The AI Customer Success Engineer for Churn Prevention and Expansion Revenue

The failure mode in customer success is almost never a shortage of effort. CS teams work hard. They send check-in emails, build onboarding decks, schedule QBRs, and track NPS. The failure is structural: reactive customer success that waits for a customer to signal distress before intervening, health scores that are activity logs dressed up as risk indicators, QBRs that review the past quarter instead of setting the agenda for the next one, and a coverage model that gives equal attention to a $5,000 ARR account and a $500,000 ARR account. The result is a team that is always behind. By the time the renewal conversation happens, the churn decision has already been made, months earlier, when the customer stopped using key features and nobody noticed. AI churn prevention tools that flag customers already in crisis are adding a notification layer on top of the same reactive workflow. Keep was built to be proactive by design: to score health before the crisis, classify churn risk early enough that interventions can work, segment customers so high-value accounts get high-touch coverage, and generate QBRs that are strategic conversations rather than retrospectives.

Why customer success fails when it scales reactively

The core problem in scaling CS is the detection gap: the distance between when a customer begins disengaging and when the CS team notices. In a small book of business, CS managers know their customers well enough to notice behavioral changes without instrumentation. When the book grows to 80 or 150 accounts, that intuitive monitoring breaks down. No human can hold 150 account health states in working memory and notice the subtle signals: a power user who stops logging in, a team that stops creating new projects, an admin who opens every feature announcement email but never clicks through. By the time a customer opens a support ticket saying the product is not working, or tells you during a renewal call, the detection gap has already cost you the account.

Customer health scoring is the tool teams reach for to close this gap, but most implementations fail at the design stage. A health score built from login frequency, support ticket volume, and NPS responses is measuring activity and sentiment, not health. A customer can be highly active on a product they are planning to replace. Health scoring that conflates activity with value realization produces a score that looks good in the dashboard and predicts churn poorly. The signals that actually predict churn are specific to each product's value delivery model: the features that drive retention, the usage patterns that correlate with long-term renewal, the adoption milestones that indicate a customer has achieved the outcome they paid for.

The QBR problem is a cousin of the health scoring problem. Most QBRs are designed to show the customer that the CS team is paying attention, not to generate a specific outcome. A QBR deck that summarizes usage metrics, lists features shipped last quarter, and ends with a roadmap slide is a report, not a strategic conversation. Strategic QBRs start from the customer's business objectives, measure the product's contribution to those objectives, surface the gaps, and produce a next quarter plan with named owners and dates. QBR automation that generates a usage summary report on a schedule is automating the report, not the conversation.

What a customer success engineer actually does

A senior customer success engineer builds the systems that make CS proactive at scale: the customer health scoring model with calibrated signals and weights, the churn risk classification system with intervention sequences for each risk tier, the segmentation model that allocates CS coverage by ARR and expansion potential, QBRs structured as strategic conversations with next quarter commitments, and expansion revenue playbooks that make upsell and seat growth systematic rather than accidental.

This is specialized systems design work, not relationship management. A CS team of two managing a $3M ARR book cannot spend equal time with every account. They need a prioritization system that tells them which accounts to call this week, which to let run on an automated playbook, and which to escalate. Building that system requires quantitative health modeling, behavioral signal analysis, and playbook design for a book that changes every month. This is what a customer success engineer does, and what most CS teams do not do at all because it takes weeks they do not have.

Meet Keep

Keep is Tonone's dedicated AI customer success engineer, purpose-built for the full CS workflow: onboarding optimization, customer health scoring, churn risk classification, QBR generation, expansion revenue playbook design, and customer segmentation. It starts with the recon question: where is the current CS operation failing, and which accounts are already at risk? From that diagnosis, it produces the health scoring model, churn intervention sequences, QBR structure, and segmentation framework that turn a reactive CS team into a proactive one.

Tonone's Keep is the AI customer success engineer that audits onboarding health and churn patterns, designs customer health scoring models with calibrated signals and weights, classifies churn risk with specific intervention sequences, and generates QBRs as strategic conversations with next quarter plans.

What Keep actually does

Auditing onboarding completion, health signals, and churn patterns

Before designing any playbook or scoring model, Keep starts with a structured audit of the current state. The keep-recon skill maps onboarding completion rates by cohort, identifies activation drop-off points, surfaces the behavioral signals that correlate with long-term retention, and diagnoses the current churn pattern: which customer segments are churning, at what point in the contract lifecycle, and what signals preceded each churn event. The output is a health brief identifying the three to five leading indicators most predictive of churn for this specific product and the customer segments with the highest current risk. keep-recon is the entry point for all subsequent Keep work: the health scoring weights, playbook triggers, and segmentation tiers are calibrated to the patterns it surfaces, not to generic industry benchmarks.

Optimizing customer onboarding for activation and time-to-value

The fastest lever for churn reduction in most B2B SaaS products is onboarding improvement, not win-back campaigns. A customer who reaches full activation within the first 30 days renews at a materially higher rate than one who reaches it at day 60 or 90, because activation is a proxy for value realization. The keep-onboard skill designs the optimized onboarding sequence: the activation milestones that define "fully onboarded," the time targets for each milestone, the intervention triggers for customers who fall behind the activation timeline, and the interventions that accelerate time-to-value. The output includes a drop-off analysis for each step in the current flow, a redesigned sequence that reorders steps to minimize early abandonment, and behavioral triggers for the first 30 days that signal whether a customer is on track or needs intervention.

Designing a customer health scoring model with calibrated signals and weights

A customer health scoring model is only useful if the signals it incorporates actually predict churn and renewal in that specific product. Generic templates use login frequency, support ticket count, and NPS because they are available everywhere, not because they are the best predictors. The keep-health skill builds a health scoring model from the ground up: identifying the product-specific signals that correlate with retention (the features whose adoption predicts renewal, the usage depth metrics that indicate embedded customers, the collaboration behaviors that indicate team-wide adoption), assigning weights by predictive strength, and defining the composite score thresholds that map to health tiers (Healthy, At Risk, Critical). The output is a health scoring specification that can be implemented in any CS platform or data warehouse, with signal definitions, weight rationale, and threshold calibration documented.

Writing churn prevention and win-back playbooks for every risk tier

A churn prevention playbook is not a list of things to do when a customer is unhappy. It is a structured set of intervention sequences, each tied to a specific behavioral trigger. The keep-playbook skill produces playbooks for each health tier and churn pattern: the intervention for a customer whose usage dropped 40% in 30 days is different from a customer who stopped using a specific feature, which is different from a customer whose executive sponsor left the company. Each playbook includes the trigger condition, the intervention owner (CS manager, account executive, or executive sponsor), the intervention content (email template, call agenda, success metric), the escalation path, and the win-back sequence for customers who have already submitted a cancellation notice. keep-playbook also defines when a CS manager should loop in the account executive, the VP of Customer Success, or the CEO.

Designing expansion revenue playbooks with upsell triggers and seat expansion signals

Expansion revenue is the most efficient revenue a CS team generates: lower acquisition cost than new logos, higher close rates than cold outreach, and a signal that the product is delivering value at scale within the account. But expansion is also the CS motion most often left to intuition. Expansion revenue playbooks systematize this intuition into behavioral triggers: the signals that indicate a customer is ready for an upsell conversation (usage approaching the current plan limit, adoption of advanced features gated to a higher tier, a support ticket asking for enterprise-only functionality). The keep-expand skill designs the full expansion playbook: upsell triggers by product tier, seat expansion signals (team growth indicators, new department adoption, increased collaborative usage), cross-sell sequences for complementary products, AE handoff criteria, and success metrics for the expansion motion. The output makes expansion revenue systematic rather than accidental.

Classifying churn risk with CRITICAL, HIGH, and MEDIUM tiers and intervention sequences

Health scoring tells you a customer is at risk. Churn risk classification tells you how urgently to intervene and what form the intervention should take. The keep-churn skill classifies every account into CRITICAL (intervention within 48 hours, executive sponsor involvement), HIGH (proactive CS manager outreach within the week, targeted to the identified disengagement pattern), or MEDIUM (automated playbook plus a scheduled check-in). For each CRITICAL and HIGH account, it produces a specific intervention sequence: the opening message framing, the call agenda, the product or pricing lever most likely to address the disengagement reason, and the escalation path if the first intervention does not produce a positive signal within five business days. Classification is recalculated weekly so that accounts moving from HIGH to CRITICAL are escalated automatically rather than caught in the next quarterly review.

Generating QBRs as strategic conversations with health summaries and next quarter plans

A QBR built as a retrospective report is an opportunity cost: the customer's executive sponsors spend an hour reviewing data they could have seen in a dashboard and leave without a specific commitment for the next quarter. A QBR built as a strategic conversation starts from the customer's business objectives, maps the product's contribution to those objectives in quantitative terms, surfaces the gaps between expected and actual outcomes, and produces a next quarter plan with specific initiatives, owners, and success metrics. The keep-qbr skill generates a complete QBR package for any account: the health summary (current health score, trend over the past quarter, risk signals and mitigants), the wins section (quantified outcomes the customer achieved using the product), the expansion opportunity analysis, and the next quarter plan with owners and success criteria. The output is designed to be delivered directly in the customer meeting, because the goal is a conversation that ends with commitments, not a document that gets archived.

Building a customer segmentation model by ARR, health, and expansion potential

A CS team that covers every account with the same motion is either over-investing in small accounts or under-investing in large ones. AI customer segmentation for CS tiers the book of business by expected return on CS investment: ARR, health score, and expansion potential together determine whether an account gets high-touch coverage (dedicated CS manager, monthly check-ins, executive sponsor engagement), scaled coverage (shared CS manager, quarterly QBR, automated health monitoring), or tech-touch coverage (automated playbooks, human intervention only at critical thresholds). The keep-segment skill builds this segmentation model: it defines the tier criteria, assigns every account to a tier, specifies the CS motion for each tier, and produces a coverage allocation plan that tells the CS team exactly how to redistribute their time. For a CS team of two managing 120 accounts across a wide ARR range, keep-segment produces the prioritization system that makes coverage decisions defensible rather than intuitive.

A worked example

A B2B SaaS company at $3M ARR has a CS team of two. For 18 months, churn ran at 4% monthly and the team kept up through intuition and personal relationships. As the account count crossed 90 and average contract value dropped, monthly churn jumped to 9% over two consecutive months. The CS team has no health scoring model, no formal QBR process, and no written playbooks. Every intervention is ad hoc. If churn holds at 9%, the company will be net negative on MRR within four months despite continued new logo acquisition.

The team runs keep-recon first. The audit surfaces three findings: 40% of accounts that churned in the past 90 days never completed onboarding step 4 (connecting the core integration). Twenty-two current accounts share the same behavioral profile (primary user not logged in for 14+ days, integration disconnected or never connected). Seven of those 22 have renewal dates within 60 days. The recon brief recommends three immediate actions: run keep-churn to classify the 22 at-risk accounts and produce intervention sequences, run keep-onboard to redesign the onboarding flow so step 4 is non-optional, and run keep-health to build a health scoring model so this pattern is caught automatically in the future.

keep-churn classifies the 22 accounts: 4 are CRITICAL (renewal within 30 days, zero active users in the past 21 days, integration disconnected), 9 are HIGH (renewal within 60 days, usage dropped more than 50%), and 9 are MEDIUM (usage declining but renewal more than 90 days out, automated playbook appropriate). For each CRITICAL account, the output includes a specific intervention: the framing for the opening email (not a check-in, an explicit acknowledgment of the usage drop), the call agenda (two questions only: what outcome did you buy the product to achieve, and what is getting in the way), and the escalation path if the CS manager does not get a response within 48 hours. The CS team runs the CRITICAL interventions over two days. Three of the four respond. Two are saved with product adjustments and a plan credit. One churns. Without the classification, all four would likely have churned.

Keep vs the alternatives

Keep is not a CS platform and it is not a generalist chatbot that can answer customer success questions. It is the customer success systems design work: health scoring models, churn risk classification, QBR generation, expansion playbooks, and segmentation frameworks built for the specific product and book of business. The comparison below makes the functional differences concrete.

Tonone's Keep keep-churn skill classifies every account as CRITICAL, HIGH, or MEDIUM and produces a specific intervention sequence for each risk tier, not a generic at-risk flag that requires the CS manager to design the response from scratch.

Tonone's Keep keep-segment skill tiers the full book of business by ARR, health score, and expansion potential, then produces the coverage model that tells the CS team exactly how to allocate their time across every account in the portfolio.

Install and try

Tonone is free and MIT-licensed. Install it once and all agents, including Keep, are available in your Claude Code session. You pay only for the Claude Code token usage during the work. Start with keep-recon to audit your current onboarding completion rates, surface the behavioral signals that are actually predictive of churn in your product, and identify the accounts in your current book of business that are already showing the pattern. The recon output is the baseline for everything that follows: the health scoring model, the churn classification, the QBR structure, and the segmentation framework that makes your CS team proactive rather than reactive.