AI Powered Knowledge Management and Insight Generation

As companies grow, it becomes a continuous challenge to keep track of everything relevant that is happening that may have an impact on their operations. Both internal and external information and data are routinely “lost in the mix”, and relevant details only come up when it’s already to late and companies are forced to take reactive measures, instead of being proactive.

In fact, research suggests that over 50% of data goes unused, which can become especially inefficient when companies reach a global size, and the data needs of each individual regional cluster become more specific.

Fortunately, the new access to an array of Gen AI technologies allows us to be more creative and inventive in the approach to topics of knowledge management and knowledge access. Our challenge was, then, to completely transform the access and engagement that employees have with information, both documents and databases.

Typically, datasets are managed by some type of internal data team that curates and maintains the databases. Since this team will be responsible for data quality and accuracy, and given that database interactions usually require some degree of technical skill, it’s not uncommon for data requests to also be directed to this team. This can lead to two suboptimal outcomes:

• Immense strain on the team to keep up with all data and analytics requests
• Delay in the time it takes between a business team having a data-related question and the provisioning of the data or analytics that will help provide an answer

On the document side, the capacity for knowledge management is usually even lower. Up until the Gen AI tools were presented, there were no tools available that could properly be used to extract, structure and leverage free-text data in an efficient manner. At least not at the speed that is now possible. Thus, organizations focused their efforts on specific, business-critical documents and employed resources aimed specifically at managing this information.

With this in mind, we partnered with a Consumer Goods global company and our goal was to approach this challenge by carefully designing the use cases around the capabilities of Gen AI and to maximize the value that it can provide to end users, even if they don’t have technical or analytical know-how.

We started with creating 1:1 mockups of each use case, complete with the look and feel of the real app. This allowed us to quickly iterate with the client the key features, graphic profile and overall layout of the app and each individual page. Throughout this exercise, we’ve also discussed and agreed on the high-level methodology for each AI feature, making sure that what was planned on the mockup was possible to develop.

This first approach was essential to set the tone for a truly collaborative ideation and development, making sure that the client was completely engaged with the vision of the app.

The key use cases developed were the following:

• Search engine – a simple, direct use case for document search, which allows users to search using any prompt. Documents were also enriched in the ingestion process with custom summaries and insights
• Chat engine – the app’s powerhouse, hidden behind the interface’s simplicity. It is a chat interface that allows users to get answers on internal documents, generate descriptive analytics on the go using data from their internal databases or just use as a standard GPT with web search. Its powerhouse status comes from its intricate pipelines that enable the generation of relevant answers even when the user’s questions are layered and require sequential reasoning. It also required the development of an agentic architecture with custom tools that produce visualizations based on internal KPIs. It’s agentic nature is what enables an adaptive generation of descriptive analytics outputs and visuals. It is also used across other use cases that require content generation.
• Report engine – Leveraging the Chat engine, this use case is centered around creating more complex stories, merging data analytics with document retrieval and content generation. It allows users to create custom reports that focus on one or several topics based only on simple user parametrization and prompting. It is also built around a workspace concept, promoting inter-user sharing and knowledge transfer.

These were complemented by other, smaller and simpler pages focused on specific and typical Gen AI interactions, like document summarization. The overarching goal was to create a productivity app, powered by internal knowledge and by customized Gen AI pipelines that integrate business needs and ease up the access to user-centric analytics.

Upon launch, the app became a central tool for day-to-day activities that require any interaction with documentation or internal data. On one side, strategic roles like directors and managers will engage with it to retrieve insights that can aid strategic, inter-regional

decisions, as well to conduct analysis to generate new hypothesis to follow through. Alternatively, more operations focused roles will depend on it to reduce the time-to-market of recurrent analyses on brand and product performance on specific geographies and specific segments, enabling faster reactive measures to shifting market conditions.

Given that the tool is the same for everyone, it also serves the purpose of standardizing the approach and outputs across the organization. This, alone, is incredibly beneficial since it reduces the likelihood of running similar analysis on different data sources or with different assumptions, reducing the time spent on identifying sources of these differences.

With this centralized and customized tool, the retrieval of insights and analysis can be more structured, more agile and employee time can be spent on more value-added tasks, instead of repetitive processes. Additionally, the app can be continuously improved with changes to existing use cases, creation of use cases or expanding the capabilities of the custom pipelines. This means that the organization can respond to developments in Gen AI capabilities with greater flexibility and adapt to its own needs.

‍