How to Reduce Churn & Boost Growth with Fast, Empathetic Customer Support — Eli Winderbaum, Captions

At four people, before any dedicated support hire, one engineer answered customer messages in Intercom from 6-9 AM and again at the end of her day, and fixed issues she saw straight in the code. Gives the fastest possible feedback loop while finding product-market fit; replaced by a support team as the company grew.

After evaluating every AI agent/chatbot on the market, Captions chose Parahelp, which tag-teams every incoming message with human agents: it answers easy ones and hands off complex ones. Program what it must not answer (e.g. 'give me Captions Max free for lifetime'). On handoff the human sees the whole conversation so the customer never repeats themselves. It drafts answers from two sources: up-to-date help center docs (Mintlify) and the most recent human agent replies, so docs must be kept current. Humans get 2-4x more time per interaction to build rapport.

Keep normal human support hours (Captions had 8 hours a day, 7 days a week) and turn on the AI agent only off-hours, so every customer gets an immediate friendly acknowledgment and humans reply the next day. Lets cautious teams test AI and market 24/7 support right away.

Follow-the-sun staffing: at least one agent in the Americas, one in APAC, one in EMEA, always at least one or two humans on. Time to first response counts even if the first reply is just 'we received your message'. Suggested ramp for smaller teams: 8 hours x 5 days, then 10 hours x 5 days plus some weekend, then 12 hours x 7 days, then 24/7. Once you market 24/7, you must keep the promise.

Localize the App Store and Play Store listings, website, help center (the docs platform handles translation) and app. Support uses a tool that translates every customer message into English and the English reply back into the customer's language. Brazil is one of Captions' biggest markets; launching the website in Portuguese produced a huge lift in Brazil installs.

The Intercom SDK on iOS, Android and desktop passes email, device, OS version and app opens. Three RevenueCat fields appear on every conversation: latest entitlement (plan), expiration date (trial or subscription) and status (auto-renew on or canceled; RevenueCat built this one at Captions' request). Rules let canceled-trial, churned or premium customers skip to the front of the line, skip the AI agent, or reconnect to the agent they last spoke with. Replies are tailored per billing platform (Apple, Google, Stripe).

One agent was a customer who kept sending feature requests and bug reports; he was offered a job. Another (in India) answered other users' questions in Discord off-hours, linking to the help center; after a week Eli DM'd him and hired him. Advice: tweet that you're hiring and post the role in your Discord.

When Captions was 10 or fewer people, every employee was in Discord, tagging each other to answer; the CEO replied directly and floated ideas ('What if we added this feature?'). This replaced surveys, calls and paid research. As the product matured and grew noisier, Discord support questions were routed into in-app messaging so customers could be identified, while the Discord stayed active.

When customers repeatedly ask and vote for something, sneak in a small fix or quick win alongside the roadmap and have the CEO post that it's done because you listened. Direct access to leadership turns vocal users into customers for life.

Inspired by Tony Hsieh's 'Delivering Happiness', a Slack channel called C4L (customer for life) collects screenshots of customers saying the app changed their business or got them more TikTok views, plus App Store reviews that name a support agent, so the whole company sees who they build for and agents get recognition.

After helpful interactions, support asks customers for App Store and Play Store reviews. Reviews that praise named agents are shared internally and signal great support to prospective downloaders; some customers later ask for the same agent by name.

Contact rate = share of subscribers who reach out in a month (Captions ~3%). Use it to forecast next month's inbound volume and staffing, and investigate swings (e.g. 1.5% to 4%): a failed release, an outage, a coupon or promotion. Marketing must tell support before promotions, billboards or big campaigns so it can staff up or hire temps.

Collect 1-5 CSAT after support conversations (Captions has switched it on and off for months at a time). Rather than chasing the absolute score, compare agents: if the team averages 4.5 and one agent gets 3.5, talk to them about product knowledge, feeling rushed or personal issues, and adjust their schedule if needed.

Captions tested a seven-day free trial against its three-day trial; the seven-day version wasn't as good, despite RevenueCat's report finding longer trials did better. No details on the metric.

Every feature Captions launches gets a two-week test period comparing variants to make sure it performs better before full rollout.

The host, doing support alone for his side project app, asks for a rating and/or review after nearly every good support experience and links directly to the review page.