Building a GTM engine for an ultra-narrow market

We designed a full signal-based GTM system from scratch, combining market intelligence, automated enrichment, and multi-channel outreach.

Market mapping & segmentation
We mapped the global TAM : every company with ML engineers, NLP specialists, MLOps professionals, or data scientists.
80,000 companies narrowed down to 25,000 qualified, handled via a dedicated ABM playbook, tiered targets across three segments:

— Tier 1 : early-stage startups ($2-5M ARR) with small ML teams in fault-tolerant markets like media, sports, and education
— Tier 2 : production-ready companies ($2-20M ARR) in manufacturing, retail, and R&D running inference and training workloads
— Tier 3 : scale-ups and enterprises ($20M+ ARR) with 10+ person ML teams in regulated industries : healthcare, financial services, defense.

Always-on acquisition engine
We set up a daily webhook via TheirStack.com capturing around 300 AI and infrastructure job postings worldwide.
Each posting triggered an automated qualification workflow, filtering by industry, validating AI product fit, analyzing tech stack,
feeding directly into personalized outreach sequences.
By cross-referencing multiple data sources through Clay and Surfe, we achieved 80% lead coverage.
Campaigns launched initially on the US market, then scaled globally.

GitHub signal tracking
We built an automated system to monitor stars and forks across key ML repositories, HuggingFace Transformers, PEFT, Axolotl, Unsloth, QLoRA, DeepSpeed, Lightning AI. Engineers engaging with compute-intensive training tools were automatically enriched and added to targeted campaigns, with personalized openers referencing their exact repo activity.
Engineers forking PEFT or starring DeepSpeed are actively working on training workflows, that's not a lead, that's a buying signal.

Hugging face model tracking
We scraped Hugging Face directly to identify organizations consistently publishing new models.
Despite a 3,000-row export limit, we ran frequent rapid scraping cycles to capture active organizations.
Companies consistently deploying models signal both technical maturity and ongoing compute needs, exactly our target profile.

LinkedIn social listening
We tracked engineers publicly expressing frustration with hyperscalers, GPU quotas, or competitor limitations lie RunPod, Together AI, Groq
and built campaigns using their exact pain points as entry points. Engineering colleagues flagged relevant posts in real time, and browser extensions allowed rapid extraction of everyone who engaged with them.

Nurturing & re-targeting
We pulled every lead from HubSpot who had previously responded to campaigns run by predecessors and launched multi-touch re-engagement sequences, typically four emails combined with LinkedIn messaging, acknowledging past interest and repositioning FlexAI's evolved product.

Event-based outreach
For every event attended at TechCrunch, Web Summit, AI Collective, Adopt AI, we scraped complete attendee and speaker lists, enriched them through Clay, and launched event-specific sequences before, during, and after each event.
At Web Summit in Lisbon, the team was on-site at the booth, converting online outreach into face-to-face conversations.

Lookalike & expansion
We launched lookalike campaigns based on existing customers via Ocean.io, with industry-specific blueprints, and integrated 11x.ai to scale outreach capacity with AI SDRs, augmenting manual campaign volume without adding headcount.