How I Built A Predictive Data App That Made Retail Forecasting Click

At Brooklyn Data, we’re always exploring how to make data more accessible, actionable, and trusted. Over the years, we’ve seen a persistent challenge among modern e-commerce companies: different departments — merchandising, marketing, regional management — are all using customer data, but rarely in the same way, or even in the same place.

So, we decided to create a data app to help retailers manage that complexity. In building the Customer 360 Embedded Data App, we want to give every team visibility into their part of the business, from online sales to customer segments to campaign results, and to make sure they can trust what they were seeing without spending time applying endless filters. One of our biggest goals was to integrate predictive analytics directly into the workflow, so they can forecast performance, refine their marketing strategies, and operationalize decisions all in one place.

Here’s how we did it.

Inside the build: Each layer unlocked the next, and operationalized AI

We’ve been building in Sigma for years, and this app really grew alongside the platform. We started with a simple customer 360 dashboard—just a few core insights, but with row-level security built in so each manager could view only their own data.

Then, once input tables became available, we added a new layer: the ability for users to enter their marketing budgets directly. That was the point where we began connecting those inputs to forecasting models hosted in Snowflake. Sigma handled the front end, Snowflake ran the models, and the flow between them was seamless.

What started as a dashboard turned into a robust operational tool.

With the release of data apps, we added even more functionality. We implemented modals to make the interface more user friendly, and layered in an approval workflow so teams can review, refine, and sign off on forecasts, all within the same workbook. What started as a dashboard turned into a robust operational tool.

One challenge came up when we started using modals. Because Sigma implements modals as additional workbook tabs, our original approach—embedding individual tabs—no longer worked. Modals simply wouldn’t load unless the whole workbook was embedded. We worked with the Sigma team to find a workaround: embedding the full workbook and using the nodeId parameter to direct users to the right tab. That’s what I appreciate about Sigma. There’s almost always a way forward.

Empowering business users to trust what they see

Because we work closely with leadership teams and non-technical stakeholders, making apps feel intuitive was always a priority. We decided early on that the experience needed to be self-explanatory—people didn’t want to guess what filters to use or what a number meant. They wanted guidance directly in the app.

So, we added instructions and plain-language descriptions throughout. We used Sigma’s description fields for visualisations, which show a little “i” icon when you hover. That let us define each metric clearly—how it’s calculated, what it includes, and why it matters. It’s a small feature that made a big difference in helping people trust the data they were seeing.

Because we work closely with leadership teams and non-technical stakeholders, making apps feel intuitive was always a priority.

We also added export buttons, which, as every data analyst knows, is both a blessing and a compromise. Analysts don’t love it, but users ask for it constantly. Being able to export the full data table gave teams the freedom to dig in further, run their own numbers, and bring the insights into their own workflows. And because everything we included came directly from user feedback, we built something that felt usable, even for folks who never thought of themselves as “data people.”

All that set the stage for my favorite feature in the app: forecasting. This allowed users to do more than just observe performance, and actually shape it. With forecasting, teams could input their marketing budgets, allocate them across different channels, and instantly see how those investments might translate into revenue. What made this especially powerful was that it didn’t require technical intervention. Business users can experiment with campaign strategies, optimize budget allocation, and get ROI projections without ever needing to ask a data scientist for help.

And because it was all built into Sigma, users don’t have to switch tools or lose momentum. The entire planning loop—from ideation to approval—stayed right there in the dashboard.

Lessons learned, and what I’d tell any new builder

If you’re building your first data app in Sigma, I’d say this: don’t underestimate what’s possible. The learning curve is so much lower than you’d expect, especially if you’ve worked in more rigid BI tools before. Sigma is flexible, intuitive, and there’s almost always a way to do what you want—even if it takes a little creativity. The payoff is an app that gives end users more control while taking full advantage of your live cloud data warehouse architecture.

And if you get stuck, use the live chat. I’ve had so many moments where I thought, “There’s no way this will work,” and someone from Sigma suggested a workaround I never would’ve come up with on my own.

What made this project really rewarding was how easily we could integrate predictive models in a way that actually got used.

But what made this project really rewarding was how easily we could integrate predictive models in a way that actually got used. I’ve spent time building machine learning models that no one ever touches because they’re too hard to operationalize. With Sigma and the Cloud Data Warehouse, it just worked. We plugged in the model, designed a simple UI, and suddenly those insights were being used to make actual marketing decisions.

It felt like the kind of build where the tech got out of the way—and the strategy came to the front.

‍