Omnichannel Conversations & AI Support
Product Manager · HighLevel · 2023–2024
A conversations module for a small-business CRM, sequenced deliberately from rules-based to predictive.
The Question
If a small business owner is juggling messages across several separate inboxes, what does that actually cost them, and can software absorb it?
Context
HighLevel is a CRM for small and medium businesses. Conversations were scattered across channels, and support was only available when someone was watching an inbox.
The Constraint
Every channel (SMS, email, chat) had its own inbox. A small business owner had to check all of them just to know if a customer was waiting.
My Hypothesis
I initially thought the fix was better notifications across the existing inboxes. It wasn't. The inboxes themselves were the problem, not how loudly they alerted you.
What I Tried
Built a consumer-first omnichannel conversations module across mobile and web. Layered a GenAI-driven chatbot on top for 24/7 support. Ran an 80-SMB beta test specifically to confirm a hypothesis (that users wanted less manual setup) before committing to build predictive lead scoring. Also shipped two-way Gmail sync once multiple inboxes surfaced as the clearest user pain point.
The Decision
Shipped rules-based lead scoring first (set rules, assign scores) rather than jumping straight to the more sophisticated predictive version, using the beta's result to decide the sequencing rather than assuming.
The Build
One omnichannel conversations module across mobile and web, a GenAI chatbot on top for after-hours coverage, and rules-based lead scoring to prioritize who to call first.
The Evidence
The conversations module now handles 463M+ messages a month. The GenAI chatbot lifted monthly active users 8% in four months, from 257k to 278k. Lead scoring lifted conversion rates by 20%. Gmail sync lifted engagement by 8%.
What Changed
Businesses stopped checking five inboxes to find one waiting customer. Monthly active users rose 8% in four months once that friction was gone.
What I Learned
Validating the simpler hypothesis before building the more sophisticated version isn't slower. The beta made the roadmap sequencing an easy call instead of a guess.