Next Best Offer Engine · AI-powered product recommendations
Know the one product every customer should be offered next
Aryng's Next Best Offer Engine uses machine learning to predict which product each customer is most likely to adopt, weighs that propensity against the profit each product brings in, and scores the result straight into your database, so marketing, sales and support all lead with the offer most likely to convert and pay off.
Proven with a Fortune 500 payments company
The problem
More products should mean more revenue, not more unsubscribes
Customers who use more than one of your products are almost always more profitable, so the natural instinct is to market every product to every customer. That blanket approach backfires: customers get buried in offers that have nothing to do with how they use you, tune out, and eventually unsubscribe, which shrinks the very audience you were trying to grow.
Blanket marketing
Every product, sent to every customer
- Generic campaigns with low conversion
- Offer fatigue, confusion and rising unsubscribes
- Budget spread across customers who will never adopt
- Sales and support pitching whatever is top of mind
Next best offer personalization
One recommended product, chosen for each customer
- Offers ranked by adoption propensity and profit
- Fewer, more relevant messages per customer
- Budget concentrated where it converts
- One model score shared by outbound, inbound and support
Cross-selling works when the second product fits the customer. AI makes it possible to work out that fit for every customer at once.
What is a next best offer model?
A recommendation that balances what each customer wants with what each product earns
A next best offer (NBO) model, also called a next best product (NBP) model, uses machine learning to predict the product each customer is most likely to adopt next. Most product recommendation engines stop there, however, the most likely purchase is not always the most valuable one. Aryng's approach multiplies each product's predicted adoption probability by its incremental profit to get an expected incremental profit (EIP), and the product with the highest EIP becomes that customer's next best offer.
Customer focus
Adoption probability
How likely is this customer to take up each product?
Financial focus
Incremental profit
What is each product worth if they do?
Optimize both
Expected incremental profit
The highest EIP is the next best offer.
One next best offer score, used across the customer lifecycle
Target outbound campaigns
Send each customer the email offer the model ranks highest for them, not the whole catalog.
Personalize inbound
Feed the same score into your website, product and in-app messaging.
Upsell on support calls
Give support agents the one product worth mentioning to the customer on the line.
Cut wasted spend
Stop paying to reach customers with offers they were never going to adopt.
Test and learn
A/B test model-driven offers against business as usual and refine with every campaign.
How it works
Aryng's Next Best Offer Roadmap
Aryng's SWAT data science team runs every engagement on BADIR, its proprietary Data-to-Decisions framework, which keeps the machine learning work tied to a marketing decision from the first stakeholder session to the model scored in your database.
-
B
Business question
We use the Business Question framework to define the real problem: the next best product for every customer, optimized for adoption as well as profitability.
-
A
Analysis plan
We brainstorm hypotheses with your product, customer support and marketing teams, then agree an analysis plan covering the data, hypotheses and modeling methodology.
-
D
Data collection
We extract transactional, account, customer support and online usage data from your data warehouse against a written data specification.
-
I
Insights
Multiclass decision trees find the segments that adopt each product, and adoption probability is combined with historical profit to rank offers by expected incremental profit.
-
R
Recommendations
Channel-by-channel recommendations for putting the model to work, with a test-and-learn plan so each campaign sharpens the next.
Case study · Fortune 500 online payments company
How Aryng built an AI product recommendation engine that drove $20M+ in profit and cut marketing spend by 70%
About the client
Our client is a Fortune 500 online payments company that enables digital and mobile payments on behalf of consumers and merchants worldwide. It offers several payment products to fit the needs of its merchants, who are its customers.
Merchants who used more than one product showed wider and deeper product usage, which made them more profitable. As a result, every product was marketed to every merchant, which confused them and led to unsubscriptions. The head of product wanted a smarter way to decide what each merchant should be offered.
The real business question
Find the next best product (NBP) recommendation for each merchant, optimized for adoption as well as profitability.
Optimizing for adoption alone favors popular, low-margin products. Optimizing for profit alone favors products merchants will never take up.
The analysis plan
Stakeholder alignment
We brainstormed with the product, customer support and marketing teams to narrow down the hypotheses on each merchant's next best product.
Analysis plan
Drafted with the consensus of every stakeholder, setting out the required data based on the goal, the hypotheses and the methodology.
The data
Transactional, account, customer support and online usage data, extracted from the data warehouse based on the data specification.
Methodology
A multiclass decision tree classifier for each base product, with adoption probabilities combined with historical profit per segment.
Why decision trees
A decision tree is explainable AI by design: every recommendation traces back to a clear set of merchant characteristics, so marketing and product teams can see why a segment gets the offer it gets.
What the model found
For each base product, the multiclass decision tree split the merchant base into segments based on which other products they had adopted. Some leaf nodes were dominated by non-adopters, while others held clear pockets of adopters of particular products.
Each segment came out with its own adoption probability for every product. Multiplying those probabilities by each product's historical incremental profit gave the expected incremental profit (EIP) per product per segment, and the product with the highest EIP became the next best product for every merchant in that segment.
From model to marketing
Inside one segment: merchant segment 39
Zooming into a single leaf of the tree shows how the recommendation is made: products A and D carry almost the same profit per adopter, however, once adoption probability is factored in, product E pulls far ahead of everything else.
| Product | Next product probability | Incremental profit | Expected incremental profit (EIP) |
|---|---|---|---|
| A | 2% | $234 | $4.68 |
| B | 0% | $0 | $0.00 |
| C | 4% | $127 | $5.08 |
| D | 3% | $232 | $6.96 |
| E · Next best product | 23% | $266 | $61.18 |
| No next product | 69% | $0 | $0.00 |
Product names anonymized. Max EIP = $61.18, so product E is the next best product for merchants in segment 39.
Our recommendations
-
01
Run an email campaign on the NBP
Send each merchant the single offer the model recommends for them.
-
02
Upsell during support calls
Give customer support the NBP for every merchant who calls in.
-
03
Test and learn
Run controlled tests on the recommendations and feed what you learn back into the model.
Results
Results from the outbound marketing campaign
The model was put to work upselling product A. Merchants the model flagged as NBP = A were compared with merchants whose next best product was something else, and the model proved itself on conversion as well as volume.
conversion rate for model-targeted merchants, 3.14% against 0.56%
volume on product A per converted account, $566 against $388
incremental volume per converter compared with non-converters
Next best offer model effect on product A upsell
| Segment | Targets | Conversions | Conversion rate |
Volume on A |
Volume on A per account |
|---|---|---|---|---|---|
| NBP = A | 143,774 | 4,511 | 3.14% | $2,551,564 | $566 |
| NBP ≠ A | 200,000 | 1,118 | 0.56% | $433,493 | $388 |
| NBP model effect | 561% | 146% | |||
Converters vs non-converters (NBP = A)
| Group | Accounts | Total volume |
Volume per account |
|---|---|---|---|
| Converted | 4,511 | $5,265,858 | $1,167 |
| Non-converted | 139,263 | $84,443,006 | $606 |
| Incremental volume from converters | $561 | ||
Model effect is the NBP = A rate divided by the NBP ≠ A rate. Volume figures are payment volume.
How BADIR turned a crowded product catalog into one profitable offer per merchant
With the next best product recommendations in hand, the marketing team finally knew which offer to send to whom, which cut marketing spend by 70%. Offering the right products using the NBP score delivered about six times the conversion and more than $20M in profit from the outbound marketing effort alone.
BADIR kept the work anchored to the business question from the start. Bringing product, support and marketing into the analysis plan meant the model answered the question they actually had, and pairing adoption probability with profit meant the recommendations were ones the business could commit budget to, rather than a propensity score that looked good on paper.
Free download
Get the complete next best offer case study
The full PDF walks through every step of the BADIR engagement, including the multiclass decision tree, the segment-level EIP math and the outbound campaign results.
Outcomes
What to expect from the Next Best Offer Engine
A next best offer for every customer
A machine learning recommendation for each customer in your base, built from your own transactional, account, support and usage data.
Offers ranked by profit, not popularity
Every recommendation weighs adoption probability against incremental profit, so the offer you lead with is the one worth the most.
Explainable segments
Decision-tree segments that show which customer traits drive each recommendation, so stakeholders trust the model and act on it.
Scored where your teams work
The model scored in your database and ready for email, CRM, website personalization and the support desk.
Leaner marketing spend
Fewer, better-targeted offers per customer, which means less wasted reach and fewer unsubscribes.
Pair it with LTV and CAC
Combine next best offers with the LTV Predictor to prioritize your most valuable customers, and the CAC Optimizer to see what each one costs to acquire.
Frequently asked questions
What is a next best offer model?
A next best offer model uses machine learning to predict which product or offer each customer is most likely to adopt next, so you can personalize what you show them instead of marketing your whole catalog to everyone. Aryng's version also factors in the profit each product generates, so the recommendation is both likely and valuable.
How is this different from a standard product recommendation engine?
Most recommendation engines optimize for the likelihood of a purchase or click. That tends to push popular products, which are not always the profitable ones. Multiplying adoption probability by incremental profit gives an expected incremental profit for every product, and the highest one wins.
Why use decision trees rather than a black-box model?
Decision trees are explainable AI: each segment is defined by traits your teams can read, which makes the recommendations easy to trust, audit and act on. Where accuracy matters more, the same framework can be extended with gradient-boosted machine learning models.
Which channels can use the recommendations?
Any channel that can read a score from your database. In the case study the model powered outbound email campaigns and inbound marketing, and was recommended for upselling during customer support calls.
How does this work with the LTV Predictor?
Predicted LTV tells you which customers are worth the most over time, and the next best offer tells you what to put in front of each of them now. Together they show where personalization effort and marketing budget will pay back the most.
Find out what each customer should buy next
Talk to Aryng's data science team about building an AI-powered next best offer model on your own customer data.
About Aryng
Aryng is a data science and AI consulting, training and advising company. Aryng's SWAT data science team solves complex business problems, builds each company's Data DNA through data literacy programs, and delivers rapid ROI using machine learning, deep learning and AI. Clients include Google, Box, HERE, Applied Materials, Abbott Labs and GE.
For Corporations
For Indiviuals
Aryng
Policies
Copyright © 2011 - 2026 Aryng LLC. All rights reserved
