Architecture Overview

The 8 stages map to 4 application systems sitting on top of a single data layer. Tristan sees Airtable. The systems run as Python processes on Annette's machine (or a VPS if uptime matters).

Data flow sketch

Humans in the loop

Stage 07 (Close) is fully human. Tristan takes the call, negotiates, and marks the deal Won in Airtable. Everything before and after that is automated. Stage 06 routing may be semi-automated depending on Tristan's preference (confirm at discovery call).

Discovery Call Agenda (90 minutes)

Run this as a structured intake. Take notes in a shared Google Doc or record the call (get permission). Every bolded item is something you need before you can build.

Block 1: Current Pipeline (20 min)

• "Walk me through how you land a sponsorship deal today, start to finish. Who does what?"
• "How many brands do you reach out to in a typical month?"
• "What's your current close rate? How many outreach attempts to one signed deal?"
• Get the Google Sheet URL. Ask them to share it with [email protected]. Look at it live: columns, statuses, how they've been tracking.
• "Which of these deals are you tracking actively vs. dead?"
• "What does a deal look like when it falls through? What's the most common reason?"

Block 2: Brand Categories (20 min)

• "Let's go category by category. Combat sports brands: who are your dream sponsors? Who have you already talked to?"
• "Equestrian: same question. Who's shown interest, who haven't you touched?"
• "Outdoor / lifestyle: who's on your radar?"
• "Any categories that are off the table? Anything you've tried that didn't fit your audience?"
• "Tell me about your best deal ever. What brand, what amount, how'd it happen?"
• "Tell me about a deal you thought was going to close and it didn't. What happened?"
• Collect: a list of 10-20 brands per category (even half-formed is fine). These seed the prospecting database.
• "What does a bad-fit brand look like? Anything you'd be embarrassed to promote?"

Block 3: Boxer Network (15 min)

• "Walk me through how you use the boxer network today. How many fighters are in it?"
• "How do you offer a deal to them currently? Email, DM, text?"
• Get: current split percentage (Tristan's cut vs. the boxer's cut).
• "Is the split the same for every deal, or does it vary?"
• "What's the threshold where you'd rather close it yourself vs. hand it off?"
• "Have any fighters turned down a deal? Why?"
• "Do boxers have content deliverables, or is it mostly appearance/endorsement?"
• Collect: boxer name, Instagram handle, email, and specialty category for each fighter in the network.

Block 4: Voice Samples (10 min)

• "Can you share 2-3 DMs you've sent to brands that you felt good about? Even screenshotted."
• "Do you have any outreach emails you've sent? Could be from Gmail sent folder."
• "How would you describe your tone when you're pitching? Casual, professional, hype-y?"
• "Any phrases or sign-offs you always use?"
• Collect: at least 3-5 real outreach samples. These train the voice model for DeepSeek.

Block 5: Zoom Setup (10 min)

• "Do you have a paid Zoom plan?" (Free plan doesn't have cloud recordings.)
• "Are you currently recording your sales calls? Cloud or local?"
• "If cloud: we can pull via Zoom API automatically. If local: we'll set up a folder you drop files into."
• Decision: cloud API vs. watchdog folder. Get Zoom API credentials if cloud.
• "How long are your typical first-call brand conversations? 20 min? 45?"

Block 6: Tech Access Checklist (10 min)

Go through these one by one. Note what's ready vs. needs setup.

Block 7: Deliverable Preferences (5 min)

• "What do you want to see in the CRM every morning? What's the daily dashboard view for you?"
• "Do you want email summaries, or will you check Airtable directly?"
• "When a deal gets routed to a boxer, do you want to approve it first, or should the system send automatically?"
• "Who else on your team needs access? Manager, assistant?"
• "Any deal categories where you'd want a human flag before outreach goes out?"

Tech Stack Decisions

Airtable Data Schema

6 tables. Create these in order (linked fields require the target table to exist first).

Table 1: Brands

Table 2: Outreach_Log

Table 3: Deals

Table 4: Boxer_Network

Table 5: Boxer_Deals

Table 6: Fulfillment

Recommended Airtable Views

• Brands: Pipeline Kanban- grouped by status. Tristan's daily view.
• Brands: Outreach Queue- filtered to status = outreach_ready, sorted by qualifier_score desc.
• Outreach_Log: Today's Activity- filtered to sent_at = today.
• Deals: Active Deals- filtered to status IN (prospect, negotiating), sorted by next_action_date.
• Fulfillment: This Week's Deadlines- filtered to due_date <= today + 7, sorted by due_date asc.

Week-by-Week Build Plan

Stage-by-Stage Technical Detail

Instagram Safety Protocol

The line: Messenger API (official) for outreach, not browser automation. Account bans are existential risk for a 2M-follower creator.

On the PSID problem

The Messenger API requires a recipient's Page-Scoped User ID (PSID), not their Instagram handle. Brands that haven't interacted with Tristan's page don't have a PSID yet. Short-term: Phantombuster handles cold outreach (browser session DMs). Long-term: as brands reply and establish a thread, they get a PSID and move to the official API. Build a field in Brands (ig_psid) that gets populated when a brand first replies. Once populated, all future messages go through the API.

Testing + Go-Live Checklist

Each item must be manually verified before go-live. "The script ran without errors" is not verification.

Per-Stage Verification

• Stage 01: 90+ brands in Airtable across all 3 categories. At least 30 per category. All have instagram_handle or contact_email (not both required, but one is required).
• Stage 02: All brands have qualifier_score populated. At least 30 brands at outreach_ready status. Spot-check 5 scores manually against what you'd expect, if they look off, revisit the system prompt.
• Stage 03: Send 10 test DMs + 10 test emails to real brands. Confirm all 20 show up in Outreach_Log with correct channel, sent_at, and follow_up_date. Confirm brand status updated to "outreached". Confirm at least one email lands in a real inbox (test with a brand you have a contact at).
• Stage 04: Set a test brand's follow_up_date to today, run the scheduler manually. Confirm a follow-up message generates and sends. Confirm a new Outreach_Log entry is created. Confirm follow_up_date advances to next scheduled touch.
• Stage 05: Drop a test audio file (even a 2-minute call recording from your phone) into the watch folder. Confirm: transcription runs, Claude extracts fields, Deal record appears in Airtable with brand_id correctly matched, all extracted fields populated.
• Stage 06: Create a test Deal with amount = $10,000, status = negotiating, routed_to_boxer = false. Run the router. Confirm: Boxer_Deals record created, split_amount calculated correctly, offer email sent to boxer's email, Deals.routed_to_boxer = true.
• Stage 08: Set a test Deal to status = won. Confirm: Fulfillment records created for each deliverable type. Run the Monday summary job manually. Confirm summary email arrives in Tristan's inbox.
• Google Sheets sync: Run the gspread export manually. Open Tristan's Sheet. Confirm all three tabs updated with current Airtable data. Confirm nightly scheduled run fires (check next morning).
• End-to-end run: One brand through the entire pipeline: uncontacted → outreach_ready → outreached → (simulate reply) responded → deal_opened → (fake Zoom file) Deal record created → amount < $15k → routed to boxer → (manually) won → Fulfillment checklist created → weekly reminder fires. All Airtable records correct at each step.
• Handoff doc: One-pager delivered to Tristan: daily system behavior, his responsibilities (mark Won, drop Zoom files, check Airtable morning view), emergency contact if system breaks.
• Process restart: Windows Task Scheduler configured for all Python processes. Simulate a reboot. Confirm all processes restart automatically.

NEED Flags: Confirm at Discovery Call