search and web-operator tools that are common in consumer-facing deep research tools, but also includes tools to interact with your proprietary data sources and endpoints.

This is analogous to a real-life toolbox - the Research Agent just has to pick the right tool for the job. Sometimes a web search is the most useful tool— you might want to quickly look at a research paper that was released today, and it might not have made it into your company’s knowledge base yet.

However, if you are working in industries such as pharmaceuticals, finance, or law, your workflow will be heavily dependent on your internal knowledge bases. For these sources, we can carefully tune the Research Agent with end-to-end reinforcement learning so that it can traverse your proprietary knowledge most effectively. And that’s where Tomoro’s Enterprise Deep Research really shines.

Integrating your proprietary data#

This is where the magic happens.

While the internet offers vast amounts of information, embedding your proprietary expertise directly into AI systems unlocks their full potential. At Tomoro, we believe that the most effective solutions come from integrating your unique knowledge and competitive strengths at every step of the process.

We work collaboratively with organisations to develop tailored integration solutions that bring together data from various sources — whether it’s structured databases like Neo4j, PostgreSQL, or Snowflake, or cloud storage and file systems such as S3 or SharePoint — into a unified ecosystem. This integration helps ensure that every facet of the system is enriched by your proprietary data and internal processes. This enables the Research Agent to follow the same steps that your internal analysts do.

Efforts to effectively harness data often raise critical questions around access rights and entitlements. Data owners are understandably concerned about the potential of sensitive information being inadvertently exposed in responses. At Tomoro, we believe in building privacy and data security into our solutions from the ground up. When indexing your internal data, it’s essential to capture data ownership and enforce entitlements, ensuring that only authorised users can access specific information. Furthermore, robust mechanisms to purge data from the index — compliant with GDPR and other regulations—are vital for maintaining privacy and regulatory compliance. This level of control and safety distinguishes our approach from public, internet-wide deep research tools.

Web search & other tools#

Web search remains a cornerstone of enterprise research workflows. Traditional web scraping has long been a dependable way to harvest structured data from online sources. These classical methods are effective for collecting large volumes of public information.

Web operator agents have recently emerged to handle the challenging data retrieval tasks on the internet. This includes OpenAI’s Operator, Anthropic’s Claude, and Convergence’s Proxy. They are designed to navigate complex web environments, tackle nuanced queries, and retrieve data that might be out of reach for standard scraping techniques.

Sometimes, LLMs aren’t the best tool for analysing data that has been retrieved. So, we can give the Research Agent access to more specialised tools such as a coding environment or even your proprietary tools to complete these analyses. This layered approach — from classical scraping to advanced operator agents and specialist tools — ensures that your enterprise research is both thorough and adaptive, no matter how intricate the data landscape may be.

Tailored research outputs#

Every enterprise has its own culture, industry nuances, and established practices that influence how insights are best consumed and applied. Off-the-shelf tools often deliver one-size-fits-all answers that may not align with the unique context of your business.

By creating a tailored deep research solution you have the flexibility to format the research findings exactly where and how they’re needed — whether that’s integrating them into existing analysis tools or creating a bespoke workflow just for this.

By aligning the research outputs with your business’ specific language and processes, you ensure that the insights are not only accurate, but also seamlessly integrated into your existing decision-making frameworks.

The path forward#

We envision a future where every company benefits from a tailored deep research solution that powers its most critical workflows.

Recent launches—such as OpenAI’s Responses API, which offers integrated web search capabilities, alongside their newly unveiled Agent SDK for building agentic AI systems — signal industry leaders’ commitment to advancing AI tooling in this space. While public tools like OpenAI’s Deep Research excel at addressing general queries, enterprises seeking a competitive edge will benefit from a bespoke approach that fully leverages their proprietary knowledge and unique strengths.

Interested in exploring how this vision can work for you? Reach out to us via your preferred channel.