I spent 2 weeks writing about why agency dashboards don't work. This week I started building the thing that does.
Week 3 of building in public. Status update: I am no longer "thinking about it." I am building it.
The tool is called Pulseboard. The thesis is unchanged from my last two posts here: the product service-business owners actually buy is "a confident recommendation for the next 7 days," not "visibility into the last 30." Almost every analytics SaaS in this category sells the wrong shape of value.
Pulseboard is the thing I would have wanted as an agency owner four years ago.
What it does (v0.1, running internally this week):
- Ingests the same data sources my agency already runs (GA4, GSC, Facebook organic, Instagram, Google Business, LinkedIn organic)
- Runs the 14-day window through a decision model I've spent two months writing
- Outputs a ranked set of recommendations, each with a confidence score, a "why," and the underlying data points
The hard part isn't ingest. Ingest is solved. The hard part is the decision model.
A decision model that tells an agency owner "double down on the lead-speed hook" needs to know that lead-speed hooks are content patterns, that the same hook appears across Reels and carousels and engagement posts, that the same topic at different reach levels means different things, and that telling someone to "post more" without telling them which hook is useless.
Most "AI insights" tools skip this step. They run an LLM over a chart. The output is grammatically a recommendation but semantically a description.
What I'm trying to ship is a tool that takes a position. With a confidence score, because positions taken without confidence scores are just opinions in suits.
Three things on my mind this week:
-
How aggressive should the model be when the data is thin? Right now it abstains when sample size is below a threshold. Beta testers want it to "still say something." I'm not sure the right answer here.
-
The dashboard-vs-recommendation tension. I shipped with zero charts. Beta testers pushed back. I now include tiny inline trend charts under each recommendation. (Wrote about this on LinkedIn today.)
-
Pricing. Service businesses pay per-seat for tools all the time, but the value here is collective ("the agency"). Per-agency pricing feels right. Per-client-managed pricing feels predatory. Per-recommendation-acted-on would be amazing if I could measure it, which I can't.
Open question for anyone here who has shipped a vertical SaaS that replaced an existing category: how did you handle the "but I want my old dashboard too" pushback in the first 90 days?
Waitlist is at getpulseboard.ai for anyone who wants to be in the first cohort.
On the “but I want my old dashboard too” pushback in the first 90 days: treat charts as receipts under a recommendation, not the product. Recommendation stays primary; the tiny trend line only exists so the client (or AM) trusts the call enough to act.
On thin data: abstaining with an explicit “not enough signal” line usually beats inventing confidence. Beta testers who want you to “still say something” are asking for comfort copy — that erodes the judgment thesis you’re selling.
Per-agency pricing feels closer to how service businesses buy ops tools than per-client-managed, which can feel like a tax on growth.
Curious whether your abstain threshold is based on sample size alone, or also on how expensive a wrong recommendation would be for that client type.
Some copy drafted with AI tools and reviewed by Isom Industries, LLC before publishing.
This direction is much stronger than another agency dashboard. The sharp part is that you’re not selling visibility, you’re selling judgment. A ranked recommendation with confidence, why it matters, and the supporting data is closer to an operating system for agency decisions than a reporting layer.
On the “old dashboard” pushback, I’d probably keep the charts as evidence, not the product. The recommendation should stay primary, and the chart should only exist to make the recommendation feel trustworthy enough to act on.
One naming thing I’d watch early: Pulseboard still sounds partly tied to the dashboard category you’re trying to escape. If the product becomes the decision layer for agencies, a cleaner SaaS name like Beryxa.com may age better than something that keeps “board” in the frame.
Selling judgment instead of visibility is a good distinction. The recommendation staying primary with charts as evidence, not the product, is exactly the direction we moved in after user testing. Appreciate the framing.