Is AI Intelligent Enough to Bolster Business Intelligence?

Business intelligence (BI) – using data about business performance, behavior, the market, etc. to make informed business decisions – is a bedrock of successful digitization in the microfinance sector and elsewhere. Customer data dashboards are a key business intelligence tool for microfinance institutions (MFIs), allowing them to leverage existing data and improve understanding of their customers’ behavior. However, these dashboards require concerted time and effort to analyze the data they provide, leading us to ask, could better AI tools ensure MFIs, regardless of size and resourcing, are able to glean insights from customer data dashboards and other business intelligence tools? 

Offered by cloud companies, Software-as-a-Service providers, and specialized tech entities, AI tools are likely to become commonplace in the very near future. Below, we lay out how business intelligence can be assisted by AI today, and where we believe human input remains irreplaceable. For that purpose, we use the stylized data analytics journey created for the CGAP microfinance digitization project which is explained in a tutorial video (see Figure 1).

Figure 1: Data Analytics Journey

Figure 1: the data analytics journey

Step 1: Gathering Requirements

Areas where AI can support:

An AI model, specifically Natural Language Processing (NLP), can assist with capturing and interpreting requirements. For instance, AI chatbots can engage with stakeholders to ask key questions and document the responses, ensuring consistency and thorough requirements gathering.

Areas where human expertise is preferred:

The process of understanding the business problem and aligning the data analyses with strategic objectives requires human involvement. Interactions between the product owners and data analysts are crucial to understanding the nuances, priorities, and context behind the requirements.

Step 2: Data Availability

Areas where AI can support:

Data analysts can ensure that the required data points are present in the data warehouse and assess the data points that can be substituted when the required data is not available. Machine learning algorithms can analyze the metadata of a data warehouse and suggest potentially relevant datasets for a given problem statement. AI can automate the discovery of correlations and dependencies in datasets to suggest additional data points that may need to be collected based on past projects.

Areas where human expertise is preferred:

The final decision on data relevance and validity often requires a human analyst who understands the business context. For instance, there might be regulatory or ethical reasons to exclude certain data points, which AI might not be capable of considering.

Step 3: Queries

Areas where AI can support:

Data analysts write codes in different programming languages such as Python, R, and SQL for data extraction. AI can help with debugging and building blocks of SQL query (and other programming languages) which can reduce the time spent writing complex queries and can reduce errors while optimizing the performance of the queries.

Areas where human expertise is preferred:

The application of business logic and the translation of business requirements into technical requirements is a task best performed by data analysts. Moreover, interpreting and managing exceptions and outliers, particularly those that are context-specific, often requires human intuition and judgment.

Step 4: Data Visualization 

Areas where AI can support:

Using the output data, analysts transform them into charts and graphs while choosing the suitable visualization to address the problem statement. AI could suggest the most effective way to visualize specific datasets based on their characteristics. For instance, AI could determine whether a line graph or bar graph would be best suited for a particular dataset, improving the speed and quality of visualization selection.

Areas where human expertise is preferred:

Good visualization often needs to be supported by storytelling that is best handled by data analysts. Understanding the story that the data tells and how best to present that story to key stakeholders often requires human creativity. Humans are also better equipped to understand and adjust to the preferences and understandings of the report’s intended audience.

Step 5: Validation

Areas where AI can support:

Data analysts and stakeholders review the visualization together and validate if there are any areas of concern, for example, missing data or data integrity issues. AI models can be used to predict potential issues in the data or in the interpretation of visualizations. By learning from previous validation sessions and stakeholder feedback, AI could flag potential issues before the validation stage (i.e., requirement gathering or data availability stage) saving time and reducing the number of iteration cycles.

Areas where human expertise is preferred:

Validating the data and analytics process requires human stakeholders who can bring their business knowledge and contextual understanding to bear. They can understand the ramifications of the findings and challenge them in ways AI cannot.

Step 6: Deployment

Areas where AI can support:

Data analysts check manually for error messages and failures in report refresh. AI could be used to monitor the deployment of the dashboards, checking for any errors or inconsistencies that occur. AI models could also dynamically adjust the refresh frequency based on the data volatility and business needs, balancing timeliness, and resource usage.

Areas where human expertise is preferred:

AI can monitor and manage errors during deployment, but when errors do occur, humans are typically better at troubleshooting and finding a solution, as they better understand the broader system in which the deployment operates.

Step 7: Publishing

Areas where AI can support:

Data analysts share the report with a wider audience, customizing the format based on the audience as well as the delivery context. AI can assist in personalizing the delivery of reports to different stakeholders based on their preferences and past interactions. The AI language model can be used to write clear, concise summaries and insights from the visualizations.

Areas where human expertise is preferred:

The final polishing and explanation of the data often require a human touch. Humans excel at understanding other humans, and therefore they are often better equipped to communicate complex findings in a manner that will be understood and accepted by the stakeholders.

In conclusion, we think that artificial intelligence is already intelligent enough to improve business intelligence. This is particularly true for tasks on the data analytics journey that require prediction rather than judgment and could be easily automated. However, (as of now) AI tools are unlikely to solve fundamental shortcomings such as outdated, dysfunctional, or indeed non-existent core banking systems. And they are unlikely to relieve MFIs (or any other financial service providers) of the necessity to employ staff with a robust understanding of technology and data analytics.

Please let us know what you think and what your experience has been with either AI or BI either using the comments from below or e-mailing



Microfinance institutions that successfully generate value for their business and customers through digitization anchor these efforts in business intelligence. This Technical Note outlines an approach for improving business intelligence with interventions that require minimum or no investment in technology. CGAP also offers a customer dashboard library with detailed instructions for data teams and a tutorial video.

Add new comment