How to design AI-first products - Podcast recap

A recap of Marc Dupuis' conversation with John Krohn on the Super Data Science Podcast

In a recent episode of the Super Data Science Podcast, I sat down with host John Krohn to discuss a problem that's plaguing data teams everywhere: the endless cycle of ad hoc analysis requests.

"Ask any data scientist what eats up most of their time, and you'll hear the same answer: ad hoc analysis requests. 'Can you pull this metric?' 'What's our conversion rate by channel?' 'How did that campaign perform?” These questions aren't particularly complex, but they pile up fast, turning strategic data professionals into ticket-taking query machines.

The conversation revealed something striking: data teams are spending 50+ hours a month just answering one-off ad hoc analysis questions. That's more than a full work week lost to repetitive analysis instead of strategic work. Here's what I shared about how AI data analysis is changing that equation.

The data team bottleneck: A universal problem

Before starting Fabi, I lived on both sides of this frustration. As a product manager at companies like Clari and Assembled, I was constantly asking my data science colleagues for "quick pulls." As someone who later partnered closely with data teams, I watched brilliant data scientists get buried under requests that kept them from doing their best work.

My co-founder Lei experienced this firsthand. He'd be in the middle of designing an experiment or building a model, then get pulled away to run a Jupyter Notebook, export data, and send it via Slack. Five times a day. Every day.

The problem isn't that these questions don't matter (they do). The problem is that data scientists shouldn't be the only ones who can answer them. And with AI, they don't have to be.

Why traditional BI platforms haven't solved this

You might be thinking: "Don't we already have BI platforms for this?" Yes, but traditional BI tools have their own issues.

Building dashboards takes forever. Data teams spend weeks or months building a perfect Python dashboard, only to watch users export it to Excel anyway. One data scientist told me they'd be "lucky if someone accidentally clicked" on their dashboards. That's heartbreaking.

Self-service analytics remains elusive. If the question isn't already answered by an existing dashboard, you're back to filing a ticket. The tools are powerful, but only if you know SQL, understand the data model, and have time to dig in.

Data warehouse costs spiral. Give business users direct access to query your warehouse and watch your bills explode. Without proper guardrails, well-meaning marketers can accidentally run queries that cost thousands.

This is where AI BI fundamentally changes the game.

How AI accelerates data analysis (without the chaos)

At Fabi, we've built what we call "vibe analytics". An AI BI platform where AI handles the heavy lifting of data analysis while data teams maintain control. Here's what that looks like in practice:

1. Turn hours of analysis into minutes

Our customers are seeing 90%+ reductions in analysis time. What used to take a data scientist an hour or more, understanding the question, writing SQL, debugging, creating an AI data visualization now takes 10 to 15 minutes. Pilot program evaluations that used to take 2 to 3 weeks now take only a few hours.

The Fabi AI Analyst Agent writes SQL and Python for data analysis, creates charts, and handles the mechanics of analysis. But it's not running wild; data scientists define which data sources the AI can access and guide the analysis direction.

2. Enable true self-service analytics

When we work with data teams, one of the first things they notice is that their ad hoc analysis request queue disappears. Teams like Aisle using Fabi have eliminated 40-50 monthly data requests saving 50+ hours a month.

How? Business users can now explore data themselves for initial questions through self-service analytics. They're no longer filing tickets for basic analysis. And when they do reach out to the data team, they're coming with much better, more informed questions because they've already done the initial exploration.

3. Automate data workflows to meet people where they work

Here's a rule of thumb we share with every data team: if you have a choice between building a dashboard and building a data workflow, always pick the workflow.

ROI on dashboards is nearly impossible to prove. But if you automate insights directly into Slack, email, or Google Sheets, wherever people actually spend their time, you guarantee your work gets seen and used.

With Fabi's Smartbooks (think Jupyter Notebooks with AI superpowers), you can build business-specific data analysis once, then convert it into an automated AI reporting workflow. Push AI-generated summaries to your company's Slack channel every morning, send executive reports via email, or update Google Sheets automatically. Your data analysis work finally gets the spotlight it deserves.

The guardrails that make this safe

Giving everyone access to data sounds risky. What about costs? What about accuracy? What about people drawing wrong conclusions?

We've built several safeguards into Fabi specifically to address these concerns:

Smart memory: When AI pulls data to answer your question, we temporarily store that information. Follow-up questions use the saved data rather than pulling fresh data each time. This keeps costs down and speeds up your analysis.

Controlled access: We encourage teams to grant the AI access only to their most important, reliable data sources, not to everything in their database. This reduces costs and helps the AI give you more accurate answers by working with clean, trusted data.

Real-time data collaboration: Our Smartbooks let data scientists and business users work together in the same workspace. If someone's unsure about their analysis, they can instantly share a link for the data team to review. It's like Google Docs for data analysis.

Undo and version history: AI can take wrong turns. We automatically save versions as you work, so you can always see what changed and go back to an earlier version if needed.

What this means for data scientists

The question we get most often: "Does this mean companies won't need data scientists anymore?" Not even close. AI is changing what data scientists do, and honestly, it's changing it for the better.

You get better questions to answer. When business counterparts can explore data themselves first, they come to you with much more informed questions. Instead of "Can you look at feature adoption?" you get "I noticed adoption drops after day three for users who skip onboarding. Can we segment by persona and test an intervention?"

You focus on high-leverage work. If AI handles writing Plotly charts and debugging SQL, you can spend time on what actually moves the business: understanding the P&L, designing experiments, building models, supervising AI outputs with your statistical expertise.

Your work makes more impact. When you're not buried in ad hoc analysis requests, you can focus on strategic initiatives. Build automated data workflows instead of one-off analyses. Work on forecasting models instead of pulling last week's numbers for the tenth time.

I think this is honestly the most exciting time for data scientists in the past decade. You're getting out of the less valuable tasks and finally getting to exercise the skills you went to school for.

Real results: how teams are using AI to move faster

One of our customers, Aisle, a retail analytics platform, reduced its data analysis time by 92% with Fabi. Their product manager, Chirag, is part of a lean 15-person tech team supporting 500+ brands. Before Fabi, his data team was drowning in 40-50 ad hoc analysis requests per month.

Now? The brand managers answer their own questions through self-service analytics. Chirag focuses on strategic growth. Pilot program evaluations that used to take weeks now take hours. And within the first month, 100% of their brand managers were using Fabi for weekly AI reporting.

That's the kind of transformation we're seeing across teams of all sizes, from early-stage startups to established enterprises.

Getting started: don't wait for perfect data

Here's something crucial: don't wait to have perfectly clean data before using AI for analysis.

Yes, spend 90% of your time on data cleaning when you're doing analysis, that's critical for good results. But you don't need a pristine data warehouse to get started with an AI BI platform. You can clean data in SQL or Python right before your analysis, even if it's sitting in messy spreadsheets.

The best data teams we work with "lean into the mess." They explore, improvise, and ship useful insights without waiting for the perfect setup. Because in fast-growing companies, even core metrics evolve constantly. Your ARR definition might change next quarter when you add consumption-based pricing. Don't let perfection be the enemy of progress.

The future of data teams

AI isn't replacing data work, it's elevating it. The repetitive, low-value ad hoc analysis tasks that filled your days are becoming automated. The strategic, high-impact work that you're uniquely qualified to do is finally getting the time and attention it deserves.

At Fabi, we're building the BI platform that makes this shift possible: where data teams escape the ad hoc analysis request queue, where business users can leverage self-service analytics for basic questions, and where insights actually drive action through automated data workflows and AI reporting that meet people where they work.

If your data team is spending more time answering ad hoc analysis tickets than driving strategy, it's time for a change.

Ready to eliminate 50+ hours of ad hoc analysis requests per month? Try it for free at Fabi.