Jun 29, 2025
Two years ago, I started Blossomer with the simple belief that startups should attack their go-to-market with the same experimental rigor they bring to product development.
I also believed AI would play a critical role in this (beyond just writing bad emails). There was an opportunity to help teams develop their positioning, research their target customers, and execute campaigns faster and more effectively.
So I embedded myself within each startup, executing their workflows manually to learn what actually worked. This taught me that while every startup's challenges are unique, there are patterns in how successful GTM experiments unfold.
After two years of refining this approach with B2B startups, we've evolved from using SOPs, chat apps, and n8n automations to building an AI-native system.
Today, we're releasing a lightweight CLI tool that showcases our methodology alongside GTM automations—it researches accounts, qualifies leads, and executes campaigns using the same framework we use with early-stage founders.
Watch the 4-minute demo above or try it for free!
My profit-maximizing wish is that once you experience what we CAN do, you'll want the full system personalized for your startup. 😀
For those of you who want to see how we got here, please read on!
The Problem: Death by a Thousand Copy-Pastes
When we first applied AI to GTM, we did what everyone does: threw ChatGPT at everything.
We crawled towards a slightly more sophisticated AI-powered workflow consisteing of carefully crafted prompts, Clay tables, and n8n automations.
When a company worked with us, we'd fork our templates and impress them with our speed in producing decent strategies, email campaigns, and analysis.
But I quickly realized we'd been measuring success by the wrong metrics.
We were producing deliverables at scale but the real value actually came from our in-person working sessions.
"Campaign X outperformed Campaign Z by 3x—let's understand why."
“Your messaging about use case A resonates with enterprise but not mid-market.”
"Based on these patterns, here's what we should test next."
The breakthroughs happened when we analyzed results, captured insights from founders, and pattern-matched across our portfolio.
We spent 90% of our time on production and documentation, leaving almost no time to systematically improve our own methodology.
The irony wasn't lost on me: we were trapped in the same exact manual execution hell we were supposed to solve for founders.
The Path Forward: Building Something Better
That's when we realized we needed to flip our approach.
Instead of sprinkling AI over a manual processes, we started building agents to handle our critical workflows.
Instead of sprinkling AI over manual processes, we built an agent-based system that could actually respond to real-world feedback and evolve with each engagement.
Today, we're packaging those core modules into this lightweight CLI tool so you can experience our methodology firsthand before trusting us to build your custom system.
How it Works
The system runs a 5-step workflow that mimics the in-person conversations we have with founders.
There are three core design principles that ensure agents maintain quality, abide by our best practices, and produce strategies that feel handcrafted rather than templated.
Context quality gates between steps prevent cascading failures—if Step 1 produces poor company analysis, the system stops rather than generating garbage through all 5 steps.
Dual storage formats (JSON for machines, markdown for humans) enable both programmatic intelligence and manual editing. You can export results to your CRM or edit the markdown files directly.
Shared company context that agents can update and reference. By Step 5, you get a strategic plan that synthesizes insights from your entire analysis journey, not just the final prompt.
Step 1: Company Analysis
Just like our actual engagements, the tool starts by deeply understanding your business.
Users input their domain and our research agent uses Firecrawl to parse multiple pages, specifically hunting for details around their business model, the customers they serve, and how they position themselves in their category.
Details around pricing, customer case studies, and even the language they use will inform how we develop their Ideal Customer Profile and their go-to-market strategy.
Since this foundational understanding shapes everything that follows, we’ll verify accuracy with the founders themselves before proceeding—just like we would in our actual proces
Step 2: Target Account Profile
Next, we develop your Ideal Customer Profile by formulating a hypothesis around which account attributes indicate a growing need for your product.
For a document processing startup, that might mean financial advisory firms emphasizing "holistic planning" (which requires extensive documentation) with 50-200 clients (large enough for process pain but small enough to lack enterprise solutions).
You get detailed ICP profiles based on firmographics and behavioral patterns, setting up the foundation for more sophisticated scoring once you begin tracking real engagement and buying signals.
Step 3: Buyer Personas
While the Target Account Profile answers "Which companies likely have this problem?", the Contact Readiness Score answers "Who inside that company is most motivated to solve it right now?"
The system identifies your champion buyers—not just by title, but by indications and attributes that they're personally invested in the problem.
Taking our financial services example, your champions might be founders with CFP certifications or VPs who came up through operations—people whose backgrounds suggest hands-on experience with document processing pain.
This context-aware approach means you're reaching out to people whose career path and day-to-day role make them natural champions for your solution.
Step 4: Email Campaigns
We believe a great outbound system can systematically uncover what language resonates with which sorts of customers.
The system generates multi-step campaigns that test different combinations of pain points, use cases, and value props using our “Lego Block” methodology.
Every line in the email is modular and swappable so you can quickly test different messaging pillars to see what drives responses.
Each campaign is tailored to the persona's relationship with the pain point you’re solving.
You may want to lead with ROI and growth metrics for an executive while focusing on workflow efficiency for individual contributors/operators.
The goal is to generate campaigns and a scaffolding for experimentation (see our next post about Multi-armed Bandits to see how these experiments get executed in practice)
In just a few seconds, you get a systematic framework to discovering which messages resonate with which personas.
In our next post, I'll show how our full system uses Multi-armed Bandits to optimize these experiments in real-time.
Step 5: Strategic GTM Plan
Finally, our advisor agent synthesizes everything into an actionable GTM playbook using our best practices.
The system pulls insights from all previous steps to create a comprehensive plan: how to qualify and prioritize accounts, which enrichment data sources and APIs to integrate into your GTM stack, and what metrics to track for rapid iteration.
It's like having an experienced GTM advisor who's analyzed your specific situation tell you: "Start with these account attributes, reach out to these personas first, test these value prop combinations, and here's exactly how to measure what's working."
For this critical last step, we recommend setting the model select to use advanced models (we really like Claude Opus 3, Claude Opus 4, or Gemini 2.5 Pro) for their superior reasoning.
While previous steps benefit from faster, cost-effective models for data gathering, this final strategic synthesis is where model intelligence and larger context windows pay off.
Try it Yourself
We encourage you to download this repo and try this tool out for yourself.
The CLI tool is completely free to use and you can even go into the code and tweak the prompts and models.
What's Next: The Full Learning System
As I mentioned before, this lightweight tool demonstrates our core methodology, but it's just the foundation.
We had bigger goals in mind and after months of late nights and dusting off my software engineer chops, we finally built a system that:
Learns from every campaign's performance data
Applies successful patterns from one startup to improve results for others
Spots opportunities I'd miss doing analysis manually
Evolves its strategies based on real-world feedback
The engineering journey was brutal. We had to solve agent orchestration at scale, prevent model drift while email best practices kept evolving, and learn how to efficiently debug multi-agent systems.
But the results have been worth it.
We finally have something that can support the unique GTM challenges of our founders while scaling our impact.
If you try the CLI tool and like what you see, I'd love to show you what we can deliver. Book a demo here or just email phil@blossomer.io if you have questions.
And if you're building with AI, stay tuned—I'll be sharing a technical deep dive on the engineering challenges and solutions in my next post.
