I am an Analytics and Data Engineer. Will an LLM replace me?
Large Language Models (LLMs) have taken the world by storm. Since early 2023, anyone with a connection to technology has heard of Generative AI (GenAI) and its potential to revolutionize industries—and, perhaps more alarmingly, take over our jobs.
As someone working in tech, I have naturally been affected by the growing discourse on AI’s disruptive potential. I am an analytics and data engineer, responsible for building cleaned datasets, dashboards, and reports that help businesses make informed decisions. Will LLMs replace me? Based on my experience, I am not convinced.
In this article, I use GenAI and LLM interchangeably. An LLM (large language model) is a type of Generative AI that uses human language as input and output medium.
What makes LLMs so powerful?
LLMs are the product of decades of research in linguistics and natural language processing (NLP), bolstered by massive datasets, relatively cheap computing power, and user-friendly interfaces. They provide surprisingly accurate and versatile responses to questions ranging from vacation planning to writing complex SQL queries. However, their impact on specific roles, like mine, depends on how they fit into the nuances of real-world workflows.
GenAI and Analytics: A Real-Life Example
In 2024, I led an internal project at my company to explore how GenAI could enhance our consulting work. Credit to Nicolas Gonzalez, who interned at Argusa during fall 2024. He did most of the work on conducting and interpreting an internal survey on GenAI and catalogued various AI assistants with capabilities that could map to our day-to-day activities. I will share the insights from this work at the end of the article. First, let me share a client project that reflects the typical trajectory and components of analytics work.
In July 2024, I delivered a dashboard project to a client. By August, the client reported it as "broken" because sales numbers had dropped to nearly zero. After debugging the issue, I discovered that the client had switched e-commerce platforms, which altered the formatting of their CSV files. Adjusting the pipeline fixed the sales data issue.
However, a new problem emerged: the profit numbers weren’t aligning with expectations. After investigating and consulting multiple business stakeholders, we identified the root cause—a product category whose costs had risen due to changes in supplier agreements and additional fees. This discovery required weeks of collaborative effort, including exploratory analysis, workshops, and detailed comparisons of 2023 and 2024 product performance.
Here is the breakdown of what this “dashboard problem” entailed:
- Timeline: 14 weeks
- Effort: Hundreds of hours of meetings, 22 Jira tickets, and 120+ emails with over a dozen stakeholders.
- Nature of Work: Highly collaborative, requiring extensive domain knowledge and problem-solving.
Key Takeaways: Why Analytics Is More Than Data
This example highlights why analytics is as much about people as it is about technology. Fixing the pipeline was a technical task, but uncovering the profit issue required:
- Collaboration with stakeholders to gather insights scattered across departments.
- Navigating diverse data sources and business logic.
- Presenting actionable insights through dashboards and reports.
These tasks are inherently human and context-dependent. Could an AI assistant have connected the dots between supplier fees and profit margins? Not likely—because that insight required deep collaboration and intuition. LLMs, while impressive, are far from replacing this level of nuanced, people-driven work.
Insights from a GenAI Survey
To better understand the potential role of GenAI in our workflow, at Argusa, we conducted a survey with 11 consultants across 40 client projects. Here’s what we found:
1. Data manipulation takes most of our time: While the final product we deliver is usually dashboards or reports, less than 25% of our time is spent building dashboards. The majority is dedicated to data engineering, cleaning, and debugging the processing pipelines.
2. Tedious Tasks = Automation Opportunities: We asked the survey participants to rate the different tasks on an axis of tediousness and on the possibility of Gen AI aiding on those tasks.
The activities perceived as repetitive and tedious are also the ones with strong potential for GenAI assistance. This is encouraging but also unsurprising.
While documentation is crucial, in most cases developers do not have the appropriate amount of time to dedicate to it. GenAI assistants that can write the documentation based on the source code, and keep it up-to-date, would add huge value to the development team’s work. Activities like creation and maintenance of data pipelines tend to have a large code component to them and can therefore be enhanced by co-pilots. And finally, even project management is perceived as a potential candidate to benefit from Gen AI because a lot of this work is document driven, where LLMs can help by summarizing or retrieving information in large bodies of text. Another aspect of project management is crafting pertinent communication, an area where LLMs excel.
Tasks like creation of visualisations and reporting do not seem likely to benefit from GenAI because of two reasons. One is the understanding of the business needs, that we acquire via diverse and loosely integrated sources e.g. meetings, emails, screenshots and other examples etc. The other reason is the interface: LLMs can write code, but so far are not able to masterfully manipulate a graphic user interface to juxtapose the relevant objects to create visualisations.
And lastly, higher level tasks like the design and architecture of databases or platforms, are quite challenging and far from tedious. As expected, the human understanding of context, constraints, and organization culture plays a huge role in such projects, and therefore this area is perceived to have a minimal benefit from GenAI technology.
3. Challenges in Integration: When asked the most important factors hindering the integration of GenAI into the workflow of analytics, our consultants cited diversity and variability of software and data flows, as well as complexity of tasks as reasons.
In retrospect, this is also quite an intuitive result. As consultants, we are exposed to many environments where similar tools and platforms are being used in different configurations depending on the client’s industry and even culture. Even within the same company/client, we encounter many ways to extract and process data and the meaning that the data is given via reports and dashboards. Without repeatable, clear and well-documented processes, it is difficult to automate the analytics workflow or involve GenAI to augment human capability.
Where Does GenAI Fit?
There are three levels at which GenAI could enhance analytics workflows:
- Personal Use: Leveraging LLMs for documentation or generating SQL queries. However, concerns about data security and lack of integration into existing workflows limit this option.
- AI-Assisted Tools: Built-in assistants in text editors, email clients, or meeting platforms that streamline tasks like summarising meeting notes. This is a secure option with data security offered via the tool vendors themselves in most cases. For data analytics workflow, this is an immediate efficiency boost be it in coding, documentation or project management tasks.
- Integrated Co-Pilots: GenAI embedded within analytics tools, such as Tableau or Power BI. While this is the most secure option, it depends on vendors implementing these features effectively. Nevertheless, tools are evolving very fast and it is only a matter of time before graphic interface based tools implement assistants with capabilities on par with coding co-pilots.
The Future of Analytics and GenAI
As analytics professionals, we must adapt to a rapidly evolving landscape. However, here is why I believe LLMs won’t take over our roles entirely:
- Analytics Is a People Business: Much of the knowledge required for analytics resides in people’s minds, not in databases. Collaboration and communication remain critical.
- Data Engineering Matters: Understanding how data is generated, modeled, and transformed is a fundamental skill that is far from being fully automatable.
The fear of GenAI replacing jobs is valid, but the reality is often more nuanced. I believe that GenAI has the potential of greatly enhancing our capabilities as analytics professionals. As humans, our ability to navigate complexity, connect disparate pieces of information, and align insights with business goals—remains irreplaceable.
Author:
Fatima Soomro, in collaboration with Nicolas Gonzalez
Large Language Models (LLMs) have taken the world by storm. Since early 2023, anyone with a connection to technology has heard of Generative AI (GenAI) and its potential to revolutionize industries—and, perhaps more alarmingly, take over our jobs.
As someone working in tech, I have naturally been affected by the growing discourse on AI’s disruptive potential. I am an analytics and data engineer, responsible for building cleaned datasets, dashboards, and reports that help businesses make informed decisions. Will LLMs replace me? Based on my experience, I am not convinced.
In this article, I use GenAI and LLM interchangeably. An LLM (large language model) is a type of Generative AI that uses human language as input and output medium.
What makes LLMs so powerful?
LLMs are the product of decades of research in linguistics and natural language processing (NLP), bolstered by massive datasets, relatively cheap computing power, and user-friendly interfaces. They provide surprisingly accurate and versatile responses to questions ranging from vacation planning to writing complex SQL queries. However, their impact on specific roles, like mine, depends on how they fit into the nuances of real-world workflows.
GenAI and Analytics: A Real-Life Example
In 2024, I led an internal project at my company to explore how GenAI could enhance our consulting work. Credit to Nicolas Gonzalez, who interned at Argusa during fall 2024. He did most of the work on conducting and interpreting an internal survey on GenAI and catalogued various AI assistants with capabilities that could map to our day-to-day activities. I will share the insights from this work at the end of the article. First, let me share a client project that reflects the typical trajectory and components of analytics work.
In July 2024, I delivered a dashboard project to a client. By August, the client reported it as "broken" because sales numbers had dropped to nearly zero. After debugging the issue, I discovered that the client had switched e-commerce platforms, which altered the formatting of their CSV files. Adjusting the pipeline fixed the sales data issue.
However, a new problem emerged: the profit numbers weren’t aligning with expectations. After investigating and consulting multiple business stakeholders, we identified the root cause—a product category whose costs had risen due to changes in supplier agreements and additional fees. This discovery required weeks of collaborative effort, including exploratory analysis, workshops, and detailed comparisons of 2023 and 2024 product performance.
Here is the breakdown of what this “dashboard problem” entailed:
- Timeline: 14 weeks
- Effort: Hundreds of hours of meetings, 22 Jira tickets, and 120+ emails with over a dozen stakeholders.
- Nature of Work: Highly collaborative, requiring extensive domain knowledge and problem-solving.
Key Takeaways: Why Analytics Is More Than Data
This example highlights why analytics is as much about people as it is about technology. Fixing the pipeline was a technical task, but uncovering the profit issue required:
- Collaboration with stakeholders to gather insights scattered across departments.
- Navigating diverse data sources and business logic.
- Presenting actionable insights through dashboards and reports.
These tasks are inherently human and context-dependent. Could an AI assistant have connected the dots between supplier fees and profit margins? Not likely—because that insight required deep collaboration and intuition. LLMs, while impressive, are far from replacing this level of nuanced, people-driven work.
Insights from a GenAI Survey
To better understand the potential role of GenAI in our workflow, at Argusa, we conducted a survey with 11 consultants across 40 client projects. Here’s what we found:
1. Data manipulation takes most of our time: While the final product we deliver is usually dashboards or reports, less than 25% of our time is spent building dashboards. The majority is dedicated to data engineering, cleaning, and debugging the processing pipelines.
2. Tedious Tasks = Automation Opportunities: We asked the survey participants to rate the different tasks on an axis of tediousness and on the possibility of Gen AI aiding on those tasks.
The activities perceived as repetitive and tedious are also the ones with strong potential for GenAI assistance. This is encouraging but also unsurprising.
While documentation is crucial, in most cases developers do not have the appropriate amount of time to dedicate to it. GenAI assistants that can write the documentation based on the source code, and keep it up-to-date, would add huge value to the development team’s work. Activities like creation and maintenance of data pipelines tend to have a large code component to them and can therefore be enhanced by co-pilots. And finally, even project management is perceived as a potential candidate to benefit from Gen AI because a lot of this work is document driven, where LLMs can help by summarizing or retrieving information in large bodies of text. Another aspect of project management is crafting pertinent communication, an area where LLMs excel.
Tasks like creation of visualisations and reporting do not seem likely to benefit from GenAI because of two reasons. One is the understanding of the business needs, that we acquire via diverse and loosely integrated sources e.g. meetings, emails, screenshots and other examples etc. The other reason is the interface: LLMs can write code, but so far are not able to masterfully manipulate a graphic user interface to juxtapose the relevant objects to create visualisations.
And lastly, higher level tasks like the design and architecture of databases or platforms, are quite challenging and far from tedious. As expected, the human understanding of context, constraints, and organization culture plays a huge role in such projects, and therefore this area is perceived to have a minimal benefit from GenAI technology.
3. Challenges in Integration: When asked the most important factors hindering the integration of GenAI into the workflow of analytics, our consultants cited diversity and variability of software and data flows, as well as complexity of tasks as reasons.
In retrospect, this is also quite an intuitive result. As consultants, we are exposed to many environments where similar tools and platforms are being used in different configurations depending on the client’s industry and even culture. Even within the same company/client, we encounter many ways to extract and process data and the meaning that the data is given via reports and dashboards. Without repeatable, clear and well-documented processes, it is difficult to automate the analytics workflow or involve GenAI to augment human capability.
Where Does GenAI Fit?
There are three levels at which GenAI could enhance analytics workflows:
- Personal Use: Leveraging LLMs for documentation or generating SQL queries. However, concerns about data security and lack of integration into existing workflows limit this option.
- AI-Assisted Tools: Built-in assistants in text editors, email clients, or meeting platforms that streamline tasks like summarising meeting notes. This is a secure option with data security offered via the tool vendors themselves in most cases. For data analytics workflow, this is an immediate efficiency boost be it in coding, documentation or project management tasks.
- Integrated Co-Pilots: GenAI embedded within analytics tools, such as Tableau or Power BI. While this is the most secure option, it depends on vendors implementing these features effectively. Nevertheless, tools are evolving very fast and it is only a matter of time before graphic interface based tools implement assistants with capabilities on par with coding co-pilots.
The Future of Analytics and GenAI
As analytics professionals, we must adapt to a rapidly evolving landscape. However, here is why I believe LLMs won’t take over our roles entirely:
- Analytics Is a People Business: Much of the knowledge required for analytics resides in people’s minds, not in databases. Collaboration and communication remain critical.
- Data Engineering Matters: Understanding how data is generated, modeled, and transformed is a fundamental skill that is far from being fully automatable.
The fear of GenAI replacing jobs is valid, but the reality is often more nuanced. I believe that GenAI has the potential of greatly enhancing our capabilities as analytics professionals. As humans, our ability to navigate complexity, connect disparate pieces of information, and align insights with business goals—remains irreplaceable.
Author:
Fatima Soomro, in collaboration with Nicolas Gonzalez