Prospects ask me what tools we use at Starfish almost every week. I'd rather publish the answer once than explain it 20 separate times on calls. So here it is. The actual stack. What's running, why it's in the rotation, where it falls short, and what we've tried and cut.
If you're vetting an AI consultant, this is the right question to ask them. What they use daily tells you more than any pitch deck. A consultant who can't name their own stack isn't operating their own advice.
GoHighLevel (GHL) is the backbone of our client operations. It handles the CRM, the pipeline, email marketing sequences, SMS campaigns, reputation management, and most of our internal automations. We've been on it long enough to know both its ceiling and its floor.
Why GHL over HubSpot or Salesforce? Price-to-feature ratio for an agency our size. We're not a 500-person enterprise. GHL gives us pipeline visibility, automated follow-up, and a unified client record without paying for a platform we'll use at 30% capacity.
Where it falls short: the native reporting is clunky. The workflow builder is powerful but verbose. If you want complex conditional logic or external API calls, you'll hit the ceiling fast. That's where the next layer comes in.
We use three tools here depending on the complexity and data sensitivity of the job.
Simple, linear triggers. One thing happens, another thing follows. We use Zapier when speed of setup matters more than logic depth. Form submission pushes a contact into GHL and fires a Slack notification. Done in 10 minutes. Reliable and obvious. Expensive if you let your task count run unchecked.
Multi-step, branching workflows with conditional paths. When a scenario needs to check a value, route differently based on the result, loop through an array, or call an API with a transformed payload, Make handles it cleanly. Workhorse for complex client build-outs.
When data sensitivity matters. For workflows touching sensitive client information, we run n8n on our own infrastructure. You own the environment, the logs, the credentials. See how this fits into our system integration approach.
We build custom AI agents that run on the OpenAI API as the primary endpoint, with Claude as the backup depending on the task type. The distinction matters: we use the API, not a ChatGPT Plus subscription, for anything in production.
At the volume we run document generation, lead qualification logic, and content drafts, API pricing is substantially cheaper than per-seat SaaS subscriptions. More importantly, API access lets us wire the model directly into a workflow. Output goes where it needs to go without a human copy-pasting it.
ChatGPT Plus is a research and drafting tool. The API is a production component. We treat them differently.
Claude handles nuanced document tasks where tone and context management matter more than raw speed. OpenAI handles structured data extraction, JSON outputs, and anything where the schema needs to be precise. We're not loyal to a single provider. We're loyal to the right tool for the job.
Scheduling runs through Calendly for external-facing bookings. For projects where we need more control over routing logic or want to avoid Calendly's branding, we use Cal.com. Calendly is faster to configure. Cal.com is more flexible if you need custom embed behavior or white-label output.
SMS goes through Twilio when GHL's native messaging isn't enough. Twilio comes in when we need programmatic control: dynamic message content generated by a workflow, high-volume outbound triggered by an API call, or a two-way conversation handler that needs to branch on reply content.
Email is split by type. Postmark handles transactional email: order confirmations, account notifications, form receipts. GHL handles marketing sequences, nurture campaigns, and broadcast sends. Mixing these in the same sending domain is how you tank your deliverability. Keep them separate.
Google Workspace runs the business: Docs for writing, Calendar for scheduling, Drive for file storage and client share folders. Nothing exotic. It works, it's cheap relative to what you get, and every client already has access.
Notion is our internal knowledge base. SOPs, client project specs, onboarding templates, agent documentation. We don't use it for task management. That's ClickUp's job. Notion is the reference layer.
GitHub holds the code we write. Custom agents, integration scripts, scraper utilities. Version control matters even in an agency context. When something breaks six months from now, you want to know what changed.
Google Analytics 4 for web traffic. Looker Studio for dashboards that pull GA4 data and display it in a format clients can actually read. We build these dashboards once and then let them run. No weekly manual reporting.
Financial tracking runs in Google Sheets. On purpose. We don't need a BI tool for an agency our size. A well-structured spreadsheet that gets updated consistently beats an elaborate dashboard that nobody maintains. The goal is decision-making, not impressive tooling.
Tools we paid for and cut, and why.
Replaced entirely by GHL. We were paying for two email platforms that overlapped at 90%. Once GHL's sequences matured for our use cases, ActiveCampaign was gone. Redundancy is expensive.
Beautiful interface. We used it for about four months before realizing we were managing the tool instead of using it. Moved to ClickUp because it integrates better with the rest of the stack and has automations that actually work without a paid add-on.
Jasper, Copy.ai, and a few others. We tried them early when the hype was loud. The output required so much editing that it wasn't saving time. We now do all AI-assisted writing directly through the API with prompts we've built and refined ourselves. Better control, lower cost.
We loved the flexibility. The problem was it required constant maintenance to keep base structures current as projects changed scope. ClickUp handles the same job with less upkeep.
This stack isn't a prescription. Your business has different constraints, different data, different team size. What works for a 7-person agency in Longview doesn't automatically port to a 40-person distribution company in Shreveport.
The point is when we talk to you about automation, we're not guessing. We've run these tools under real conditions. We've hit the walls. We've switched when something better came along. The consulting work we do is informed by years of operating our own version of this stack.
When we recommend a tool to a client, we've usually already broken it ourselves. That matters. It means we know where the edge cases live before you pay to discover them.
If you want to see how we've applied this stack to real client problems, the case studies are the right next read.
We'll map your current tools, find the gaps, and tell you exactly what to add, cut, or replace. No generic recommendations.
Schedule a Consultation Take the Free Assessment