The state of analytics engineering in 2025: A summary

Every year, dbt Labs releases our State of Analytics Engineering report. In our latest 2025 report, we identified a number of key trends, including:

• A growth in data investment after a period of caution;
• The use of AI to augment (not replace) data teams; and
• A continued concern over data quality and an emphasis on building trust in data.

We also wanted to get some thoughts from leaders in the industry on what they’re seeing beyond these numbers. We held a webinar roundtable recently featuring our own Senior Manager, Developer Experience, Jason Ganz, as well as Yannick Misteli, Head of Engineering, Go-to-Marke at Roche and Jenna Jordan, Senior Data Management Consultant at Analytics 8.

You can watch the full webinar yourself. Here are some of the main takeaways from our participants.

The biggest changes in analytics engineering

In all, 2024 was a transition period for analytics. Instead of any major new developments, the industry mostly laid the groundwork for the next transformations to come. Given how overwhelming sudden change can be, this stability was refreshing.

For the most part, organizations in 2024 focused on developing fully mature data pipelines. Observability and data quality became more prominent concerns across the board.

Others were excited by some of the changes announced by dbt Labs over the past year. In particular, more large organizations took advantage of dbt Mesh. Enterprise found that new features incorporated into Mesh enabled them to manage a growing number of data transformation models, and introduce the additional flexibility and distributed approach to data that a data mesh architecture offers.

There was also excitement around dbt Labs’ acquisition of SDF. SDF is a high-performance toolchain that can represent various SQL dialects. That means SDF can faithfully emulate all popular data warehouses locally, providing data pipeline developers with immediate feedback before they even run a single line of SQL.

The analytics engineer role in 2025

Analytics engineering is a fairly new industry whose practitioners focus on providing clean data sets to end users, using software engineering best practices to maintain a clean analytics codebase.

Originally, this role took on some of the scope of data analysts and some of the scope of data engineers. Different organizations have chopped this up differently. The result is we’re seeing more of a blurring of lines between these three roles.

An analytics engineer in one org, for example, may continue to move left and take on more traditional data engineering tasks. Others, however, may move more to the right and start interfacing more often with data stakeholders, bringing more of the business perspective into their work.

We’ve seen similar movements in other roles as well. Some data analysts, for example, have been drifting more leftward, taking on dbt modeling and some of the other tasks typically done by data and analytics engineers.

At larger organizations, however, we’re seeing a clearer delineation between these roles. At these companies, analytics engineers are becoming more focused on data modeling.

We’re also seeing the emergence of visual analytics engineers who excel at turning data into reports in tools such as Qlik, ThoughtSpot, and Tableau. Typically, at such organizations, we see a 3:2:1 ratio between analysts, analytics engineers, and visual analytics engineers.

AI's impact on analytics engineering

AI isn’t just changing how we do business. It’s changing analytics engineering as well. Our latest report found that 80% of data practitioners are using AI in some way as part of their workflows.

AI is impacting all roles in the data lifecycle - data and analytics engineers, analysts, and business stakeholders. Participants have found they can use AI to:

• Better define requirements
• Improve code quality and efficiency
• Connect BI tools to data using text-to-SQL capabilities

The primary use of AI currently in data workflows is to reduce drudgery and generate code (dbt YAML files, for example) for repetitive tasks. Offloading the drudgery to tools like dbt Copilot has given analytics engineers and data engineers more time to focus on the creative side of their work. This can involve an increased focus on data handling, for example, or on revising the company’s data architecture.

Additionally, AI is empowering more people who are just starting out on their journey with dbt or even SQL. Even if someone doesn’t know exactly what goes into a dbt project YAML file, for example, they can use AI to give them a boost and move up the stack, enabling them to contribute to the company’s dbt projects more directly.

The importance of context in AI applications

On the other hand, in many ways, we’re not there yet. We can ask LLMs any questions we may have about our data. But they don’t always return the correct answers.

We continue to see accuracy issues with AI when LLMs and AI agents don’t have full access to our codebases or standards. Passing data context to LLMs - e.g., using technologies such as Model Context Protocol (MCP) - will become increasingly important over the next year.

The semantic layer also has a role to play in improving AI results over your data. In the past, different departments in an organization have had their own, slightly divergent calculations for key data points, such as sales volume or revenue. A semantic layer standardizes vocabulary and meaning for these key metrics across teams. Connecting LLMs to this layer is one way of increasing the likelihood that, when we ask them a question, we’ll get the right answers back.

Finally, good documentation will become more important than ever. Writing out detailed descriptions of models, fields, etc. provides more context to a query engine on how to query that data.

Analytics 8’s Jordan said she’s not a huge fan of using AI for documentation. In many cases, however, she says data teams can get a limited Ouroboros effect going by using AI itself to stub out some of the initial documentation (which you then refine and supplement with your own human knowledge).

Roche’s Misteli concurred by citing his team’s experience in learning the importance of documentation firsthand. Roche built a chatbot on top of its technical documentation. Engineers panned it, however, saying it wasn’t useful.

After analyzing why, Roche came to a simple conclusion: its documentation wasn’t up to snuff. That led to an effort to clean up the docs and make them more valuable - for both humans and LLMs.

“If your documentation isn’t good,” Misteli concluded, “your chatbot won’t be, either.”

Data quality remains a top issue

Surprising to no one who deals with data for a living, data quality still remains a top issue. 56% of survey respondents identified data quality as a problem.

Participants emphasized that data quality isn’t just a technology problem - it’s a process problem. You need a mature analytics workflow in place to find and fix problems early in order to minimize their business impact.

Jordan from Analytics 8 agreed and said the key is “bringing your analytics systems closer to your operational systems and organizing by domain.” That includes using data mesh architectures, data contracts, data quality testing, and defined Service Level Agreements (SLAs) for data. It’s also important to get all data stakeholders in the room early on - i.e., when the data contracts are being written.

Predictions for the future

In terms of what’s coming up in the near future, the fusion of AI agents with semantic web technology holds a lot of promise in realizing the true value of AI agents. Connecting these agents to the ontologies that exist inherently in our web-based data could provide agents with true semantics and reasoning capabilities.

On the analytics workflow front, improvements such as SDF and the Visual Studio add-on for dbt promise to cut down dev time for analytics code changes even further. We should also expect to see LLMs take on a larger role through the analytics development lifecycle, particularly via ChatGPT-style interfaces for analytics.

No one can predict the future, of course. Given current trends, though, we can expect that by the end of 2025, it’ll be easier to transform, publish, find, and utilize data than we ever previously thought possible.