Every week there’s a new AI tool promising to “revolutionize your workflow.” Most of my job, honestly, is telling clients what not to buy. Here’s the stack I’ve converged on after years of implementations — the tools that survive contact with real businesses — and the categories that consistently disappoint.
The foundation: one source of truth
Everything starts with a CRM that every lead, conversation, and deal flows through. For most SMBs and clinics: GoHighLevel, HubSpot Starter, or Pipedrive. The specific choice matters less than the rule: if it’s not in the CRM, it doesn’t exist. No leads living in email inboxes, WhatsApp threads, or someone’s memory.
The connective tissue: workflow automation
The second layer connects your tools and executes the repetitive work. My default is Make (or n8n for clients who want self-hosted control and lower running costs). Zapier is fine for simple cases but gets expensive and limiting at scale.
What this layer actually does in production:
- New inquiry → create CRM contact → send instant SMS/WhatsApp → notify the right person
- Appointment booked → confirmation + reminder sequence → post-visit follow-up
- Deal stalled 7 days → reactivation task → manager alert at 14 days
- Weekly: pull ad spend, leads, bookings, revenue into one dashboard
None of this requires AI. It requires honest process mapping and a few afternoons of building. This layer alone usually produces the majority of the measurable ROI.
The intelligence layer: where AI earns its keep
With the foundation working, AI adds leverage in specific, bounded places:
- Conversation drafting. LLMs draft replies to inquiries within approved templates and tone guidelines — especially valuable for after-hours coverage and multi-language businesses.
- Call intelligence. Transcription + summarization of sales calls (tools like Fireflies, or Whisper + GPT pipelines) turns every conversation into searchable CRM data: objections, commitments, next steps.
- Lead qualification. Scoring inbound inquiries on fit and intent so the team calls the hot ones first.
- Content operations. First drafts of review responses, social posts, and follow-up sequences — always human-reviewed before anything goes out.
What to skip (for now)
- All-in-one “AI employee” platforms. The demos are dazzling; the production reliability isn’t there for client-facing work. They hallucinate details, go off-script on pricing, and create liability. Revisit in a year.
- Custom model training. Almost no SMB needs fine-tuned models. Well-prompted frontier models with your business context outperform expensive custom training for these use cases.
- AI chatbots as your front door. A bot that can’t book, can’t answer pricing, and can’t escalate is a lead-killer. Simple forms + instant human-routed follow-up outperform mediocre chatbots every time.
- Tool sprawl. If your team needs six logins to follow up with one lead, you’ve automated yourself into a worse process. Fewer tools, deeper integration.
The build order that never fails
CRM → automation → AI. In that order, each layer pays for the next. Skip a layer and you’ll join the long list of companies with impressive AI demos and unchanged revenue.
The boring version of this stack, properly implemented, is worth more than any revolutionary tool launched this year.
Add comment