Experiment · unwindGTM × Bitscale

What happens when your
AI BDR agent gets real data?

I built the AI BDR agent using free tools — Google News RSS, homepage scraping, Claude. It works. But the data ceiling was obvious within a week. This is what I learned when we plugged in an enterprise enrichment layer.

▶ Run the experiment yourself Read the full writeup →
Interactive demo No signup needed Free tools vs enterprise data — side by side
A collab between
unwindGTM × Bitscale

unwindGTM built an AI BDR agent on free open-source tools. Bitscale is an enterprise GTM data layer. This page is the honest account of what happened when we combined them.

The demo — what you're actually looking at

The same 4-step agent.
Two very different data layers.

The AI BDR agent runs the same workflow either way — ICP setup, research, outreach generation, objection handling. The toggle in Step 2 shows you what changes when the data underneath improves.

1 Setup
Define your ICP

Tell the agent what you sell, who you're targeting, and your differentiator. This stays the same regardless of data layer.

Same in both modes
3 Outreach
Claude writes from the data it has

Claude personalises using whatever signals Step 2 produced. Better input → better copy. The difference in output quality is visible.

Claude Haiku
4 Objections
Context-aware responses

Paste any objection. The agent diagnoses it, applies the right sales playbook, and gives you a response — including what not to say.

Playbook-driven
Under the hood — the open-source research stack

What the free version of this agent actually uses

This isn't a knock on open-source tools — it's an honest account of what they give you and where the ceiling is. Every vibe coder building outbound AI hits this wall eventually.

📰
Google News RSS

Free, public, no API key needed. We query Google News for the company name and pull the last 5 headlines. Good for catching funding rounds or product launches that made the press. Misses everything that didn't.

Coverage: public news only · ~5 items · no real-time signals
🌐
Homepage meta scrape

We fetch the company website and pull the meta description tag. That's it — what the company chose to put in their SEO description. Useful context. But it's their marketing copy, not intelligence.

Coverage: 1 sentence · what the company says about itself · no signals
🤖
Claude synthesises from what it has

Claude Haiku takes the news and the homepage blurb and infers signals — hiring, product momentum, expansion. They're often directionally right. But they're inferences, not detections. The confidence score reflects this.

Confidence: 40–75 · signals inferred, not detected · no email found
💧
With enterprise enrichment (Bitscale)

18-provider waterfall for verified emails. Live signals actually pulled from job boards, LinkedIn, SEC filings, and CRM migration forums. ICP scoring. The same Claude prompt produces noticeably sharper output.

Coverage: verified email · real signals · <3 sec per prospect
Free tools vs. enterprise enrichment

What the toggle actually shows you

Both sides use the same Claude prompts and the same 4-step agent. The only variable is the data going in. Here's an honest breakdown of what each mode gives you.

✕ Free open-source stack
📰
Google News RSSUp to 5 public headlines. Great if they've been in the news. Silent if they haven't.
🌐
Homepage meta scrapeOne sentence from their own website. Useful, but it's their marketing copy — not intelligence.
🤖
Claude infers signalsClaude makes educated guesses from limited context. Confidence scores: 40–65. Signals are inferred, not detected.
📭
No email foundNo email discovery in the open-source stack. Rep still needs to look it up manually.
📝
Decent but generic copyClaude writes well from what it has. But "from what it has" isn't much. Output is solid, not sharp.
✦ With enterprise enrichment (Bitscale)
💧
18-provider email waterfallZoomInfo → Apollo → Lusha → 15 others. Cascades until verified. <3 sec per contact.
📡
Real signals, actually detectedHiring from job boards. Funding from SEC/Crunchbase. CRM migrations from intent data. Not inferred — pulled.
🎯
ICP scoringFit scored against your ICP criteria. Lets you prioritise the 5 out of 10 worth calling first.
✍️
Signal-anchored copy"Noticed you're hiring 3 SDRs — the timing feels relevant…" That hook only exists because the signal is real.
10 prospects in 90 secondsEnrichment runs in parallel, not sequentially. The scale difference becomes real at 100+ prospects/day.
18+
Data providers
in the waterfall
90%+
Email coverage
across Indian & SEA SaaS
<2 min
Full workflow
Setup → enrichment → outreach
3
AI context layers
Company · Role · LinkedIn hook
Try it yourself

See both modes side by side

The demo is fully interactive. Run the free-tools mode first, then toggle to enrichment and watch the same 10 prospects get re-enriched — row by row. The output quality difference in Step 3 is what makes it interesting.

▶ Open the demo

No signup. No install. 4 steps, under 2 minutes.

bitscale-demo.html · Step 2 — Research
Setup
2
Research
3
Outreach
4
Objections
PS
Priya Sharma
VP Sales · LeadSquared · priya.sharma@leadsquared.com
📋 Hiring 3 SDRs ICP 4/5 ✓ Email verified
RM
Rahul Menon
Head of RevOps · Chargebee · rahul.menon@chargebee.com
💰 Series B closed ICP 5/5 ✓ Email verified
Powered by Bitscale ↗

Run the experiment yourself

Toggle between free tools and enterprise enrichment. See what changes in the outreach Claude writes. Form your own opinion.

▶ Open the demo Read the founder writeup →

Built by unwindGTM · Enrichment layer by Bitscale · No signup required