Respond to the signal before the customer does
Accounts go silent weeks before they leave, and in the PingPong KYC funnel the first sign was usually a verification that stalled. I build lifecycle systems that read that signal in real time and act on it, and every signal lands in Salesforce so a human can follow the automation.
Churn is a signal you ignored weeks ago
When I inherited lifecycle at PingPong, retention meant a shared MailChimp account serving three regions with no owner and no CRM sync. Sends went out on calendar dates to whoever was on the list. Nobody knew which accounts were stalling, because nothing was listening.
I rebuilt it on one rule: journeys trigger on account state and usage signals. When a verification stalls, the account is telling you something hours or weeks before it would ever write in. I ran this repeat-usage engine inside the $2M → $76M → $250M+ cross-border payments line. Marketing on that line was mine end to end; sales and finance ran their own sides of it, and my journeys fed them. Every journey I shipped carried a measurable job: registered → first payment → habitual use.
How a save starts
before anyone writes in
01 · Instrument the signals
Before any cadence exists, every engagement signal gets a home. I sync opens, clicks, document submissions and usage events back into CRM scoring across HubSpot, Salesforce and Pipedrive, so account managers can see which accounts are moving and which ones went silent after submitting documents. I write every stall signal back to CRM the day it happens. I never had a churn baseline to measure the saves against, so I won't quote a save rate.
02 · Segment by account state
By where the account is. I built the segments around state: I orient pre-KYC users, unstick the ones in verification, and run activation once compliance approves. Those three journeys run off five state triggers, laid out in the activation principle. Each state has its own job, and an account moves segments the moment its state changes.
03 · Branch on the behavior
A document submitted, a verification stalled, a first payment pending. Each signal branches to its own cadence. The engine I built ran as real-time automation at 100–300 new registrations a day, with no manual intervention and no weekly list pulls. I wanted saves to fire without anyone remembering to open a report.
04 · Measure and tighten the loop
Every cadence I ship carries its number, and the lowest performers go into my A/B queue. AI refines the segments and scores the leads, so the loop tightens. The 200+ workflows I built cut lead handoff from ~12 hours to 1–2 minutes. I point the same machinery at keeping accounts.
Two places I caught the signal first
The support queue was a churn-risk feed nobody read
Account state is the signal inside the product. The support queue is the one outside it. At a later company, the earliest churn signals were buried in support tickets, mixed in with sales ops requests, accounting questions and product bugs. I built an AI routing layer on the queue (GPT-4o mini plus Zapier) that classified and tagged every ticket on arrival. Misrouted sales ops, accounting and product work disappeared from the support queue, and my classifier's tags gave the first honest picture of what support actually spent time on: an engagement and workload signal no one had counted before.
With my classifier absorbing the triage, the same rep went from replying to reviews on one platform to three. Trustpilot, G2 and Capterra were covered, with no added headcount. The replies were her work; the hours were mine to give back. I never wired a dashboard for it, so I won't quote a deflection percentage. The queue got quieter and the replies got faster; that part I watched happen.
Review coverage 1 → 3 platforms · same single rep · zero added headcountRebuilt onboarding around real-time account status
The biggest churn point was a verification that never finished. New registrants would start KYC, hit friction, and drift. I rebuilt the onboarding journeys around real-time account status: the moment a verification stalled or a document sat unprocessed, my journeys sent the right support message, and I wired the signal back into Salesforce so a human could follow the automation.
I instrumented it, and KYC submissions went from roughly 10% of registrations to 35–40%, on completion times of 24–48 hours instead of 140. That half of the rebuild belongs to the activation principle. Customer behavior stayed the same. The system I rebuilt finally answered them on time.
Submissions ~10% → 35–40% · completion 140h → 24–48hDaniel is a professional, knowledgeable individual, that takes pride in the quality of his work. His reliability is second to none and he is always up to the challenge.
Kathy SongOperations Manager
Losing accounts before anyone notices?
Behavior turns weeks before revenue does, so I build the system to watch the behavior.