The power of meaningful information: Case study of a data analysis AI assistant

Product & Design Apr 14, 2021

The dimensions of data continue to grow. So is the possibility of what more can be done with the data that comes into your organisation's environment every second, every day. With information being the fuel that drives an enterprise, many insights go unnoticed in the noise, and we keep going in loops of storing, searching, analysing data.

This case brief represents a product that brings real-time work actions and contextual interpretation into the plain world of business intelligence and analytics with a conversational chat agent program. This platform presents you real-time information on what's going on with your business, course of actions such as what is working, not working, what can be improved and so on. The crux of this product is to help businesses make informed decisions with the help of a conversational chat agent.

Challenges faced by enterprises today.

Information gets lost in the noise.

Data volumes continue to grow; storing all the vast data sets is a severe challenge faced by many businesses. This data sits in data centres, databases. But they evolve with time which makes it challenging to handle. Many times the data is unstructured and comes from a multitude of sources such as documents, audio, video, text files, to name a few. Meaning, many times, these data cannot be found in databases and, then the information gets lost in the chaos.

Data Silos hold back your enterprise from working as an integrated unit.

As data rises from a multitude of sources, structured, unstructured, it is often a tedious job to transform the data into relevant, consumable information and present it to all the levels of management in an enterprise.

A lot of enterprises store the information in isolated, unrelated units that have nothing to do together. Hence, it isn't easy to find insights since they are not integrated. As a result, the data analysis team has to crunch numbers to put together a monthly, quarterly, yearly sales report. And like a ripple effect, it slows down the entire decision-making process across various levels of management. It is also why multiple teams in enterprises work as separate units rather than working as an integrated one.

The pictures below show two aspects of data visualisation. The first one is a sales data dashboard that is made painstakingly by the relevant team. The second one depicts a conversation between a bot and a person internal to the organisation looking for sales figures.

A view of an enterprise's sales data dashboard

Understanding the problem statement while building this product.

The major obstacle was to enable the ability to plug into multiple data sources (application with diverse functions and database structures), transform the data into relevant, consumable sets and present the same on a conversational chat agent.

The work plan for the product was drafted into a discovery phase and build phase. The intention behind the discovery period was to understand the extent of conversations, data retrieval and process flow maps.

This system's primary leading points were around the following:

  1. Effective data transformation.
  2. Appropriate natural language models to efficiently address the use case.
  3. Data security and partitioning considering this system would sit on multiple client-side data sources.
  4. Efficient call to action integration to manage the specific business process.
  5. A system UX and UI design that works like an enterprise insights and intelligence platform but is fronted by a simple chat experience.

Today, there are very few designs and implementation cases that effectively combine the chatbot experience with deeper business process management and data exploration techniques. The scope of opportunity for the above case may rise up to more domains than the initial sales intelligence case.

Designing for conversational agents: UX and UI best practices:

Designing conversational agents across a diverse set of implementation formats - chat, voice or gestures-driven, doesn't follow a one-size-fits-all best paradigm. However, some of the key points that were followed here while designing a conversational experience and interface are as follows:

  1. Type and nature of the conversation - Is it objective or subjective? For instance, does the endpoint involve a transaction or a query closure etc., or is it an exploratory conversation?
  2. Characterisation without excess of humanisation.
  3. Consistency of language.
  4. Conversation or re-directs/CTAs.

Conversations should ideally engage/convert in a sequence of minimal chat exchange. However, it depends on the use case but KISS- Keep it short and simple worked here.

Theoretical solution input:

Best practices estimation for the current problem statement.

The end system revolves around the following generalisations:

  1. Relevant information retrieval and presentation
  2. Walkthrough to actions (like meeting requests, enquiries etc.)
  3. Understanding from natural language what to show or where to take to
  4. Information Retrieval from multiple data sources
  5. Data transformation
  6. Presentation (UI)

With these insights, the apt design mix in this case, works around the following:

  1. Consistent language exchange.
  2. Option to opt-out of the conversation and act on action points, however, with the persistence of the derived context.
  3. A healthy mix of text and GUI elements.
  4. Chatbot layout with a more extensive application holding interface frame.  

Introducing a data analysis AI assistant.

A BPM, data analytics crossover with a conversational chat agent program

What it does:

a. Make sense of unstructured data dumps.

b. Derive insights in a natural language format with intelligent interpretation.

c. Help the right people access contextual insights at the right time and take action.

d. Connects to any data source.

e. Creates dynamic data lakes & pipelines.

f. Lets your customers/stakeholders ask in natural language and help them with qualitative insights.

g. Aids decisions - Powerful decision support system inside.

P.S.- This case represents a sales function in an enterprise, but a similar solution could be implemented across multiple functionalities based on a businesses' needs.

If you want to understand how a solution like this can transform your business, let's talk at

Manisha Dash

On a mission to help build a meaningful world for the people.