Scaling a WhatsApp-First Partner Acquisition Motion
How I redesigned a manual, WhatsApp-first partner acquisition motion in Brazil and built the AI-assisted system that supports it, with human judgment kept where it matters.
- Role
- Strategy, operating model & implementation
- Context
- Live production
- Channel
- Measured window
- 31 days
- Human in the loop
- Final contract
- Market
- Brazil
Executive summary
Business context
In Brazil, B2B buying conversations happen on WhatsApp. A marketplace expansion motion in this market means high-volume prospecting of small and mid-size partners, on a channel that punishes generic messaging and rewards fast, contextual replies.
I had run exactly this motion by hand: 80+ cold calls a day earlier in my career and a full-cycle funnel every day since. That frontline experience is what the operating model is built on, knowing which steps are repeatable and where a human must stay in the loop.
Commercial challenge
- 01High lead volume that manual research could not keep up with
- 02A channel sensitive to generic, templated messaging
- 03Conversations that require real commercial context to advance
- 04Repetitive manual work: research, follow-up, scheduling, CRM updates
- 05Risk of losing quality when scaling volume
- 06A motion dependent on individual seller knowledge
- 07Actions that needed to land in the CRM without extra admin
The challenge was never just building an agent. It was redesigning how the motion operates, then deciding where technology fits.
Diagnosis
- Mapped the commercial flow end to end and timed where seller hours actually went.
- Separated process problems from technology problems: qualification and prioritization had to be fixed before any automation.
- Identified which tasks were genuinely repeatable (research, first touch, follow-up cadence, scheduling, CRM sync) and which depended on judgment (negotiation nuance, gatekeepers, final commitment).
- Documented the playbook that top-performing conversations followed, so it could be encoded instead of improvised.
GTM strategy
Operating model
- Capture
- Enrich
- Qualify
- Converse
- Book
- Sync
- Measure
Human vs automated responsibilities
An operator override silences the system the moment a human replies within 15 minutes. Escalation is designed in, not an afterthought.
System implementation
With the operating model defined, the system was built serverless and integrated with the CRM and calendar. Technology choices followed the strategy, not the other way around.
Interface screenshots
Screens from the interactive demo environment. All data shown is synthetic; no real prospect or customer information is displayed.
Pipeline visibility
Opportunities move through explicit stages with values attached, so pipeline reviews reflect the real state of the motion.
Agent configuration and guardrails
Where the operating model is enforced: what the AI may do alone, what waits for approval and which guardrails apply.
Measurement
The feedback loop: reply rates, funnel conversion and workload signals that show whether the motion is improving.
Metrics and observed outcomes
System figures come from a single measured 31-day production period and describe activity within that window; they should not be read as a strict sequential funnel, and contracts are not attributed to the AI alone. Quota attainment is the seller's monthly figure over the same period. Human approval remained required at the final contract stage.
Risks and guardrails
Every guardrail traces to a real production risk. They exist so the system fails safely instead of confidently.
anti-hallucinated-bookingA reply claiming an invite was sent without a booking tag forces a retry, then falls back to honest text.
anti-false-closeNo contract is marked without a verified tax ID in the conversation.
anti-pitch-to-gatekeeperThe system won't run a pitch at someone identified as a gatekeeper.
command-tag firewallInternal command tags are stripped so they can't leak or be injected.
deliverability pacingPer-number daily caps, business-hour windows and delivery verification protect the channel.
operator overrideThe system goes silent the moment a human replies within 15 minutes.
Lessons learned
- Fix the process before automating it. Automating a broken qualification step scales the problem, not the pipeline.
- A delivery API accepting a message does not mean it was delivered. Delivery verification had to be built after learning this in production.
- Channel limits are a design constraint, not an afterthought: pacing, caps and business-hour windows exist because volume without them breaks the channel.
- Automation without context produces generic communication. The playbook and guardrails exist to prevent it.
- Attribution between system activity and revenue must be measured carefully and claimed conservatively.
- Human review stays where judgment creates value: sensitive conversations and the final contract.
What I would improve next
The transferable asset is not the codebase. It is the method: map the motion, separate process from technology, encode the playbook, define human decision points, automate the repeatable work and instrument the funnel. That sequence applies to any team scaling a high-volume, conversation-driven motion.
Want to walk through the operating model, the guardrail design or the measurement notes? Book a short call or email me directly.