What I'm Building
I'm not theorizing about operational strategy. Right now, I'm orchestrating a SaaS transformation, advising an AI platform, and building community. Here's the log.
Multi-tenant brand voice
Same content, 10+ distinct brand voices. Each brokerage gets voice-tuned versions without manual rewriting. The trick: MEALL tuning parameters work per content type. Market updates need different voice calibration than advice posts. One system, many outputs.
Voice calibration UX
People can't describe their voice, but they can recognize it. So we stopped asking them to define it and started showing options. Three variations—warmth, authority, brevity—and you pick what sounds like you. Way better than a blank text field asking 'describe your brand voice.'
Brand voice extraction shipped
System analyzes writing samples and extracts a reusable voice definition. Feed it 3-5 emails, get back a voice profile that actually sounds like the person. Three variations—warmth, authority, brevity—let users tune without starting over. Switched from Sonnet to Haiku for 3x speed, no quality loss.
SLA architecture for enterprise
Built tiered response time commitments for a major channel partner. Three layers: technology SLAs (uptime, performance), customer service SLAs (call answer, resolution), mortgage-specific SLAs (processing, disclosure timing). Vendor compliance data feeds from Zendesk, GHL, Twilio into Domo. Enterprise partners want the receipts.
The memoryless therapist
Built an NVC-based chatbot to see what happens when people talk to something that doesn't remember or judge. Turns out they open up immediately. No memory isn't a bug—it's the feature. Like journaling in invisible ink. Testing whether this unlocks better signal in recruiting conversations.
Pricing by funnel position
Different customers use the product at totally different points in their process. Boutique agencies run final slates—5-15 candidates, high margin. SMB screens top-of-funnel—50-100+ candidates, lower margin. So we built a matrix: more reqs when runs are low, more runs when reqs are low. Enterprise self-selects based on how they actually work.
M&A integration as a wedge
PE firms have a 60-90 day window after close where employees are still open to change. Miss it and they either leave or dig in. We're testing whether values alignment predicts who stays better than the standard HR playbook. Side benefit: integration analysis surfaces recruiting gaps—that's the flywheel back to Ikigai's core product.
Co-branded distribution economics
In fragmented markets, co-branded distribution beats direct sales CAC by roughly half. Partner brings the audience, we bring the infrastructure. Revenue share aligns incentives. Nobody builds their own mortgage operation when they can bolt onto one that already works.
M.E.A.L.L. framework in production
Five content pillars for agent marketing: Market, Events, Advice, Local, Listings. Each has its own tuning parameters so market updates sound different than advice posts. First cohort generating 30 days of content in under an hour. The framework forces consistency without feeling templated.
Local market intelligence at scale
Built a workflow that researches 54 markets monthly for $4.50 total. Perplexity does the research, Claude checks quality, sparse results trigger automatic follow-ups. Agents don't care what the Fed did—they care about Plymouth Township. Hyperlocal beats generic every time.
The Strategy
Transforming a niche mortgage SaaS (Avenu) into a multi-product ecosystem.
Active Projects
Building something similar?
Compare Notes