Generative AI and Knowledge Management

Knowledge Management (KM) is a term and discipline that emerged in the 1990s. Its essence is to capture new information as it happens while enlisting internal experts to share their wisdom and expertise. Join us as we explore how AI and knowledge management can be paired together to increase team efficiency. 

Over the years, significant strides have been made in this field, with billions of dollars being spent to solve this problem. We have seen a proliferation of KM tools like Yammer, Slack, Teams, Confluence, Notion, SharePoint, etc. This list goes on and on. Many of the original concepts that knowledge management represented have been realized at many levels. It is easier to create, share, and protect information internally and externally than ever before. Search tools have made things easier, but processes and data quality challenges have produced reams of information without proper context.

At UDig, we recognize these challenges acutely. Our consultants frequently need to switch contexts and grasp new information quickly, which is essential in our field. However, sharing and finding information remains difficult, leading to wasted hours and inefficiencies. Let’s dive into our approach to AI in knowledge management.  

 In this article we will cover:

• Current Challenges with Knowledge Management
• Our Approach
• Challenges in Implementing Gen AI for Knowledge Management
• Technical Approach: Building on Our Microsoft 365 Foundation
• SalesGenie: Our Chatbot MVP
• Impact

Current Challenges with Knowledge Management

Sharing and finding information at UDig can be difficult, leading to wasted hours. Our business thrives on information, and our consultants frequently need to switch contexts and grasp new information quickly. Here are three of the main challenges we’ve come across in KM:

1. Multiple Secure Environments: Our work requires us to operate in various secure environments, complicating communication and information sharing.
2. Tool Diversity: While we have core tools like Salesforce, Kantata, Slack, and Office365, our team’s preference for diverse tools creates challenges in consolidating and sharing work effectively.
3. Balancing Flexibility & Consistency: We strive to keep our organization lean and efficient, but this must be balanced with necessary process and risk management.

Our Approach

We strive to keep our organization lean and make it easy to get things done at UDig, hence our exploration into generative ai and knowledge management. Our proposed solution internally involves the development of context-specific chatbots trained on our data. This will necessitate a comprehensive data consolidation effort and metadata validation. We are currently building and testing chatbots to gain a better understanding of the quality and accuracy of our data. This initiative underscores the process challenges we face in storing and retrieving data and the importance of addressing these challenges.  

Sales & Marketing Bot

A bot dedicated to helping our Growth and Client Services teams find sales and marketing content quickly and efficiently. Example prompts:  

• “I am proposing an automation workflow for a healthcare client.  Summarize all proposals for the healthcare vertical that refer to payer automation.”
• “We are in the final negotiations with a client.  They want to check references. Build a list of client references for Financial Services.”
• “The quarter is closing in 15 days. What is our progress against the current bookings goal?”
• “Show all RFP responses for modernizing applications for the State Government vertical.”

People Bot

A bot dedicated to answering questions about our internal HR policies and process workflows. Example prompts: 

• “My great uncle is sick. What is our bereavement policy?”
• “I want to continue to grow my design skills. What’s the process for getting training approved?”
• “I want to publicly recognize one of my co-workers for helping me with a critical problem with a client. How would I do that?”
• “Can you show me all training documents related to negotiating competing priorities with a client?”

Delivery Bot

A bot dedicated to our delivery processes and artifacts and teach it to audit our portfolio and projects. Example prompts: 

• “Review our retrospectives and share themes I should be aware of when starting a new project.”
• “Can you provide sample work products for effectively documenting business processes?”
• “Share our project closeout process and who should participate.”
• “What is our average team velocity for the software engineering over the past 12 months?” 

Challenges in Implementing Gen AI for Knowledge Management

Data Quality & IntegrationUser AdoptionData Governance & ComplianceOperational IntegrationROI Measurement

Data Quality & Integration

Generative AI’s effectiveness depends on data.  Our data is stored in multiple cloud providers. Integrating that data and making it available in near real-time is a critical requirement. Re-indexing and data movement can be an incremental expense.  

User Adoption

Even the most advanced AI systems are only effective if employees use them. Driving user adoption often requires significant change management efforts, including training, clear benefits communication, and integrating AI into existing workflows. 

Data Governance & Compliance

As AI systems require access to large datasets, ensuring compliance with data governance standards and regulations becomes crucial. This includes managing data access, ensuring security and confidentiality, and adhering to industry-specific regulations, which can be complex and resource-intensive. 

As our knowledge base grows and our thinking evolves, our historical data will be important but also outdated in certain aspects. For example, our brand evolves, and our messaging can change.  How do we ensure our current brand elements are utilized in future communication?  

Operational Integration

Integrating Generative AI into existing knowledge management systems and processes can be technically challenging. Our commitment to being lean and nimble must be factored into every system we deploy at UDig. In addition to curating and disseminating information, AI tools must be capable of initiating workflows.  

ROI Measurement

By saving time for all our employees, we can support our mission of accelerating impact. We don’t have a baseline or data to support our claims that we waste time looking for information, but our perspective is universally accepted by the team.

Technical Approach: Building on Our Microsoft 365 Foundation

Leveraging Existing Infrastructure

As a Microsoft 365 organization, most of our knowledge resides within SharePoint and OneDrive. This ecosystem forms the backbone of our daily operations, from document collaboration to project management. Our reliance on these platforms has created a rich, albeit sometimes fragmented, repository of information that holds immense potential for enhanced knowledge management. Recognizing the synergy between our existing infrastructure and Microsoft’s cutting-edge AI services, we’re capitalizing on this familiar ecosystem to build our GenAI-powered knowledge management system, leveraging Azure’s powerful suite of AI tools.

Data Consolidation & Integration: The Foundation of AI-Powered Knowledge Management

In AI-driven knowledge management, data consolidation and integration form the bedrock upon which all other capabilities are built. This process involves bringing together disparate data sources, harmonizing their formats, and creating a unified, accessible repository of information. For AI systems, particularly those powering knowledge management, the quality and organization of data directly impact their effectiveness. Well-consolidated and integrated data allows AI models to draw more accurate insights, make better connections between pieces of information, and provide more relevant responses to user queries. It’s not just about quantity; it’s about creating a coherent, contextual data ecosystem that AI can navigate and understand. With proper data consolidation and integration, even the most sophisticated AI models could deliver meaningful results, akin to trying to solve a puzzle with pieces from different sets.

Current State & Path Forward

At UDig, we’re starting our GenAI-powered knowledge management journey from a familiar place. Our data resides in Microsoft 365 and SharePoint, organized using basic folder structures and manual file management. While this system has served us well, it comes with limitations. Our metadata tagging is a manual process, often inconsistent or incomplete. Data silos exist across departments and tools, making it challenging to get a holistic view of our information landscape. Search functionality, while available, is confined to individual platforms, requiring our team to navigate multiple systems to find what they need. We recognize these constraints but are also mindful not to overengineer our solution prematurely. By acknowledging our current state, we can build a flexible foundation that allows for gradual improvements without locking ourselves into rigid systems or processes. This approach ensures we can adapt to emerging technologies and evolving business needs as we progress in our knowledge management transformation.

Cognitive Search & Advanced Indexing: Powering Intelligent Information Retrieval

At the heart of our GenAI-powered knowledge management system lies a sophisticated cognitive search and indexing engine built on the robust foundation of Azure AI Search. This platform serves as our cornerstone for transforming raw data into actionable insights. We’re embarking on a journey with fundamental content indexing and keyword-based search capabilities, laying the groundwork for more advanced functionalities. 

As we progress, we’re leveraging Azure AI Search’s cognitive skills to enhance our metadata extraction processes, enabling a deeper understanding of our content. This allows us to implement semantic search capabilities, moving beyond simple keyword matching to grasp the intent and context behind user queries. We’re also developing custom skill sets tailored to UDig’s unique content, ensuring our search engine speaks our language and understands our business nuances. 

Our vision extends to the cutting edge of search technology with the implementation of vector search. This approach allows us to represent documents and queries as mathematical vectors in a high-dimensional space, enabling highly nuanced similarity-based searches. By combining traditional keyword search with semantic understanding and vector representations, we’re creating a hybrid search system that can handle complex, context-dependent queries with remarkable accuracy. 

The goal is a fully cognitive search capability that leverages advanced AI models for real-time indexing and content understanding. This system will find relevant information, understand relationships between different pieces of content, identify trends, and even predict information needs before they arise. Through this evolving approach to cognitive search and indexing, we’re not just finding needles in a haystack—we’re weaving those needles into a tapestry of knowledge that drives UDig’s success.

SalesGenie: Our Chatbot MVP

As we embarked on our journey to harness the power of Generative AI for knowledge management, we decided to start with a focused Minimum Viable Product (MVP) chatbot named SalesGenie. This MVP is a proving ground for our concepts and a steppingstone towards a more comprehensive solution. 

Here’s how we approached the development of SalesGenie, considering various crucial factors:

Technical RequirementsConstraints & LimitationsSecurity of InformationLearning & Iteration

Technical Requirements

Deployment Strategy

• A cloud-based approach using Azure services and a web-based application accessible through our internal portal. This approach allows for easy updates and ensures that all users always have access to the latest version of the chatbot.

Scalability

• Azure’s cloud infrastructure allows us to scale our resources as usage proliferates.

Integration

• As a Microsoft 365 organization, Azure seamlessly integrates our existing SharePoint and OneDrive data sources. 

Managed Services

• As a Microsoft 365 organization, Azure seamlessly integrates our existing SharePoint and OneDrive data sources. 

Monitoring & Analytics

• Azure’s built-in monitoring tools allow us to track usage, performance, and errors in real-time.

Constraints & Limitations

In developing SalesGenie, we carefully navigated several constraints. First, we limited the chatbot’s scope to sales-related information to ensure a manageable dataset for our initial training. This focused approach allowed us to fine-tune the AI’s responses more effectively. We also set clear boundaries on the types of queries SalesGenie could handle, primarily focusing on factual information retrieval rather than complex decision-making processes. Additionally, we implemented rate limiting to manage server load and costs during this experimental phase.

Security of Information

Given the sensitive nature of sales data, security was paramount in our MVP design. We leveraged Microsoft Entra ID (formerly Azure Active Directory) for authentication and authorization. This integration ensures that only authenticated UDig employees can access SalesGenie, with role-based access control determining the level of information each user can retrieve. We also implemented encryption for data in transit and at rest, ensuring that sensitive sales information remains protected throughout the interaction.

Learning & Iteration 

We’ve focused on gathering user feedback and analyzing interaction logs throughout the MVP process. This data is crucial for identifying areas of improvement, uncovering new use cases, and refining the AI’s responses. We’ve scheduled a regular review cycle to incorporate these learnings into SalesGenie’s development roadmap. Starting with SalesGenie as our MVP, we’ve gained valuable insights into the practical applications and challenges of implementing a GenAI-powered chatbot within our organization. This experience informs our strategy for expanding the chatbot’s capabilities and developing similar tools for other departments. As we refine and broaden SalesGenie, we’re laying the groundwork for a more comprehensive, company-wide knowledge management solution that leverages the full potential of GenAI technology. 

Impact of AI Knowledge Management

If we can save two hours a week per employee, that savings will improve productivity, reduce response time to client needs, and allow more creative problem-solving time.  As UDig continues to scale, the benefit of context-specific GenAI-based chatbots will continue to grow.

Ready to unleash the power of AI? Let’s Talk.