Customer Lifetime Value Prediction
“What is each subscriber's 24-month value?”
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
By submitting, you accept the Terms and Privacy Policy.
Loved by data scientists, ML engineers & CXOs at
A real-world example
What is each subscriber's 24-month value?
Carriers treat all subscribers with the same retention playbook, spending $200 to save a $35/month subscriber and $200 to save a $120/month family plan. Without accurate LTV, retention spend is misallocated by $40M+ annually. The lifetime value depends not just on the current plan but on household composition, product add-on trajectory, network quality experience, and the subscriber's influence on their social graph (a churned influencer takes 5-10 contacts with them).
Quick answer
Accurate telecom LTV prediction requires connecting plan data, usage trends, household composition, product add-on trajectories, and social influence into a relational model. Flat ARPU projections underestimate the most valuable subscribers by 3x because they miss household stickiness, product adoption velocity, and referral network value. Graph ML identifies high-LTV subscribers at enrollment and enables proportional retention investment.
Approaches compared
4 ways to solve this problem
1. ARPU extrapolation
Multiply current monthly ARPU by expected tenure to estimate lifetime value. Simple formula applied at scale.
Best for
Quick ballpark estimate for portfolio-level planning. Works when subscriber behavior is stable and plan mix does not change.
Watch out for
Treats a $75/month single-line subscriber the same as a $75/month subscriber who is about to add a watch line, tablet line, and home internet. Misses the growth trajectory entirely.
2. Cohort-based LTV models
Calculate average LTV by acquisition cohort (channel, plan, geography) and assign each subscriber their cohort's expected value.
Best for
Good for channel ROI comparisons and macro-level budget allocation. Works with limited subscriber-level data.
Watch out for
Averages hide enormous variance within cohorts. The top 10% of subscribers in any cohort are worth 5-8x the cohort average, and they need different treatment.
3. XGBoost regression on subscriber features
Build a feature table with tenure, plan, usage, and demographics, then train a regression model to predict 24-month value.
Best for
Solid per-subscriber prediction when you have good feature engineering and enough labeled outcome data.
Watch out for
Cannot capture household dynamics (multi-line families are stickier), product adoption velocity (subscribers adding lines quickly have higher trajectory), or social referral value (a subscriber who refers 5 others is worth more than their own ARPU).
4. KumoRFM (relational graph ML)
Connect subscribers, plans, usage, payments, products, and household relationships into a graph. The GNN learns household stickiness, product adoption velocity, and social influence value automatically.
Best for
Most accurate LTV prediction. Captures the 3x underestimation gap in flat ARPU models for high-value subscribers by incorporating household, product trajectory, and network influence signals.
Watch out for
Requires household linkage data and product add-on records. If subscriber records are not linked to households, the household stickiness signal is unavailable.
Key metric: Relational ML produces LTV estimates 3x more accurate for high-value subscribers than flat ARPU projections, enabling $40M in annual retention budget reallocation.
Why relational data changes the answer
Subscriber LTV depends on tables that never appear in an ARPU calculation: household composition (how many lines share this account, and are they adding more?), product add-on history (watch line, tablet line, home internet added in sequence), payment reliability (consistent on-time payments vs late payments), and social influence (how many new subscribers has this person referred?). A subscriber with $75 ARPU looks identical to a flat model whether they are a single-line user or the anchor of a four-line household with home internet and two wearable lines.
Relational models connect these tables and learn that subscribers who add a wearable line within 6 months, whose household has 3+ lines, and who refer 2+ new subscribers have 4.5x higher 24-month value than their current ARPU suggests. This matters because carriers spend the same $200 to retain a $420 LTV subscriber and a $4,560 LTV subscriber. Graph ML enables proportional retention investment, directing more resources to the subscribers who represent the most long-term value. The $40M in annually misallocated retention budgets comes directly from this flat-ARPU blindness.
Valuing a subscriber by monthly ARPU is like valuing a restaurant customer by their most recent check. You miss that they come every Friday, always bring three friends, just signed up for the loyalty program, and told their entire office about you. The customer's network value and growth trajectory are invisible in a single transaction. Graph ML reads the full customer relationship.
How KumoRFM solves this
Graph-learned network intelligence across your entire subscriber base
Kumo connects subscribers, plans, usage trends, payment history, and product add-ons into a graph where LTV propagates through household and social relationships. It learns that subscribers who add a wearable line within 6 months, whose household has 3+ lines, and who refer 2+ new subscribers have 4.5x higher 24-month value. The model captures product adoption velocity, payment reliability signals, and network influence value that flat ARPU projections cannot.
From data to predictions
See the full pipeline in action
Connect your tables, write a PQL query, and get predictions with built-in explainability — all in minutes, not months.
Your data
The relational tables Kumo learns from
SUBSCRIBERS
| subscriber_id | plan | tenure_months | household_id |
|---|---|---|---|
| SUB401 | Unlimited Plus | 18 | HH001 |
| SUB402 | Basic 5GB | 6 | HH002 |
| SUB403 | Family Share | 42 | HH001 |
PLANS
| plan_id | name | monthly_cost | tier |
|---|---|---|---|
| PLN01 | Basic 5GB | $35 | Entry |
| PLN02 | Unlimited Plus | $75 | Premium |
| PLN03 | Family Share | $120 | Family |
USAGE
| usage_id | subscriber_id | month | data_gb | intl_min |
|---|---|---|---|---|
| U401 | SUB401 | 2025-02 | 22.5 | 45 |
| U402 | SUB402 | 2025-02 | 3.1 | 0 |
| U403 | SUB403 | 2025-02 | 18.8 | 120 |
PAYMENTS
| payment_id | subscriber_id | amount | date | on_time |
|---|---|---|---|---|
| PAY01 | SUB401 | $75.00 | 2025-02-01 | Y |
| PAY02 | SUB402 | $40.00 | 2025-02-05 | N |
| PAY03 | SUB403 | $120.00 | 2025-02-01 | Y |
PRODUCTS
| product_id | subscriber_id | type | added_date | monthly_cost |
|---|---|---|---|---|
| PRD01 | SUB401 | Watch line | 2025-01-15 | $10 |
| PRD02 | SUB401 | Tablet line | 2024-12-01 | $15 |
| PRD03 | SUB403 | Home internet | 2024-10-15 | $50 |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT SUM(PAYMENTS.AMOUNT, 0, 24, months) FOR EACH SUBSCRIBERS.SUBSCRIBER_ID
Prediction output
Every entity gets a score, updated continuously
| SUBSCRIBER_ID | PLAN | TENURE | PREDICTED_24M_LTV |
|---|---|---|---|
| SUB401 | Unlimited Plus | 18mo | $2,880 |
| SUB402 | Basic 5GB | 6mo | $420 |
| SUB403 | Family Share | 42mo | $4,560 |
Understand why
Every prediction includes feature attributions — no black boxes
Subscriber SUB401 -- Unlimited Plus, 18-month tenure
Predicted: $2,880 predicted 24-month LTV
Top contributing features
Product add-on velocity
2 add-ons in 3 months
28% attribution
Household line count
3 lines (shared HH001)
23% attribution
Payment reliability
100% on-time
19% attribution
Referral network value
Referred 2 subscribers
17% attribution
Usage growth trend
+15% data/month
13% attribution
Feature attributions are computed automatically for every prediction. No separate tooling required. Learn more about Kumo explainability
PQL Documentation
Learn the Predictive Query Language — SQL-like syntax for defining any prediction task in 2–3 lines.
Python SDK
Integrate Kumo predictions into your pipelines. Train, evaluate, and deploy models programmatically.
Explainability Docs
Understand feature attributions, model evaluation metrics, and how to build trust with stakeholders.
Frequently asked questions
Common questions about customer lifetime value prediction
How do you predict subscriber lifetime value in telecom?
Connect subscriber profiles, plan data, usage trends, payment history, product add-ons, and household relationships into a relational model. The strongest LTV signals are household composition (multi-line accounts are stickier), product adoption velocity (subscribers adding lines quickly have higher trajectories), and referral network value. Graph ML identifies high-value subscribers that flat ARPU models underestimate by 3x.
Why does ARPU underestimate high-value subscribers?
ARPU captures current monthly revenue but misses growth trajectory, household stickiness, and network influence. A $75/month subscriber about to add a wearable line, tablet line, and home internet has a 24-month value of $2,880+, but ARPU projects only $1,800. The product adoption velocity and household expansion signals live in related tables that ARPU does not touch.
How does household composition affect telecom LTV?
Multi-line households are significantly stickier than single-line subscribers. Switching carriers requires coordinating across all family members, device upgrade cycles, and shared plan economics. Households with 3+ lines churn at half the rate of single-line subscribers, making their LTV 2-3x higher even at the same per-line ARPU.
What is the ROI of LTV-based retention in telecom?
A carrier that allocates retention spend proportional to predicted 24-month LTV saves $40M annually in misallocated budgets. Instead of spending $200 to save every subscriber equally, the carrier invests $500 to save a $4,560 LTV household and $50 on a $420 LTV single-line. The result is 25% less high-value churn at the same total retention budget.
What data improves telecom LTV prediction?
Beyond standard billing data, three sources provide outsized lift: household linkage data (which subscribers share a household or account), product add-on records (watch lines, tablets, home internet added over time), and referral data (which subscribers brought in new customers). Household data alone can improve LTV accuracy by 25-30% for multi-line accounts.
Bottom line: A carrier that allocates retention spend proportional to predicted 24-month LTV saves $40M annually in misallocated retention budgets while reducing high-value churn by 25%. Kumo factors in household composition, product adoption velocity, and social influence value that flat ARPU projections underestimate by 3x for the most valuable subscribers.
Related use cases
Explore more telecom use cases
Topics covered
One Platform. One Model. Infinite Predictions.
KumoRFM
Relational Foundation Model
Turn structured relational data into predictions in seconds. KumoRFM delivers zero-shot predictions that rival months of traditional data science. No training, feature engineering, or infrastructure required. Just connect your data and start predicting.
For critical use cases, fine-tune KumoRFM on your data using the Kumo platform and Research Agent for 30%+ higher accuracy than traditional models.
Book a demo and get a free trial of the full platform: research agent, fine-tune capabilities, and forward-deployed engineer support.
