The Situation
Open source AI has a conversion problem.
Thousands of developers might star your repo, fork your code, or spin up your model on Hugging Face. But the path from "ML engineer playing with your API" to "enterprise procurement signing a six-figure contract" is murky at best.
I joined a visual generative AI company with strong open source traction—downloads were healthy, the GitHub community was engaged, and ML practitioners genuinely loved the product. But the enterprise pipeline was anemic. Marketing was running the same playbook they'd use for any B2B SaaS: gated whitepapers, webinar registrations, demo request forms.
The problem? Developers don't fill out forms. They evaluate tools by reading docs, cloning repos, and running code. By the time they're willing to talk to sales, they've either already decided you're the answer—or moved on.
We needed to see the buying process that was happening in the dark.
The Insight
Enterprise AI purchases don't start with a Google search for "visual AI vendor." They start with an engineer solving a problem.
That engineer finds your model on Hugging Face. Downloads it. Runs inference. Maybe files an issue on GitHub asking about fine-tuning. Shares it in their team's Slack channel. A week later, their manager asks "what are our options for scaling this?"
The enterprise buying journey had already started—we just couldn't see it because none of those signals existed in our CRM.
The second insight was about who matters. In developer-led purchases, the technical evaluator and the economic buyer are different people with different information needs at different times.
We needed a system that could:
- Detect developer engagement signals where they actually happen
- Connect individual developers to their organizations
- Score accounts on both technical engagement and commercial readiness
- Trigger the right message to the right person at the right time
The System
Layer 1: Discovery Signal Capture
Most marketing teams treat their website as the top of the funnel. For developer products, your website is the middle of the funnel. The top is wherever developers discover and evaluate tools.
We built integrations to capture signals from:
GitHub Activity
- Stars and forks on our repos (awareness signal)
- Issues filed and PRs submitted (active evaluation)
- Questions in discussions (buying signal—they're trying to make it work)
Hugging Face Engagement
- Model downloads (basic interest)
- Space usage—when someone deploys our model in a Hugging Face Space, they're past curiosity into prototyping
- Model card views vs. actual downloads (tire-kickers vs. evaluators)
Product Signals
- API trial signups
- Free credit usage patterns—someone burning through credits on realistic workloads is testing for production
- Specific endpoints called (certain API patterns indicate enterprise use cases)
Layer 2: Identity Resolution and Enrichment
A GitHub username isn't a lead. We needed to connect developer activity to companies.
Developer Signal (GitHub user: ml_engineer_jane)
↓
Email domain extraction (from Git commits, if public)
↓
Company identification (Clearbit, Apollo, or manual)
↓
Firmographic enrichment:
- Company size
- Industry vertical
- Tech stack (AWS? Azure? PyTorch shop?)
- Existing AI/ML investment signals
↓
Account matching in Salesforce
Layer 3: Dual-Track Scoring
Traditional lead scoring fails for PLG because it treats all engagement equally. We built parallel scoring tracks:
Technical Engagement Score (Developer Track)
- Points for: repo activity, API usage depth, documentation engagement, community participation
- High score = strong technical validation happening
Commercial Readiness Score (Buyer Track)
- Points for: pricing page views, security documentation requests, enterprise feature inquiries
- High score = procurement process starting
The magic was in the combination:
| Technical | Commercial | Action |
|---|---|---|
| High | Low | Arm the champion: case studies, ROI tools, "how to pitch internally" content |
| Low | High | Technical proof: sandbox environment, solution architecture call, pilot scoping |
| High | High | Enterprise sales engagement |
| Low | Low | Nurture: educational content, stay visible |
Layer 4: Lifecycle Orchestration
Signals flowed into Salesforce as lead and account activities. Marketo handled lifecycle automation based on score thresholds and trigger events.
Champion Enablement Track
Trigger: Developer at enterprise company with 3+ engagement events, no commercial signals
Content: Case study → ROI calculator → "Questions your CFO will ask" guide → Customer advisory call offer
Layer 5: Sales Handoff with Context
When an account hit threshold for sales engagement, the rep didn't just get a name. They got:
- Which developers at the company were active, and what they'd done
- Which use case the account was likely evaluating (based on API patterns)
- Competitive intel (had they also evaluated alternatives?)
- Recommended talk track based on vertical and technical profile
- Champion contact (the developer most engaged) for potential intro
The Takeaway
Developer-led growth isn't about getting developers to fill out MQL forms. It's about building visibility into the evaluation process that's already happening—and orchestrating the right interventions at the right moments.