Alexander Kučin

"You Don’t Have a Data Problem. You Have a Decision Throughput Problem". Most mid-market companies already have plenty of data. What they don’t have is enough decision throughput. Ask yourself: - How many high-stakes decisions did your company make last quarter? - How many did you delay? - How many were made with partial confidence because time ran out? That gap is the real constraint. Growth doesn’t stall because dashboards are missing. It stalls because decisions arrive too late to matter. And when every meaningful decision queues behind the same analytics team, velocity collapses. Your growth rate is capped by how many high-quality decisions you can make per quarter. Not how accurate your forecasts are. Not how sophisticated your models are. Decision throughput is the bottleneck whether you measure it or not. #decisionintelligence #decisionframeworks #graphitenote

3

1 Comment

CFO: “Why doesn’t the dashboard match the UI?” Data Analyst: “I’ll investigate.” 3 months later… Still off. By 0.04%. Now every number needs explaining. What’s interesting is that in these cases, it’s rarely about perfect data. It’s about confidence in the numbers. 99.96% accurate can still feel wrong - if no one understands why there’s a gap. Instead of chasing a perfect 1:1 match: - Show the variance - Explain the definition differences - Agree on what “close enough” means - Quantify the delta Example: Revenue (Dashboard): €1,017,907 Revenue (UI): €1,017,500 Δ = 0.04% Reason: Different day cutoffs (UTC vs local time) Perfect alignment with UI logic is expensive, fragile, and usually unnecessary. The better question is: How accurate does this need to be to support the decision? Board reporting isn’t the same as a marketing experiment. And neither is the same as a daily ops dashboard. Different decisions require different precision. Clarity beats perfection.

71

9 Comments

Two weeks ago I wrote about a guy who named his company "AI Slop" backwards. At the time Polsia was doing $1.25 million in annual run rate. Three days later I posted an update. It had crossed $2 million. The founder went dark for a few days and just posted an update. He's hit $4.5M ARR. Solo founder, a team of AI agents, and zero employees. Let me put the timeline in perspective. Mid-December 2025: Polsia launches. Late February: $100K run rate. Three days later: $1M. Early March: $2M. Today: ~$4.5M with 3,774+ active companies on the platform. 147,000 tasks completed. 107,000 emails sent. Growth running at 30% week over week. Last week almost broke him. Infrastructure partners hitting rate limits everywhere. Every bug that could surface, surfaced. Investors throwing term sheets at him. Customers flooding every channel at once. He had to go dark. Just him and his AI agents, fixing things one by one. This is the moment that separates the "AI replaces everyone" fantasy from what is actually happening. Ben spent years managing global operations for Travis Kalanick at CloudKitchens. He knows exactly what it looks like to throw headcount at a scaling crisis. He did none of that. He sat down with AI agents that have narrow, well-defined authority and worked until the system stabilized. Then he came back online and the revenue curve kept climbing. When I wrote the first post, people in the comments said it would not last. That the model would break at scale. That one person cannot run thousands of companies. The number was $1.25 million then. It is $4.5 million now. The employee count has not changed. I am going to keep tracking this. Because whatever you think about Polsia as a product, the operating model is the story. One human plus AI agents plus the right architecture is outscaling teams of 50. That is no longer a future prediction. This is reality.

12

A significant part of "unarmed" AI is the old concepts that hinder the transition to the next level. Here we need a rethinking.

1

Fridays are feedback days. Here are the top pieces of feedback I received from customers of Flaunt’s AI Data Analyst last week: Klaviyo data analysis: This is one of the most important use cases. They want the data to tie out with key metrics in dashboards and they want to do deep analysis of event level data. We want to enable this without doing traditional transformations and pre-aggregations to scale. Gorgias data analysis: We are syncing custom field data so you can do deep analysis on “contact reason” for tickets. Like Klaviyo, these data sets are big and messy and amazing use cases for AI-assisted data analysis. Automatically recalling past context/queries: As a user, I want things snappy and I want to say “hey remember that analysis we did on reviews? Let’s refresh that.” And I want to say that without going through a lot of effort to save analyses and design a dashboard. More business context, better recommendations: Brand operators want to rely on, trust, understand, and use the recommendations that our analyst provides. But sometimes, like a new employee, more business context on how certain products are prioritized, how the website funnel works, the rules of their subscription and loyalty program works, etc can lead to much better recommendations. Speed: To go from early adopters to the middle majority, I think speed will be an area we need to chisel at over time. Some analyses take seconds, some take minutes. But the faster we get, the more useful the tool becomes. Because one question begets another. Landing page dashboard / proactiveness: Our landing experience is simple. But sometimes there’s friction when having to come up with “what should I be analyzing?” Enabling a rich way to continue past conversations, display metrics that we’ve learned are the users favorite from past conversations, and displaying anomalies in this metrics will all get there. Also- the first time user’s experience will need to be more considered over time. Finally, several integrations are getting added: * Google Analytics * Google Ads * Meta Ads * Yotpo * Stay AI * KnoCommerce * Digioh & others

7

4 Comments

Successful AI projects will require real leadership from executives, because these are Transformative projects….not Optimization projects. Process and Role transformation are required.

13

Generally, I like to post about Data Compression. But sorry. It's Friday. So here's a nice graph I stole from Thomas Vles today. Not sure if Thomas also stole the idea too. Possibly originated from a non-human? Next week, back to full data-compression mode & going to tackle: - What does AI-First actually mean - Who owns the travel customer in an age of AI (It's obvious, isn't it?)

61

12 Comments