Giving every team a data scientist with Grok Bot

At SpaceXSI, our data science team works across the company, helping teams answer difficult questions and finding new ways to incorporate data into our products. Over time, we've developed an intuition for our data, including how to interpret it and where its limitations lie.

But our small team cannot answer every question that arises in a fast-growing company. People come to us throughout the day with questions that may be straightforward but still require our expertise to answer accurately. This can easily make us a bottleneck.

A few months back, we started exploring how AI agents could change that relationship. We wanted to give everyone a data scientist, available around the clock, that could draw on our collective knowledge, while shifting more of our own work toward building and improving that AI teammate.

We built Data Bot, a Team Bot in Grok Bot that we've taught how our data science team works. It can query our warehouse, conduct analyses, and return results in the form each person prefers. Unlike personal bots, which depend on what their owners know to teach them, Data Bot draws on a shared set of skills that our data team maintains and improves for everyone.

Teams can also add Data Bot to their Slack channels, where it answers questions and contributes proactively to ongoing work. As we improve its skills, everyone benefits from the same growing body of expertise.

Teaching Data Bot how the data team works

Data Bot needs enough context to approach a question the way our team would. We give it that context through skills, which cover these main areas:

• Metrics and analysis. We define our core metrics across the business, such as active users and revenue, and identify the source tables and query logic behind them. The skills also guide Data Bot through investigating changes in those metrics, from breaking down a spike or drop by segment or region to checking for holidays or logging changes before attributing it to a product change.
• Experiments. These cover how we set up A/B tests, run power analyses, check on exposure balance and sample bias, plan rollouts, deep dive into unintuitive results, and write up readouts once experiments have reached power.
• Charts and dashboards. We equip Data Bot with data visualization best practices, such as using line charts for trends over time and horizontal bars to compare categories, along with our team's chart themes and style guides. Other skills guide Data Bot through creating dashboards, so it can turn a chart in chat into a Databricks or Hex dashboard the whole team can use.
• Data engineering. These cover our best practices for writing and reviewing dbt models, our conventions for automated pipeline alerts, and how to add analytics event logging so we can measure how people use new features from the moment they launch.

Data Bot learns from experience

We also give Data Bot shared memory so it can retain knowledge that emerges through day-to-day use. For example, it might learn which teammate owns a particular pipeline or that a previous drop in an important metric turned out to be a regional holiday. The next time it investigates a similar change, it can draw on that experience and check the holiday calendar before attributing the drop to a product change.

As the Team Bot's managers, we can review these proposed additions to its memory and approve or reject them before Data Bot shares that knowledge with everyone who uses it.

Helping PMs better understand their products

Product managers have a constant stream of questions about how people use their products, many of which fall outside what their existing dashboards can tell them. Answering those questions has traditionally meant digging into the data themselves or asking someone on the data team for help.

Data Bot gives PMs a way to explore those questions as they arise.

Kevin Niparko, a PM at SpaceXSI, can message it in Grok Bot much as he would a teammate, get an answer in minutes, and follow up without starting a new analysis. He describes the experience as having a data scientist in his pocket.

The ease of asking means PMs can investigate many more questions throughout the day. Previously, even opening a dashboard or starting an analysis in Cursor introduced enough friction to discourage some questions. Now Kevin can explore things he might otherwise have let go, without having to ask a data scientist for help.

Data Bot also adapts to how people prefer to consume information. Kevin likes a concise headline accompanied by a chart, so Data Bot presents its findings in that format, letting him move quickly from a question to an answer he can use.

Helping marketers understand what their campaigns accomplish

Marketing teams need to understand what customers do after receiving a campaign. Email platforms track opens and clicks, but those metrics don't always show whether the campaign achieved its goal.

Jon Brenner, who works on lifecycle marketing at SpaceXSI, uses Data Bot as an on-call analyst. He can give it a Braze campaign ID and ask what recipients did afterward. Data Bot queries the warehouse to see whether people who clicked went on to use the feature.

For our recent launch of The X Subscription, Jon wanted to know how many email recipients actually switched plans. Braze's built-in conversion metric couldn't answer that question, so Data Bot queried the warehouse to track how many recipients received the email and how many went on to switch plans.

Katy Abbaszadeh, our director of revenue growth, uses Data Bot to measure the impact of events on sales. She tracks how events contribute to pipeline generation and help accelerate existing deals. She also uses Data Bot to monitor weekly progress against targets, investigate what might be slowing progress, and turn the results into charts for presentations.

With Data Bot, our marketing team can now answer more of their own questions without waiting for the data team, while still drawing on the team's expertise.

Letting Data Bot handle entire analytics workflows

Data Bot can take on work that goes well beyond answering individual questions. The data team has given it the tools to conduct more substantial analyses, including projects that would ordinarily require several rounds of collaboration with a data scientist.

A PM investigating a change in product adoption, for example, can ask Data Bot to work out what happened. Data Bot can plan the analysis, query the warehouse, and explore possible explanations. It can then produce a written report with charts or build a dashboard in Hex that the team can continue using.

Data Bot joins team channels and contributes proactively

Teams can add Data Bot to their Slack channels and give it responsibility for ongoing reporting. For example, a team can ask it to track an important metric and post weekly updates, following the reporting conventions our data science team has established.

Data Bot can run that workflow on a schedule and contribute updates directly to the group conversation. Colleagues can then ask follow-up questions or have it investigate an unexpected change, much as they would with another teammate.

Data science expertise for all

Data Bot adoption is growing quickly within SpaceXSI. Last week alone, we estimate it handled about 70 hours of work that would otherwise have landed on the data science team.

We still handle complex analyses that require more judgment, and we're learning where Data Bot needs additional guidance. Our goal is to give colleagues across the company an AI teammate who can draw on our expertise whenever they need it, while leaving the Data Science team more time to focus on the questions that need our direct attention.