Senior Product Manager at @databoxhq | AI Product Strategy, Growth & PLG Expert Free AI Analytics Skills: https://teapawt.net/t.co/FLk8fawO6c
Joined February 2012
- Tweets260
- Following766
- Followers148
- Likes228
Pinned Tweet
Adding AI to analytics is easy now. Building an analytics platform AI can actually rely on is the hard part.
That's what we have been heads down on for the last couple of months.
The new @databoxHQ is live today.
Performance data, governed metrics, and business context, all in one connected foundation. From there, AI can analyze results, run your recurring reporting, and turn insights into action on its own.
This is what we've been calling an agentic analytics platform: built for people who own a number and need to explain and improve it, fast.
Try it and tell me what breaks. databox.com/
We finished #1 Product of the Day on @ProductHunt 🏆
This year we launched on Product Hunt 4 times before this one:
#3 Genie
#3 Databox MCP
#2 Skills Marketplace
#2 Artifacts
Yesterday, MCP Connectors took #1.
Here's what it does and what we learned 👇
Your dashboard shows you that a number moved. It doesn't tell you why.
The why lives somewhere else. A stalled deal in HubSpot. A pricing objection in a Slack thread. A bug in Linear that shipped on Tuesday. Finding it means opening three tools and piecing it together yourself.
With MCP Connectors, you connect those tools to Genie, our AI Analyst.
Ask "why did MRR drop 8%?" and you get the number and the reason, with links to the actual deal or thread behind it. Genie can also act on what it finds.
3 things the launch comments made clear:
1. Almost nobody finds the "why" in their dashboard. It's always in the CRM, Slack, support desk, or a doc somewhere.
2. People want more than answers. They want AI to take action, like pausing an ad campaign that's burning budget.
2. Trust matters most. The best questions were about where an answer came from and which tool you'd let AI touch first.
Thank you to the @databoxHQ team for shipping this, and to everyone who upvoted, commented, and tried it.
So, which tool would you connect first?
Signups drop 18%. The dashboard shows it in a second.
The reason takes an hour. It's in a Linear ticket, a Slack thread, a HubSpot note.
Last week I asked the @ProductHunt community where they go when a number moves. Almost nobody said the dashboard. They said the CRM, Slack, the support desk, Linear, "a doc somewhere."
So that's what we built. MCP Connectors by @databoxHQ is live on Product Hunt today.
You connect the tools your team already uses, like HubSpot, Slack, Notion, Linear or Klaviyo, or add any MCP server by pasting its URL. Genie, our AI analyst inside Databox, checks those tools when it answers your question. Ask why signups dropped and you get the number and the reason in one answer.
It can also suggest what to do next, like filing the fix or flagging an at-risk account. Nothing runs until you approve it. I don't think anyone should let an AI act in their CRM without a human saying yes first.
Which tool would you connect first? My bet is Slack. That's where most of the "why" ends up.
We're live here if you want to check it out or leave feedback: producthunt.com/products/dat…
Gave Claude Code one prompt on a completely empty folder. Opus 5.5 wrote its own animation code, screenshotted my real product for content, and composed the music. Zero manual edits besides one correction I gave it afterward.
The prompt: make a modern slick and punchy video for a startup that works on agentic analytics, just javascript, playwright and ffmpeg, code the animation and take a video of it, download assets from the site or use playwright for screenshots, include music aligned with the video.
Ziga Potocnik retweeted
Most reporting problems aren't reporting problems.
They're 12 tools that don't agree with each other.
Ziga Potocnik retweeted
Every Monday I opened three tabs before coffee. GA4, HubSpot, a spreadsheet someone built two roles ago.
Not because I needed more data. Because I didn't trust any single view of it.
That's not a dashboard problem. That's a trust problem.
Most MCP servers connect one tool to one data source.
That's the wrong unit. Nobody has one data source, and nobody uses one AI tool.
So you end up rebuilding your entire stack inside every client you touch.
@databoxHQ MCP goes the other way. One connection, 130+ sources, any MCP client.
We're bringing "set it and forget it" to AI analysis.
Before building it, I asked people how they keep tabs on their numbers day to day. Most said one of two things. Either they open the same dashboard every morning out of habit, even when nothing's changed. Or they don't check often enough, and by the time something's off, it's already cost them a few days.
Routines fixes that. You tell Genie, our AI analyst inside @databoxHQ, what to check and how often. From there it runs on its own, no re-prompting, and tells you what actually changed, good or bad.
Live on @ProductHunt today 👉 producthunt.com/products/dat…
If you could put one recurring check on autopilot right now, what would it be?
Everyone says "launch on the weekend."
I checked. Pulled every @ProductHunt launch from Jan 1 to Sep 1, 2026: 4,962 launches.
Saturday needs 165 points to break top 5, Thursday needs 235. Weekends aren't easier because of some algorithm quirk though, they're just less crowded.
Didn't build this by hand either, asked @databoxHQ 's AI analyst one question and it queried the whole dataset.
Everyone says "launch on the weekend." I checked, pulled every Product Hunt launch from Jan 1 to Sep 1, 2026: 4,962 launches across 244 days. Some of what I found confirmed the myth. Most of it didn't.
Before you run a campaign, do you know what normal looks like for that metric this time of year?
Not last month. Not last quarter. The actual typical range, adjusted for seasonality, with a confidence level attached.
We just launched a free Claude skill built by Laszlo Fazakas from Arcanian that sets exactly that. It pulls the metric's real monthly history from Databox MCP (ideally 13+ months for a full year-over-year pattern) reads the seasonal cycle, and outputs a baseline band with the low-to-high range, the central tendency, and a plain-English note on how to judge a future result against it.
The technical detail worth calling out: it never fabricates a trend to fill a gap. If the history is too short, it labels the baseline provisional. That honesty is the whole point.
Free to download: databox.com/skills-marketpla…
If you're doing your data analysis in Claude, you know the hidden work in cross-channel paid ads analysis is always CONTEXT.
The numbers are usually the "easy part". Spend, conversions, CPA, ROAS, CTR.. The harder part is the LLM / AI knowing what those numbers are supposed to mean for your business.
Is LinkedIn there to generate pipeline or create demand?
Is Google Search expected to carry efficiency?
What CPA is actually acceptable?
Which channels are allowed to look “worse” because they play a different role in the funnel?
Without that context, cross-channel reporting can turn into a leaderboard that compares platforms and campaigns as if they were all doing the same job.
If you don't know answers to all of the questions above (any beyond) - no AI is going to help you get the insights and analysis you need.
And that’s exactly what we had in mind when building skills for the Databox marketplace: AI shouldn’t just see the data. It should understand the context behind it.
One of those skills dropping today is the Cross-channel paid ads performance report. Before it analyzes anything, you give it the context that matters: your conversion goals, CPA/CAC targets, budget, and the role each platform plays in your funnel.
It then connects to your live paid data through Databox MCP and builds one structured report across platforms.
Good analysis isn't just data + AI. It's data + business context + a methodology for deciding what matters.
The Skill is free if you want to try it on your own paid accounts: databox.com/skills-marketpla…
We just launched a free Claude skill for LinkedIn Ads reporting and it's one I've been wanting to share with customers for a while.
The thing I see most with B2B teams running LinkedIn Ads: the monthly report exists, but it's not built the way someone who actually knows LinkedIn Ads would build it. Blended CTR that includes Text Ads. No frequency check. Benchmarks that don't reflect B2B LinkedIn reality. The numbers are there, the story isn't.
The Winbox team — a LinkedIn partner agency — built this skill specifically to fix that. It pulls 30 days of live data through Databox MCP, benchmarks correctly, and produces a one-page client-ready PDF with talking points. The kind of report that usually requires someone who really knows the platform.
Free to download: databox.com/skills-marketpla…
Google Ads reporting doesn't have a data problem. It has a structuring problem.
Everything you need is already in the platform. Brand vs non-brand, CPA vs target, budget pacing, anomaly flags. Nobody has time to pull it into one place.
Built a skill that does it in under 90 seconds. Free.
Here is the link: databox.com/skills-marketpla…
