The Pitfalls of Self-Serve Analytics & Why the Virtual Chief Data Officer is the Game-Changer

it's a pleasure to have everyone, and we're really grateful for you taking the time for the half hour. We're excited to have you here so that we can dive into the pitfalls of self-serve analytics and why the virtual data officer is a game changer.

Whether you're here to learn, ask questions, or explore new ideas, we hope you'll find this session valuable and engaging. Over the next thirty minutes, we'll talk about the risks and limitations of dirty data, introduce the virtual data officer and its role in data strategy, and show real-world examples of how organizations are using virtual CDOs to transform their analytics.

On the right-hand side of your screen, you should see a Q&A box. Feel free to ask any questions, and we will answer them as quickly as we can. These questions will only be shared with the Tamr team, so no need to be shy.

Now, let me introduce ourselves. Hi, I’m Eleni Partakki, a solutions engineering lead at Tamr. I’ve been with the company for over two and a half years. I have a background in machine learning research and human-machine interaction, with an emphasis on visualization and how people perceive data, using tools like Tableau, Power BI, and Looker.

I’ll pass it on to Alex to introduce himself.

Alright, I’m Alex Pagan. I’m one of the cofounders of Tamr, originally from MIT’s databases lab, where we worked on the Data Tamr project. Over the past decade, I’ve had the chance to work on platform engineering, user experience design, AI applications, and closely with customers to solve some of their toughest data challenges. I’m excited to talk about the future of data democratization with Tamr.

So without further ado, let’s get started.

I was at Big Data London last week and had conversations that inspired some of the thoughts I’ll share today. I was chatting with a company representative who mentioned a report from a few years ago, stating that only 27% or 30% of organizations were able to make full use of their data to generate actionable insights. Even with a growing data skills gap cited as a primary reason, I think after these conversations, if only 30% of companies can use their data, there’s a bigger problem.

I kept thinking back to when I used to build visualizations for testing. What do you do when there’s an outlier? Right-click and exclude. So, in your experience, what do you think is going on with that? Have you seen anything similar?

Yeah, it’s funny you mention that because I was reading a study published by Fivetran. They looked at publicly available data from Snowflake and Redshift regarding the scaling properties of their data, and found that only about 20% to 30% of the workload that people are paying for is what’s produced by BI tools or used by data scientists. The rest is primarily getting data into the platform and transforming it for downstream use. I see that as a proxy for what you’re talking about—the reshaping needed to make data useful.

It’ll be great to dig into why people aren’t getting more value from these data applications.

Well, why? Let’s get into it. As the title says, we’re going to discuss the pitfalls of self-serve analytics and how a new path forward involves the virtual chief data officer.

To frame our conversation, let’s look at how businesses have leveraged data over the past 20 years. Competitors began gaining leverage by modeling business processes as data, leading to star and snowflake schemas. The competitive advantage at that time was having the infrastructure to store and query data, making data-driven decisions.

Moving forward, internet companies became proficient in data analysis, building sophisticated distributed systems now available to consumers. As a result, businesses can now store and query petabytes of data. But just having a data warehouse isn’t enough—you need to work the use of that data into everyday processes. That’s where self-serve analytics comes into play. Why should only data teams build new insights? Business users with the most context should also access data and build what they need.

But now we’re seeing rapid innovation with AI, and everyone’s focus has shifted to leveraging AI alongside data for differentiated experiences. It’s exciting but also presents the question: Do we really trust our data?

That’s a great introduction to why self-serve analytics projects fail and the challenges people face when trying to make data access a core part of their business.

To simplify, you’re saying the challenge was once isolated to specific technical teams accessing results, but now we’re moving toward empowering business users to create their own dashboards and get the answers they need. Is that correct?

Exactly, and data teams are building complex pipelines, buying tools, and working toward this goal. However, if the central view of the data doesn’t match business users’ needs, you still get queries and emails asking for specific slices of data.

The tools end consumers use to build analytics are complex. You can do amazing things, but it takes experience to get comfortable with them. Even though spreadsheets are simple and widely used, ensuring their data quality is difficult, especially when shared. Once a spreadsheet is sent around, who knows if the data is still correct?

Data quality is a constant challenge with self-serve analytics. Often, BI users right-click to ignore outliers in data, which papers over interesting problems. Without feedback in analytics tools, dashboards can lead to missed insights.

This is where we see overloaded data teams managing many tools while trying to keep up with data quality, frustrated consumers, and eventually fragmented data silos. If we're not careful, we could end up with just spreadsheets circulating and no single source of truth.

Now, I’d like to introduce the virtual chief data officer and how you can reframe your approach to your business’s data.

I’ll jump over to Tamr to set the stage. Let’s say I’m a sales manager selling hardware to multinational companies, and I want to work with a prospect. I can look up their information in Tamr, get external and internal data, and leverage AI to ask questions about the company.

For example, I can ask if ABB was involved in a merger and get detailed information based on the AI’s understanding. From there, I can consult the virtual CDO to develop a strategy for the prospect.

Forgive me for interrupting, but why use this instead of something like ChatGPT? Why not just load the data into ChatGPT or use an API from OpenAI?

The models are improving, but fundamentally, they don’t know your business. Even the most advanced model doesn’t know your customer’s preferred shipping address or recent sales transactions. The virtual data officer allows you to leverage your improved data quality and provide context for AI to interact directly with data consumers.

‍

‍