The ‘Sarah’ Protocol: Architecting the DIY AI Sales Agent on a Freshman Budget
- Govind Davis
- Feb 9
- 5 min read
Updated: 5 hours ago
There is a fundamental shift happening in the sales world right now. It isn’t just about automation; it’s about personification.
It’s about taking the rigid, linear workflows of the past—the BDR grinding out cold emails, the SDR chasing qualified leads—and fusing them into a digital entity that sleeps less, costs a fraction of a human salary, and follows instructions with algorithmic precision.
We recently sat down for a ‘Morning Scrum’—a live build-in-public session—to tear down the theory and actually architect this solution.
Joined by Gary Walkup, a veteran sales executive who has seen every methodology from the IBM days to the modern SaaS era, we mapped out a plan to build “Sarah.”
Sarah isn’t a person. She is a logic flow. She is an orchestrated set of API calls. But to the market, she wears two very distinct hats. If you are a solopreneur, a consultant, or a sales leader trying to scale without bloating your headcount, this is the blueprint for the ‘Crawl’ phase of AI adoption.
The Concept: One Agent, Two Hats
In traditional sales organizations, you have a split. You have the Hunter (the Business Development Representative or BDR), whose sole job is to kick down doors and generate interest. Then, you have the Nurturer/Closer (the Sales Development Representative or SDR), who takes that interest, qualifies it, and sets the appointment for the executive.
Mike’s vision for this project was specific: he didn’t want a generic email bot. He wanted an entity that could handle the friction of the hand-off autonomously. He calls this the ‘Sarah Two-Hat Workflow.’
Hat 1: The BDR (The Hunter)
This is the front line.
In our architecture, the BDR function is entirely text-based. It’s about leverage and volume, but with a crucial compliance twist. We aren’t in the Wild West anymore; you cannot simply unleash an AI voice bot to cold call thousands of numbers without facing significant legal peril (and fines).
Therefore, Sarah’s first job is Consent Acquisition. She uses cold email to identify the Ideal Customer Profile (ICP), engaging them with a compelling offer—an audit, a proof of concept, or a valuable assessment. The goal isn’t to sell the product yet; it is to sell the *conversation*.
Hat 2: The SDR (The Closer)
Once consent is granted—digitally signed or explicitly stated via email reply—Sarah swaps hats. She becomes the SDR. Now, leveraging tools like Retell AI or Synthflow, the agent picks up the phone. This isn’t a robocall; it’s a generative AI voice conversation.
She asks the BANT (Budget, Authority, Need, Timeline) questions. She qualifies the lead. And if the lead passes the test, she books the meeting for the human executive.
As Gary Walkup pointed out during our session, the fundamentals haven’t changed. Whether it’s a 6’6″ rugby player in a scrum or a Python script running on a server, the goal is moving the ball forward. But now, we can automate the grunt work.
The ‘College Freshman’ Budget
The biggest barrier to entry for most solopreneurs is cost. Enterprise AI solutions can run thousands of dollars a month.
Mike H laid down a challenge: Build this entire stack for under $225 a month. He calls it the “College Freshman Budget”—enough money for beer and pizza, or in our case, enough money to disrupt an industry.
To make this work, we have to embrace the Crawl, Walk, Run methodology.
Phase 1: The Crawl (Current Focus)
We are stripping away the vanity metrics. We aren’t trying to boil the ocean with LinkedIn automation (which carries a high risk of account bans) or complex multi-channel sequences initially.
The Lean Stack:
* Data & Email: Apollo. For roughly $99/month, you get the contact data and the email sequencing engine. It’s the bread and butter.
* The Brain: Google Gemini (via API). Why? Because it’s cost-effective and integrates seamlessly if you are already in the Google ecosystem. It acts as the “Advisor,” helping craft the personalized messaging.
* The Voice: Retell AIor Synthflow. These are usage-based. You pay by the minute. If Sarah only talks to people who have consented, your wasted spend drops to near zero.
* The Glue: Make.com. This is where the magic happens. Make replaces the need for a dedicated engineering team. It listens for a webhook (e.g., “Lead Replied in Apollo”), processes the logic, and triggers the next step (e.g., “Add to Call Queue”).
Under the Hood: Solution Architecting on the Fly
During the session, we moved from theory to the actual whiteboard. It’s one thing to draw a cartoon of a robot; it’s another to wire up the API keys.
We looked at the ‘Sarah’ workflow not just as a sales funnel, but as a data pipeline.
The Data Repository
This is the piece most people miss. Sarah needs a memory. You cannot just fire and forget. When a prospect replies to an email, that data needs to be extracted, enriched, and stored.
We discussed using a simple database structure within Make or pushing it back into a CRM like HubSpot. This “backpack” of data is what allows the AI to sound intelligent on the phone later. It knows who it’s talking to.
The Orchestration Layer (Make.com)
We did a live demo of setting this up.
It starts with a Trigger: *New Email Response*.
Then, a Filter: *Is this positive sentiment?* (Using Gemini to analyze the text).
If yes, we move to
Action: *Update Record & Trigger Call Flow*.
Connecting Google Gemini requires setting up a Google Cloud Console project to get your API key. It sounds daunting, but as we showed, it takes about 10 seconds to generate the key and plug it into Make. Once connected, you aren’t just sending templates; you are generating context-aware responses on the fly.
The Human in the Loop (HITL)
Despite our love for automation, Gary brought us back to reality with a critical question: “What are the standards for qualification?”
An AI agent can process data faster than any human, but it lacks intuition. This is where the Human in the Loop is vital, especially in the ‘Crawl’ phase.
We designed a checkpoint in the workflow. Before Sarah makes that phone call, or before a meeting is officially confirmed on the calendar, a human (Mike) reviews the lead. It’s a quality assurance step.
We want Sarah to do the heavy lifting of finding and warming up the lead, but we don’t want her burning bridges with high-value accounts because of a hallucination.
As Mike put it, “I’m selfish. I want to produce enormous amounts of revenue.” Sometimes that means letting the bot run; sometimes it means the human executive steps in to take over a “Tier 1” prospect personally. The architecture must support both paths.
The Reality of ‘Building in Public’
This isn’t a theoretical exercise. We are building this for real-world application (specifically for the EHP product line). The friction we encountered on the call—debating whether to use SMS, how to handle the ‘consent gate,’ and navigating the complexities of email deliverability—is the exact friction you will face.
But the conclusion is clear: The tools are here. The budget is manageable ($225/month is accessible to almost any serious business). The only missing ingredient is the will to wire it together.
We are moving from a world of “Dialing for Dollars” to “Architecting for Outcomes.” Sarah is ready to work. Are you?
