Daniel Kravtsov

Targeting is automated. Bidding is automated. Creative is the last lever that still moves the needle. We combined stellar, integrated cross-channel data with Creative Analytics. Improvado runs the full creative intelligence loop on clean data: 1 — Fatigue detection flags creatives before performance collapses. 2 — Archetype extraction pulls personas, narratives, emotional triggers, visual patterns from every ad. 3 — ICP-creative matching finds which arc works for which buyer. 4 — Creative scoring evaluates new variants against historical benchmarks before launch. 5 — Variant generation starts from creatives that actually worked. Drop 'creative' in the comments, and I'll DM you the details.

Tracking marketing performance is not a human job anymore. For 20 years, CMOs have had a ritual of going through dashboards. This ritual is a problem. Dashboards were meant to tame data, yet 76% of marketers are still overwhelmed by the volume they receive. Dashboards were always a workaround, a way for humans to process big data at the cost of tens of hours spent weekly on spotting insights and anomalies. AI agents spot exponentially more — continuously, with minimal human attention. They enable an entirely different relationship with data by running continuous monitoring and delivering insights and proposed single-prompt actions. The ritual is hard to break, as dashboards feel like a form of control. Yet, the companies pulling ahead rely on AI agents as analysts and operators. Marketers don’t need a better dashboard; they need a different operating mode. What was your biggest failure caused by delays in reviewing dashboards and acting upon insights?

AI allows you to massively increase the speed of experimentation in marketing. Without a deep business context, it’s running expensive mistakes at scale. If your agent doesn't know which products sell best, which audience segments repurchase, and which customer profiles return their orders, you’re not scaling experiments, but scaling failure. When your agent is backed by business context, speed and intelligence compound. That's what gets you from 1 experiment each week to running 5 daily. Improvado AI Agent comes with a knowledge graph that integrates all the business context that matters. To do so, it integrates data from 700+ data sources spanning beyond performance data — your landing pages, style, copy, documentation, reviews, and more. In a single prompt, you can run experiments backed by more business knowledge than you can get from 20 hours of cross-team meetings. Improvado MCP accounts for deep business context in every interaction. Drop ‘mcp’ in the comments, and I’ll dm you a guide with use cases it unlocks.

We're living through a phase shift, entering the AI future. What are the signals of that future? One of them, I believe, is the death of backlog. With AI Agents, executing tasks is easier than adding them to your backlog. The cost of execution collapsed, and even though not every task can be executed straight off, we’re getting there. This is a mindset shift. The agents are ready, but for some reason, it's emotionally hard for us humans to execute immediately. Make this shift. Kill the backlog. Execute on creation. What's one task in your queue an AI agent could ship before you close this tab?

We’re living inside the “singularity loop.” Improvado MCP runs marketing for Improvado MCP. Wait, what? A single AI Agent runs the campaign: ideation, creative, setup, and optimization. Our prompts go like this: "With Improvado MCP, create a landing page and run a Meta Ads campaign about Improvado MCP." It works because the agent sits on a knowledge graph that includes product specs, style guides, value props, ICP definitions, and insights from past experiments. One prompt pulls from all the needed context. The recursive part is the real unlock: the product improves marketing, which improves the product, which improves marketing. Feel the AGI.

We built the agent agencies actually need: lp.improvado.io/openclaw-agenc…

Improvado AI Agent = OpenClaw for marketing agencies. I keep hearing stories of people trying OpenClaw for agency business and losing access to ad accounts or burning their budgets. Real money, gone. OpenClaw is a great personal agent. It's not built for companies managing other people's money. We've been building AI agents for marketing at Improvado for three years. Our current version is what OpenClaw would be if it were built for agencies from day one. Same three things that make people excited, built for business: - Out-of-the-box integrations. OpenClaw connects to Slack and your browser. Improvado connects to 700+ marketing-specific platforms. Every source your clients use, already built. - Heartbeat. OpenClaw runs on schedule, so does Improvado AI Agent. Downloads data, analyzes weekly reports across your portfolio, generates A/B tests, flags pacing issues at 2 AM and pauses the campaign before anyone wakes up. - Dangerously skip permissions. When skipping permission checks, OpenClaw chains tasks in a powerful, yet insecure way. Improvado's agent is even more powerful: it manages campaigns, creates dashboards in three minutes, generates creatives, analyzes video narrative, runs full reporting. And, it’s more secure, as credentials live in a vault it can't reach. SOC 2, GDPR, HIPAA. Personal assistant vs. operations system. (link in comments)

@DVCorg i need this!

🔗 DataChain open-source release 🤖 AI-Driven Data Curation: Local models, LLM APIs 🚀 GenAI Dataset scale: Millions and billions of files 🐍 Python-friendly: Python objects instead of JSON Try it out github.com/iterative/data… 👇1/7

we are hiring VP of Marketing jobs.lever.co/improvadoio/55…

Exploring gender dynamics in Game of Thrones with Improvado AI: a shift from patriarchy as women take control of castles by the finale. Thoughts? 🐉👑 P.S. We're hiring a VP of Marketing! JD in first comment. #GameOfThrones #DataAnalysis #ImprovadoAI

Recorded a Loom with a practical example— a business question from a board meeting. Answered instantly, it moved us forward. AI enhances decision-making, empowering leaders to tackle questions efficiently and close the loop on pending decisions. #AI #DecisionMaking #Leadership

AI saves resources, lowers transaction costs, and reduces risk. We connected our AI Assistant to Improvado data, giving x-ray vision in decision-making. I handle most questions independently, even #2 #3, chatting with our data. Closing loop on unanswered questions brings clarity.

#1 offers simple answers. #2 leads to more questions. #3 spirals into endless queries, stalling decisions. Key choices often end up in #3, forcing leaders to either surrender, guess, or invest heavily in data-driven decisions. #AI #Leadership #DecisionMaking

AI helps streamline leadership decision-making by cutting transaction costs. It categorizes queries into three: quick and cheap, time-consuming and costly, and those too complex or unfeasible. Saves time and boosts efficiency. #AI #Leadership #DecisionMaking

@1eo @tana_inc How to see a recording of it?

Tana AI for Builders is disrupting the TfT space 🗓️ Join this Thursday's Tana JAM to learn commands & prompt engineering in @tana_inc . → Register to get a Zoom link & calendar invite → Bring your ideas & questions on the topic to share See U🚀 lu.ma/ptnpqvgo

@laurenbalik Every one who will invest in ML will die.

Most of the "correlation does not equal causation" people on either side do not understand 🌈contribution margin.🌈 If you're investing in ML and don't have 🌈contribution margin🌈 to order, unit, customer then you are all hat and no cattle with data.

@ImmanuelTrummer Hi Immanuel, this is very interesting; we are doing similar things with text2SQL at Improvado. Would you be interested in a quick call to share ideas?

#CodexDB demo: I use CodexDB to generate Python code for SQL processing while customizing code via natural language instructions! #GPT3 #OpenAI #NLP #AI #ML #DB #SQL #Python #DataScience #FoundationModels #LanguageModels

@tombielecki How?

I figured out how to get hierarchical summarization with #GPT 💪

@josephmoon_ai @hyperquery Make sense. But there is a question , why build new tool instead of improve existing ones? Like extension for VS code or google sheet / google docs addon? Easier adoption, less MVP. I understand there is no right answer curious how do you think.

Check us out at @hyperquery and DM me if you want to learn more!

The most important tool in data analytics is the #document. Yes, the document. It's where all the thoughts and insights are written down, organized, and shared with everyone in an organization. There are #dashboards, but those are good for communicating repeating insights.

@josephmoon_ai @hyperquery Make sense, it depends on level of abstraction here. But because almost nobody yet combine spreadsheet and doc nicely, I think it make sense to talk about them separately. Coda and quip did a decent job, but still even coda and quip users use google sheets a lot.

@DanielKravtsov @hyperquery Good point. I see spreadsheets as secondary primitive to the document. Important, but the core is the document as the system of record (because text is the most important type of information).

Data Twitter - we need a system of record for ad-hoc analytics, the "missing half of BI". What do I mean by that? Let's dive in 👇🧵: