Next Best Offer
“For each customer, which offer will they most likely redeem in the next 14 days?”
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
For each customer, which offer will they most likely redeem in the next 14 days?
Marketing teams blast the same offers to broad segments. A premium customer who always buys full-price gets a 20% discount they never needed, destroying margin. A price-sensitive customer gets a free-shipping offer when they actually respond to percentage discounts. Mis-targeted offers cost retailers $15-30M annually in unnecessary discounting while leaving redemption rates below 5%. The right offer to the right customer at the right time changes everything.
Quick answer
Next best offer prediction ranks promotional offers for each customer by redemption probability. Graph-based models learn from the full redemption-purchase-customer graph, capturing category-specific discount preferences, offer fatigue signals, and cross-customer redemption patterns to boost redemption rates 25-50% while cutting unnecessary discounting by 30-40%.
Approaches compared
4 ways to solve this problem
1. Segment-Based Offers
Assign offers by customer segment: VIP gets 20% off, standard gets 10% off, new gets free shipping. The default for most CRM-driven marketing teams.
Best for
Simple loyalty programs with clear segment boundaries and limited offer variety.
Watch out for
Every customer in a segment gets the same offer. A VIP who only buys full-price footwear gets a 20% off electronics coupon they will never use. Redemption rates sit at 3-5% because the matching is too coarse.
2. A/B Testing Offer Variants
Run controlled experiments testing different offers against customer segments. Measure redemption rates and roll out the winner. The gold standard for validating offer effectiveness.
Best for
Teams that want statistical rigor in offer selection. Required for measuring the true incremental impact of each offer type.
Watch out for
Slow iteration. Testing 5 offers across 10 segments requires 50 test cells and weeks of data collection. By the time results arrive, seasonal preferences may have shifted. Cannot personalize at the individual level.
3. Recommendation Engine (Collaborative Filtering)
Apply standard collaborative filtering to the customer-offer interaction matrix. Recommend offers that similar customers redeemed. Uses the same technology as product recommendations.
Best for
Platforms with dense offer-redemption data and many offer types. Leverages existing recommendation infrastructure.
Watch out for
Only sees the customer-offer matrix. Cannot incorporate purchase recency, category-specific preferences, offer fatigue, or discount sensitivity from purchase data. A customer who bought at full price 10 times should not receive the same discount offers as a coupon-driven buyer.
4. KumoRFM (Graph Neural Networks on Relational Data)
Connects customers, offers, redemptions, and purchases into a relational graph. Ranks offers per customer by learning from category-specific redemption history, purchase recency, discount sensitivity, offer fatigue, and cross-customer redemption patterns.
Best for
Retailers with diverse offer types, rich purchase data, and the goal of maximizing redemption while minimizing unnecessary discounting.
Watch out for
Requires historical offer-redemption data with customer IDs. If your promotions are store-wide discounts with no customer-level tracking, there is no individual redemption signal to learn from.
Key metric: Graph-based offer matching lifts redemption rates by 25-50% while cutting unnecessary discounting by 30-40%, driven by category-specific, fatigue-aware, discount-sensitivity signals from the relational graph.
Why relational data changes the answer
Customer C001 (Sarah Chen, Gold tier) has redeemed 4 of 5 past offers in the Footwear category. Her last Footwear purchase was 42 days ago, matching her typical replenishment cycle. A flat model might predict high redemption for any Footwear offer. But the relational graph reveals that C001 specifically responds to percentage discounts (not free shipping), has a discount sensitivity threshold of 15%+, and has a low offer fatigue index (0.12), meaning she has not been over-contacted. Meanwhile, 78% of similar Gold members in her graph neighborhood redeemed the Spring Footwear 20% offer.
These signals come from multiple tables. Category-specific redemption history requires joining REDEMPTIONS to OFFERS. Purchase recency lives in PURCHASES. Discount sensitivity is inferred from the pattern of which offer types were redeemed vs. ignored across OFFERS and REDEMPTIONS. The offer fatigue index requires counting recent offer exposure from the OFFERS table. Graph neural networks traverse all of these connections simultaneously, learning that the combination of category match + discount type match + replenishment timing + low fatigue is what drives the 0.89 prediction score. Individual signals are noisy; the compound pattern is reliable.
Segment-based offers are like a waiter who recommends the same wine to every table based on price range. Graph-based offer matching is like a sommelier who remembers you preferred Pinot Noir last time, your dinner companion ordered fish (suggesting white wine pairings), and guests at the next table with similar taste profiles loved the new Burgundy. The recommendation is specific, contextual, and far more likely to land.
How KumoRFM solves this
Relational intelligence for true personalization
Kumo ranks offers for each customer by learning from the full redemption-purchase-customer graph. It captures that Customer C001 redeems category-specific discounts but ignores free-shipping offers, while customers in C001's graph neighborhood respond to bundled deals. The model simultaneously considers offer fatigue, purchase recency, tier-based behavior, and cross-customer redemption patterns to produce ranked offer recommendations.
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
CUSTOMERS
| customer_id | name | tier | signup_date |
|---|---|---|---|
| C001 | Sarah Chen | Gold | 2023-06-15 |
| C002 | James Wilson | Silver | 2024-02-10 |
| C003 | Maria Rodriguez | Platinum | 2022-04-22 |
OFFERS
| offer_id | offer_name | category | discount_pct |
|---|---|---|---|
| OFF01 | Spring Footwear 20% | Footwear | 20 |
| OFF02 | Free Shipping Weekend | All | 0 |
| OFF03 | Bundle & Save 15% | Outdoor | 15 |
REDEMPTIONS
| redemption_id | customer_id | offer_id | revenue | timestamp |
|---|---|---|---|---|
| R001 | C001 | OFF01 | 143.99 | 2025-01-28 |
| R002 | C002 | OFF02 | 67.50 | 2025-02-05 |
| R003 | C003 | OFF03 | 289.00 | 2025-02-12 |
Write your PQL query
Describe what to predict in 2–3 lines — Kumo handles the rest
PREDICT LIST_DISTINCT(REDEMPTIONS.OFFER_ID, 0, 14, days) RANK TOP 5 FOR EACH CUSTOMERS.CUSTOMER_ID
Prediction output
Every entity gets a score, updated continuously
| CUSTOMER_ID | CLASS | SCORE | TIMESTAMP |
|---|---|---|---|
| C001 | OFF01 | 0.89 | 2025-03-12 |
| C001 | OFF03 | 0.72 | 2025-03-12 |
| C002 | OFF02 | 0.81 | 2025-03-12 |
Understand why
Every prediction includes feature attributions — no black boxes
Customer C001 (Sarah Chen, Gold tier)
Predicted: Will redeem OFF01 (Spring Footwear 20%) — score 0.89
Top contributing features
Category-specific redemption history
4 of 5 past redemptions were Footwear
33% attribution
Days since last Footwear purchase
42 days (replenishment cycle)
24% attribution
Graph neighbors' offer affinity
78% of similar Gold members redeemed
21% attribution
Discount sensitivity score
0.71 (responds to 15%+ discounts)
14% attribution
Offer fatigue index
0.12 (low — hasn't been over-contacted)
8% 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 next best offer
What redemption rate improvement can next-best-offer models deliver?
Graph-based offer matching lifts redemption rates by 25-50% compared to segment-based offers while reducing unnecessary discounting by 30-40%. The improvement comes from matching the right offer type to each customer's demonstrated preferences rather than blasting the same offer to an entire segment.
How do you prevent offer fatigue?
Graph models track offer fatigue as an explicit signal: how many offers a customer received recently, what percentage were redeemed vs. ignored, and how the response rate is trending. Customers with high fatigue scores receive fewer, more precisely targeted offers. This preserves channel effectiveness for when it matters most.
Should you offer discounts to customers who always buy at full price?
No. Customers with low discount sensitivity (who consistently buy at full price) should receive non-monetary offers: early access, exclusive products, personalized styling. Giving them 20% off destroys margin on a sale they would have made anyway. Graph models learn this from purchase history and adjust offer selection accordingly.
How many offer types should you test?
Most retailers benefit from 5-10 distinct offer types: percentage discount, dollar-off, free shipping, buy-one-get-one, bundle deals, early access, loyalty point multiplier, and category-specific offers. Graph models rank all offer types simultaneously for each customer, eliminating the need for sequential A/B testing.
Bottom line: 25-50% lift in offer redemption rates while reducing unnecessary discounting by 30-40%. Drives $10-20M in incremental annual revenue for mid-size retailers.
Related use cases
Explore more personalization 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.
