This post was originally published on LinkedIn.

WHAT WE'RE BUILDING RIGHT NOW:

Full signal → action architecture for a client using Clay + 6sense Email AI + Actively AI + AI harvested signals. The interesting part is we're testing two approaches side by side:

1. deterministic signal based workflows (if X signal fires, do Y action, report on signal success) vs

2. AI-model prioritization (feed all signals/context into an AI model and let it decide what to prioritize and how to craft the sales sequence).

Both have their place but we're reporting on them side by side to see which drives the highest pipeline//conversions.

• More and more third party signal harvesting from unstructured data. AI agents to find buying signals that don't live in a database. Job postings. Earnings calls. Press releases. Negative reviews. Only AI can do this at scale. I believe the best buying signals do not live in databases.
• Trying to solve the context layer problem. Every AI tool needs ICP context, persona details, messaging frameworks. But if that context lives in a dozen different places you end up with context silos and misalignment. Looking at different tools and methods to build a central context repository that all tools can tap into via API.

LESSONS LEARNED:

• The bottleneck isn't always AI. Sometimes AI part is working great but there's human middleware between steps. For example: People copy/pasting AI copy from one tool and into another. That's not an AI problem. It's an automation problem. Old school problems still matter.
• MAPs can't do true 1:1 AI personalization (and that's probably fine). For example: you want to send 5,000 people each a unique AI-written email, you'd need 5,000 email assets in Marketo/Hubspot. Or crazy complex token architecture. Most clients doing real 1:1 AI personalization use sales engagement tools instead.

WHAT'S NEXT:

• Building signal architecture and reporting in Salesforce. Custom object for signal → meeting → pipeline → revenue reporting.  We need to know which signals actually work. I don't see enough people talking about how to report on signals.
• Continue working on v1 of our AI data analysis agent. Push pipeline data in, AI runs analysis, finds the insights, builds a summary report with actions. Hours of analysis ready in minutes.
• Continue to balance smaller AI pilots vs. systematically embedding AI deep into GTM architecture for our clients. Both have their place.

𝘛𝘩𝘪𝘴 𝘪𝘴 𝘢 𝘴𝘦𝘳𝘪𝘦𝘴 𝘐'𝘮 𝘵𝘦𝘴𝘵𝘪𝘯𝘨 𝘵𝘰 𝘴𝘩𝘰𝘸 𝘩𝘰𝘸 𝘸𝘦'𝘳𝘦 𝘦𝘷𝘰𝘭𝘷𝘪𝘯𝘨 𝘊𝘚2 (𝘢 𝘎𝘛𝘔 𝘰𝘱𝘴 𝘢𝘨𝘦𝘯𝘤𝘺 𝘧𝘰𝘳 𝘉2𝘉 𝘵𝘦𝘤𝘩) 𝘪𝘯𝘵𝘰 𝘵𝘩𝘦 𝘈𝘐 𝘦𝘳𝘢. 𝘛𝘳𝘺𝘪𝘯𝘨 𝘵𝘰 𝘧𝘰𝘤𝘶𝘴 𝘰𝘯 𝘱𝘳𝘢𝘤𝘵𝘪𝘤𝘢𝘭 𝘴𝘵𝘶𝘧𝘧 𝘷𝘴 𝘩𝘺𝘱𝘦 𝘢𝘯𝘥 𝘴𝘩𝘢𝘳𝘦 𝘸𝘩𝘢𝘵 𝘸𝘦'𝘳𝘦 𝘢𝘤𝘵𝘶𝘢𝘭𝘭𝘺 𝘣𝘶𝘪𝘭𝘥𝘪𝘯𝘨 𝘢𝘯𝘥 𝘭𝘦𝘢𝘳𝘯𝘪𝘯𝘨.