Skip to main content
CUSTOMER STORY

Supporting customers with next-generation financial services

MNP uses Databricks Mosaic AI to deliver business insights tailored to clients’ needs

<6 weeks

To GenAI solution deployment

SOLUTION: RAG Applications
PLATFORM USE CASE: Generative AI
CLOUD: Azure

“Because of Databricks state-of-the-art architecture, we were able to develop a GenAI solution that delivers accurate responses in less than six weeks.”

— Colin Wenngatz, Vice President, Data Analytics, MNP

To strengthen their market leadership in professional services, MNP chose to adopt a lakehouse architecture that could support the integration of large language models (LLMs) and provide contextualized insights to their clients. Because of the complexities involved, the national accounting, consulting, tax and digital services firm partnered with Databricks to build a client-focused platform to enable data-driven decision-making across various industries.

Data limitations impede the rollout of advanced client services

MNP, a leading Canadian professional services firm with nearly 9,000 team members and 127 offices nationwide, wanted to modernize their conventional data and analytics platforms to better enable the new capabilities they wanted to implement. Most importantly, MNP’s systems couldn’t support the processing required to accommodate the advanced data analytics and machine learning (ML) applications like the language-based prototype it was fine-tuning on Llama 2 13B and 70B.

MNP faced several roadblocks in their initial GenAI deployment. First, the foundation model’s tight integration with their existing data warehouse hindered experimentation with other solutions. Moreover, the “time-to-first-token” evaluation metric caused lag issues, negatively affecting user experience. Finally, the total cost of ownership (TCO) became prohibitive due to significant GPU usage from the large data volume required for a reliable corpus and frequent retraining and serving of the model. Add in the complexity of MNP’s diverse clients, and the organization was at a crossroads with their data and AI approach.

Despite these challenges, MNP remained committed to their goal of enabling clients to uncover insights and make informed and data-driven business decisions. Colin Wenngatz, Vice President of Data Analytics at MNP, explained, “We work with everyone from smaller owner-managed businesses to some of Canada’s largest national and multinational enterprises. We want to ensure we equip all our clients with the information they need to make their businesses successful. We view our role as the trusted adviser for our clients. Given that their industries and needs vary, we needed an automated process to create reports, calculate key ratios and empower everyone to ask critical questions.”

Aside from facilitating generative AI innovation, a new data lakehouse architecture would improve data processing speeds and help resolve limitations with large data queries. The right solution would also handle the firm’s large datasets, scale with their growing data needs and remain adaptable to future technologies and methodologies. By choosing Databricks, MNP aimed to gain control over their analytical roadmap and decisions, offer more-advanced services to their clients, and further enhance their competitive position in the market.

Using LLMs to help clients understand complex data

MNP’s partnership with Databricks focused on leveraging advanced AI to transform how the firm delivered services to Canadian private enterprises. To support their clients in their continued growth journey, MNP aimed to simplify data analysis to ultimately provide faster, more contextual and precise client insights.

The Databricks Data Intelligence Platform served as a unifying force to consolidate MNP’s structured, semi-structured and unstructured data into a single repository. The new approach simplified data governance, management and security in handling business data. Additionally, it allowed MNP to process and analyze data as it became available, ensuring client insights were based on the most current data.

The solid data foundation MNP built with Databricks would also allow it to adopt the Vector Search feature, a sophisticated repository for data embeddings that optimizes data formats for rapid AI processing. This new addition to the Databricks stack is a highly performant serverless vector database with built-in governance. It will help the firm’s clients and client-facing teams pinpoint the most relevant information for specific queries. Vector Search provided seamless access to optimized data, crucial for deploying LLMs, specifically MNP’s eventual choice of Mixtral 8x7B.

A mixture-of-experts (MoE) model like Mixtral 8x7B would enable MNP to tailor the model for context-specific sensitivity, broaden the parameters to accommodate a wider range of instructions and enhance parallel processing to support simultaneous workloads. After the firm completed their model evaluation and selection, their data team selected retrieval augmented generation (RAG) as their preferred refinement strategy. This choice was influenced by the need to work with datasets that required regular updates or additions.

The ability to continually integrate new data, refresh the embedding database and employ the information to provide contextual relevance was fundamental in MNP’s strategic initiative to deploy RAG within the Databricks Data Intelligence Platform. MNP recognized the potential of LLMs and their summarization capabilities and wanted to offer an intuitive, language-based interface for clients to make insightful, data-driven decisions about their businesses and industries.

Databricks Foundation Model APIs helped MNP monitor the essential components for deploying and managing LLMs. Acting as a conduit between the raw computational power of the lakehouse infrastructure and the end-user application, it ensured the models deployed and interacted smoothly. This was especially crucial for the new RAG approach, which required integrating retrieved data into the generative process in real time.

Mosaic AI Model Serving was also implemented to ensure that the models were constantly available and responsive to queries, manage the allocation of resources and maintain performance during peak usage times. Each component worked together thanks to help from the Databricks GenAI Advisory Program. The initiative began as a resource to help customers adopt GenAI technologies effectively.

“The GenAI Advisory Program has been invaluable to our early successes with Databricks,” Jason Aird, Senior Manager of Data Engineering at MNP, and the program’s technical lead, said. “The support we’ve received has vastly accelerated our path to deployment and adoption.”

Transforming data handling for efficient insights

Integrating Databricks lakehouse architecture with MNP’s data warehouse systems facilitated a shift from traditional data handling methods to a more dynamic, efficient model. This necessary transformation was crucial in addressing MNP’s challenges, such as managing large volumes of complex data, providing client-facing teams with resources for timely and accurate insights and building models that helped clients with discoverability. It also allowed the firm’s data team to build a model from start to quality assurance testing within four weeks.

With a stable data environment, the adoption of LLMs and RAG applications within the Databricks Data Intelligence Platform ensures that MNP’s data remains isolated and secure via “Private AI” standards. More importantly, the firm has a concrete understanding of how foundation models are trained and on what data. This approach, powered by Unity Catalog, enabled MNP to maintain the highest levels of information security while optimizing their AI capabilities.

Wenngatz remarked, “The impact of Databricks technology enabled us to deliver enhanced services to our clients, demonstrating the powerful relationship between advanced AI and effective data management.”

Looking forward, MNP is evaluating DBRX, an open, general-purpose LLM created by Databricks, as their foundation model. DBRX also offers a fine-grained MoE architecture, significantly improving ML training and inference efficiency. With 132 billion total parameters and 36 billion active parameters per input, DBRX can handle complex data tasks with high precision and speed. It is expected to provide improved model-serving capabilities through Databricks Foundation Model APIs. The continued collaboration with Databricks positions MNP to remain at the forefront of utilizing AI-driven data solutions within the Canadian accounting and business consulting sector — further enhancing their service offerings and boosting their bottom line.

About MNP and Databricks

MNP Digital’s Applied Data team has partnered with Databricks to support the development and implementation of enterprise-grade large language models for the Canadian business market. Clients who work with MNP benefit from the industry-leading insight of both firms. To learn more about MNP Digital and how they guide, protect and empower organizations of any size along their digital journey, visit MNPdigital.ca.