A leading electrical giant needed to serve its sales and channel partners with accurate information about its complex product catalog. Ideas2IT joined forces with them to build a semantic product information retrieval system that served relevant catalog information just-in-time to the partners and clients. This significantly boosted their sales conversions and relevance of information
The project encompassed the establishment of a platform (powered by a centralized repository with an efficient information retrieval system) housing a diverse range of essential resources including documents, assets, catalogs, pricing sheets, and sales scripts of the catalog containing all the products.
This platform was to serve as a valuable tool for various stakeholders such as channel partners, sales teams, technical teams, marketing departments, and global stakeholders.
By offering seamless access and real-time outputs, the platform introduces efficiency into the workflow, rendering traditional communication methods like emails, escalations, and communication loops unnecessary. The incorporation of a user-friendly chat interface further simplifies interactions, bolstering convenience and collaboration.
Cut to the challenge: 2000+ categories. 120000+ products, all ranging anywhere between $1 to $1 Million in value. Information on any of these had become next to impossible to retrieve.
Pivotal stakeholders, like channel partners, sales consultants, price experts, and tech teams they worked with, all needed to find the right answers from this collection. But it wasn't easy – they needed to be sure they had the correct information, at the right time to make sure conversions improve.
This wasn't just about fixing a problem. It was about changing the way information was accessed. All this while ensuring that it had an extremely user-friendly interface.
The Way Forward
We established that the traditional information approaches will not be effective and it's a great opportunity for innovation leveraging the latest in LLMs. We wanted to create a solution that is not only accurate but can seamlessly scale as a catalog of products grow and new query requirements emerge.
- Develop an intelligent chatbot leveraging LLM’s Retrieval-augmented generation (RAG) to handle users’ product-related queries
- The chatbot should have self-learning and, auto-logging capabilities
- The chatbot is to be completely built on-premise by making use of open source stack of Dolly 12B model, Langchain, Bert Sentence transformer embeddings and FAISS
- The chatbot integrates with already-built search services by hitting their API without any need for middleware applications
- The products attributes are to be loaded into Graph db with relations, to maintain the hierarchy and traverse easily to find the related products
The Impact We Had - 74% Precision Enhancement In Search Results, 23% Sales Conversion Surge
- Amplifying Sales Conversion: Perhaps the most compelling marker of success was the 23% surge in sales conversion. As the sales teams did not have to depend on or wait for information on pricing, discounting, geography-specific data points, objections, or scripts, they were able to make decisions and close deals faster. This striking transformation underscored the real-world impact of our solution, translating into palpable business expansion.
- Redefining Efficiency: The process of mining documents for answers underwent a rapid transformation, thereby reshaping how information was relayed and tasks were executed. What once consumed hours was now accomplished within seconds.
- Elevating Relevance: The precision of search results witnessed a remarkable 74% enhancement. Our solution not only offered information but delivered the right information, minimizing uncertainty and fortifying decision-making.
How did we do it? Well, Let's read on to see.
The Solution: Precision Context through a Custom LLM
As a first step, we began by understanding the requirements of all pivotal stakeholders, who would be the actual end users of the centralized repository.
In line with the listed requirements, we envisioned an action plan to build a central repository platform for the client’s employees, relevant stakeholders, and channel partners to retrieve any information, as well as an intuitive chatbot interface for users to perform all these operations.
For this, a solution was developed by delivering a conversational generative chatbot that harnessed the capabilities of a custom Large Language Model (LLM) built on the foundation of Dolly 12B.
The system's approach to handling diverse document types exhibited meticulous preprocessing, wherein the distinct structures of each document, encompassing hierarchies, attributes, and textual content, were analyzed and accommodated.
To facilitate efficient information retrieval, the integration of Bert Sentence Transformer technology was employed to extract embeddings, which were subsequently stored within FAISS.
Notably, Dolly 12B underwent a process of fine-tuning; adapting it to the nuances of domain-specific documents and enabling comprehension and the generation of specialized terminologies.
This comprehensive solution ingeniously engineered prompts for various intentions and conversation trajectories too, enhancing the chatbot's conversational prowess. Moreover, the user experience was elevated through the customization of the chatbot's UI, culminating in an intuitive platform.
We Saw A Happy Client!
Given the complexity of the problem our client had and the objectives they wanted to address, Ideas2IT ensured that the collaborative efforts of various stakeholders played a pivotal role in the project's success.
From addressing the objectives to initiating a custom solution, we worked in tandem with our client to come up with a fully proofed solution that delivers impact and maps measurable business objectives.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Suspendisse varius enim in eros elementum tristique. Duis cursus, mi quis viverra ornare, eros dolor interdum nulla, ut commodo diam libero vitae erat. Aenean faucibus nibh et justo cursus id rutrum lorem imperdiet. Nunc ut sem vitae risus tristique posuere.
