Retour

iM Global Partner – Responding to RFPs using generative AI

iM Global Partner – Responding to RFPs using generative AI

The transformation of tender processes through AI demonstrates how technology can create a competitive advantage in the financial sector.

KEY FIGURES

  • 50% increase in number of tenders processed

  • 30% improvement in success rate

  • 38 billion dollars in assets under management

ABOUT THE CLIENT

A global leader in asset management, iM Global Partner manages a portfolio of 38 billion dollars across 16 locations in Europe and the United States. Its specialization in strategic partnerships with independent management companies makes it a reference player in the international financial sector

CHALLENGE

In a sector as competitive as asset management, the ability to effectively respond to tenders is crucial. iM Global Partner faced a significant challenge: processing an increasing volume of multilingual documents while maintaining impeccable response quality. The precise extraction of questions and building a robust knowledge base from existing documents required an innovative solution capable of adapting to the specificities of the financial context.

SOLUTION

At the heart of the system, an intelligent matching process combines the power of embeddings for document vectorization with the capabilities of generative AI Mistral Nemo. This alliance automatically identifies the most relevant matches between tenders and financial products.

For RFP (Request For Proposal) processing, we implemented an intelligent RAG (Retrieval-Augmented Generation) system. This system deeply analyzes questionnaires, searches for relevant information in the knowledge base, and generates contextualized responses. The entire platform is deployed directly on iM Global Partner's infrastructure, thus ensuring the security and confidentiality of sensitive data.

Intelligent AI Matching

  • Use of embeddings for document vectorization

  • Automatic matching via generative AI (Mistral Nemo)

  • Enriched and contextualized knowledge base

Automated RFP Processing

  • Analysis of tender questionnaires

  • RAG (Retrieval-Augmented Generation) system for relevant information search

  • Generation of contextualized responses

Security and Compliance

  • On-premise deployment on client infrastructure

  • Protection of sensitive data

  • Complete control over information processing

EXAMPLE PROMPT TO NORMALIZE RFPs

The strength of the following prompt lies in its combination of different prompting techniques, creating a particularly effective tool for analyzing financial tenders. At its core, the design includes few-shot learning which uses a concrete case to illustrate exactly the expected format. This approach is reinforced by methodical structuring in steps, following a chain-of-thought principle that guides the analysis in a logical and sequential manner. The prompt also relies on a system of explicit constraints and precise specification of the JSON output format, ensuring standardized and easily exploitable results. The whole is framed by a clear definition of role and context, ensuring that the analysis always remains aligned with the specific objectives of financial tenders.

This multifaceted approach thus creates a robust and precise system, capable of efficiently processing tender responses while maintaining a high level of standardization and reproducibility, essential elements in the financial sector:

Objective

Read and analyze documents to identify key investment parameters according to four main categories:

  1. Asset Classes

  2. Investment Styles

  3. Capitalization Categories

  4. Geographic Focus

Analysis Steps

  1. Asset Classes

  • Examine references to asset classes

  • Note all identified asset classes

  • Must match the list of specified values

  1. Investment Styles

  • Examine references to investment strategies

  • Note all identified styles

  • Must match the list of specified values

  1. Capitalization Categories

  • Examine references to market capitalization

  • Note all identified categories

  • Must match the list of specified values

  1. Geographic Focus

  • Examine references to geographic regions

  • Note all identified locations

  • Must match the list of specified values

  1. Optional Parameters

  • Identify optional preferences or acceptable alternatives

  • Note items that are:

    • Preferred but not mandatory

    • Acceptable but not priority

    • Permitted as secondary options

  • Categorize optional items within the four main categories

  1. Handling Ambiguities

  • Ignore any unclear or ambiguous mentions

  • Include only clearly identifiable items

Output Format

The result must be formatted in JSON with the following categories: