Grok's Data Bot looks past email opens to what customers did

On November 2, SpaceXSI, the company behind Grok, published a post about Data Bot, a shared AI teammate its data science team built inside Grok Bot. Anyone at the company can message it a question, and it queries the company's data warehouse, runs the analysis and sends back the answer in the form each person prefers. One product manager gets a short headline with a chart.

The data team taught it through what SpaceXSI calls skills: how the company defines its core metrics and which tables they come from, how it runs and reads A/B tests, its chart standards, and how it builds dashboards. Data Bot also keeps a shared memory of what it learns, and the data team reviews and approves those additions before everyone else gets them. Teams can add it to their Slack channels and have it post a weekly update on a metric they care about.

One product manager described it as having a data scientist in his pocket. The company estimates that in one week, Data Bot handled about 70 hours of work that would otherwise have landed on its data science team.

The useful part for marketers is what happened to the questions people never bothered to ask. SpaceXSI says that when getting an answer meant opening a dashboard or starting an analysis, people let some questions go. Now they ask more, and follow up without starting a new analysis.

The same post has a section on marketing, and it's the most useful part for anyone with an email list. SpaceXSI's lifecycle marketer gives Data Bot an email campaign ID from Braze, the email platform the team uses, and asks what recipients did afterward. Data Bot then checks the warehouse to see whether the people who clicked went on to use the feature the email promoted.

For the launch of The X Subscription, the marketer wanted to know how many email recipients actually switched plans. SpaceXSI says Braze's built-in conversion metric couldn't answer that, so Data Bot counted who received the email and who went on to switch. The company's revenue growth director uses it the same way for events, tracking how they add to pipeline and speed up deals already in motion.

The company frames the win simply: its marketing team now answers more of its own questions without waiting in line for the data team, but still draws on that team's expertise. That second half is the important one. Speed without that expertise just gets you wrong answers faster.

Open rates have been a shaky number for a while. Apple's Mail Privacy Protection hides the reader's IP address and, in Apple's words, "prevents senders from seeing if you've opened" their message. So when Apple Mail readers turn it on, your open rate is partly a guess wearing a percent sign.

Apple also notes that the protection doesn't extend to links. Clicks are a better signal than opens, but a click is still a visit, not a sale. The question SpaceXSI's marketers are asking is the right one for every business: of the people who got this email, how many did the thing we wanted?

Marketing insight: Stop reporting opens as a win. Report clicks as interest and the booked call, sale or signup as the result.

Data Bot runs on Team Bots, which SpaceXSI launched on September 28 in public beta on its Teams and Enterprise plans, with pre-built Team Bots for sales, product management, marketing and data analytics to start from. That post says SpaceXSI's version uses shared, read-only credentials to query approved tables, and that its data team had already spent two years building a library of skills for working with more than 45,000 tables before handing them to the bot.

The bot is only as good as the metric definitions and data you give it, and most small and mid-sized businesses don't have a data warehouse or a data team writing skills for one. The 70 hours is SpaceXSI's own estimate about its own team, with no outside measurement, and the post doesn't share an error rate. Treat it as a strong example of the workflow, not proof of the payoff.

• Define the outcome first. For each campaign, write the one action that counts as success before you hit send.
• Tag every link. Add UTM parameters to the links in your emails, as Google Analytics explains, so each visit carries the campaign name and you can see which emails send traffic.
• Connect the click to the result. Match campaign visits to the bookings, orders or signups in your CRM or store, even if it's a spreadsheet for now.
• If you try an AI analyst, start read-only. Give it written metric definitions, and check its first answers against a number you already know.

What AI changes is the price of asking the question that always mattered: did this campaign change what customers did? The businesses that answer it every week will know which emails to send again, and which to quietly retire.

Want help measuring what your marketing actually does? Talk to our founders below. See you in the next edition.