Harnessing AI for Data Analysis: Expert Insights from Shahtaj as Data Analyst

UL: Please tell us a little bit about yourself and the company that you work for 

My name is Shahtaj. I'm working with HBL as a data analyst in the financial inclusion and innovation department. I started my career in advertising and marketing. I was working as a digital marketing executive. As we were getting a lot of data from SEO, Google Ads, Facebook Ads, etc., I also analyzed that for the clients. Later, I decided to do a few short courses related to data analysis and interned at Ipsos MORI as a researcher for the consumer insights department. I also completed my MBA and focused my research on the financial inclusion and innovation of women in Pakistan. After completing my traineeship, I joined HLB. I also volunteer with AKGN –the Aga Khan Developer Network – for the Aga Khan Education Board as a data analyst, where I look after the data of school children in remote areas of Pakistan. 

UL: Can you share specific examples of how AI has transformed data analysis in your organization or industry and what measurable impacts you have observed?  

Regarding data analysis in banking, we deal with considerable data. The variety and velocity are huge, and we generate vast amounts of new data. AI has definitely transformed data analysis in terms of streamlining the data pipelines. For example, when you see a lot of data coming before doing analysis or building any consensus with it, you have to extract the relevant data. Second is the presentation of the data. In the past, we had done a lot of manual decks. Now, we can automate this process with AI. I would say this had transformed my life. Third is the analysis itself. Let me give you an example: We have to submit daily reports, and we used AI to help us write Python scripts, which immensely helped us assimilate the data. Finally, AI also allows us to benchmark. There are two ways to get a benchmark – your own historical and industry data. With AI algorithms and several data points, we can create a prediction and benchmark against it. These predictive models help us decide whether to make short-term adjustments, like running a new marketing campaign to boost sales.  

UL: How can businesses leverage AI to become more data-driven, and what kind of benefits can they expect from this transformation? 

When it comes to banking, there's a lot of data. However, for any data to work out, it must be of good quality. The first step is to minimize data quality issues. There are many frameworks regarding data governance, and it is important to implement them correctly. The second step is regular audits of all data sources to ensure the quality doesn't slip. After those steps, the business can implement machine learning models. I have seen many times that companies focus on employing staff who are very good with software, like PowerBI champions, but this is secondary. First is ensuring data quality. The second pitfall I have seen is that companies don't make good decisions on where to implement AI. They know that they need AI data analysis tools and invest a lot of money into them, but they don't make the right decisions about using them. In other words, it is the prioritization of AI. The third risk to be wary of is the lack of model maintenance. Models can be overfit or underfit for the analysis we want to perform. Especially when data comes with great velocity, we must know that we can rely on the models to be up-to-date and well-maintained.  

Chatbots are vital if we are speaking about them, but companies shouldn't rely only on them. Chatbots often fail to realize the context of complaints. For a while, we saw a boom in chatbot use among FinTech startups, as they wanted to minimize the human effort in the call center or complaint departments. But down the line, they realized that chatbots could not cope with the contextual matter of the inquiries and would guide customers in the wrong direction.  

Finally, I feel there's a great gap in human resources. I don't mean just a gap in awareness but also in implementation. Training staff on new relevant skills to give them new capacity to use these tools is crucial for success.  

UL: What role does AI play when we're talking about setting, tracking, and achieving KPIs? 

This is a very interesting question. For me, AI is very helpful in debugging my code. It helps to streamline the ETL process and makes me more productive overall. Another perspective is to look at business-level KPIs. For example, you have a banking system app, and you want to predict system errors. A model can take past data to indicate the number of transactions and the error rate for a set period. Now, divide the system errors by the total, and it will give me the overall error rate. This is my benchmark. Let's say my model gives me an accuracy of 85%, and it's not an overly underfit or overfitted model. Now, if the model predicted an error rate of 4% but my actual error rate is 3%, then everything is green. But if the actual error rate is 7%, I might have to realign the KPI.  

AI is now part of the broader organizational strategies as well. When we talk about process transformation or customer service – AI is there,  like chatbots or automated tracking. AI implementation itself has become a broader KPI to achieve.  

 UL: How can businesses use AI to gain deeper insights about their customers and then use these insights to improve customer satisfaction? 

There are a lot of ways. I can share examples from FinTech and banking. One way is a sentiment analysis. It basically means that we analyze all customer communication – the complaints, emails, messages, etc – and find out what the sentiment is. We can also do a more qualitative thematic analysis. Once we understand how customers feel overall and the major pain points, we can dive deeper with a root cause analysis and improve the process design to make it a better experience for the customer.  

Another analysis we can perform is segmentation or clustering. We place customers in different groups so we can serve them the right product or service offer. For example, we have customers who haven't done any transactions in a while. Their last transaction is bill payment or topping up their mobile balance, so we can send them a specific push notification promoting our mobile payment business. That kind of hyper customization is really aiding us to devise and sell products. 

Thirdly, is service design or simplifying the process. I think AI has really helped us to improve and ease customer journeys at various levels. We could see the data at different transaction steps and measure the performance. We could simplify it, automate push notifications, and improve timing.  

UL: What best practices are there for companies to ensure their data analysis projects are successful? 

I think the best advice is to tackle the right problem. You just cannot have data on the dashboard and say the numbers are great. You must identify the right issue to tackle and the right question to answer through your data. Otherwise, whatever numbers you put out it won't make any sense. Number two is to create an information exchange between departments. There must be a synergy between departments. When sharing data, we also need to know the backstory and context of it. Finally, remember who the target audience of that data is. Keep in mind that your target audience is not a researcher. Make sure there is a sense of relativity, because most of the time, data only makes sense in comparison.   

UL: Is there any kind of trend or new development in AI data analysis that excites you most this year or in the upcoming years? 

There are a couple of things that I'm aware of that keep me really excited. Number one is synthetic data. As a data analyst, I haven't encountered any perfect data set yet. There are always some quality issues, like missing headers, wrong formatting, numbers entered as text or similar issues. The headers are different. The numbers are converted to text, you know, these kinds of issues. So, instead, you have synthetic data that reproduces the original characteristics and structure but doesn't have these issues. It makes data analysis so much easier.  

The second thing that has been really exciting is Large Language Models – LLM - obviously, we all know about those from ChatGPT and Gemini. But there has also been a surge in Short Language Models – SLMs – and the development this sector has been seeing is tremendous.  

The number three trend, which is really exciting, is computer vision. It is a very old concept, and it's been on the market for a while, but the pace at which computer vision has been developing recently is amazing. The applications in digital forensics or in customer data analysis are really exciting. For example, I was reading yesterday that it is possible to analyze a customer's micro-expressions with computer vision.  

However, with these developments also come a lot of compliance and data governance issues. There are a lot of ethical questions about AI. But it has already penetrated every aspect of our lives, and we have to make the best use of it.