PERSONALIZED OFFERS
Next best offer: the right offer for each customer
Next best offer with TIMi ranks the offers each customer is most likely to accept, respects stock and margin constraints, and shows the reasons behind each ranking.
One model per offer
Modeler estimates the probability that each customer accepts each offer in your catalog.
Constraints included
Limited stock, margins and a fixed number of offers per customer are handled in Anatella.
Explained ranking
Each recommendation comes with the factors behind it.
HOW IT WORKS
How next best offer works.
Three steps, in one engine, on a standard server with no cluster.
1. Prepare the data
Anatella gathers purchase and subscription history, customer interactions and the catalog of available offers, then structures them into variables the model can use.
2. Build one model per offer
For each offer, Modeler builds a model that estimates the probability that a customer accepts it. The models are either one offer against all the others, or one per pair of offers.
3. Rank and allocate
Applying all the models gives each customer a probability per offer. The most likely offer comes first and the next ones remain available as alternatives. Many models can be applied in a single pass over the customer base.
WHEN IT GETS REAL
Real catalogs come with constraints.
Taking the highest probability for every customer only works when nothing limits what you can offer.
Limited stock
When the quantity of an offer is limited, the offer with the highest probability cannot go to everyone. Anatella includes an assignment solver that allocates offers to customers under such limits.
Different margins
The solver can take the margin of each offer into account, not only the acceptance probability.
A set number of offers per customer
If each customer receives exactly a given number of offers, the solver respects it. It handles campaigns with millions of customers and thousands of products.
EXPLAINABILITY
A ranking is more useful with its reasons.
A white-box model shows the factors behind the ranking for each customer, such as recent purchase behavior or similarity to customers who accepted the same offer.
For sales and marketing teams
An offer can be presented with the argument that makes it relevant to this customer, not just pushed as is.
For transparency requirements
A model that justifies each result variable by variable gives you the documentation to answer transparency expectations such as GDPR and the AI Act. The compliance assessment itself stays with your legal team.
When the catalog changes
Adding or removing an offer means regenerating the matching model with AutoML. The Drift Report tells you when existing models need a refresh.
DATA
What data a next best offer model uses.
Anatella connects to more than 80 sources, so this data can be gathered without prior custom development.
Purchase and subscription history
Products already held, frequency and amounts.
Customer interactions
Channels used and responses to past campaigns.
Offer catalog
The available offers and their characteristics.
Context signals
Stage in the customer life cycle and seasonality.
PROOF
Seen in production.
Results from our published customer stories.
Cross-sell revenue doubled in 15 days
30 product models, a campaign to 2 million customers and an offer allocation engine, built by 2 data miners on one standard desktop PC.
+10% store visits from personalized coupons
Offers matched to each shopper’s purchase history beat alternative models on store visits.
Conversion up from 4% to 34%
Models built per product rather than per product category are now the standard at BNP Paribas.
FAQ
Common questions.
What is the difference between next best offer and next best action?
Next best offer recommends a product or a commercial offer. Next best action is broader: it can also recommend a non-commercial step, such as a retention contact or a service action, depending on the customer’s situation.
Does the model recommend one offer or a ranking?
In general a ranking. It lets you propose an alternative when the first offer is unavailable or was already declined, without rebuilding the model.
Do we need a rich history for each customer?
Individual history helps, and the models also rely on customers with similar profiles, so a recent customer can still receive a relevant recommendation.
What happens when the catalog changes?
Adding or removing an offer means updating the corresponding model. Modeler regenerates models quickly with AutoML, and the Drift Report shows when existing ones need a refresh.
Can it handle limited stock or different margins?
Yes. When quantities are limited or margins differ, the assignment solver in Anatella allocates offers to customers under those constraints instead of simply taking the highest probability.
RELATED USE CASES
Keep exploring.
Next Best Action
Choose the best next step per customer: an offer, a retention contact or a service action.
Cross-Sell Prediction
Find the customers most likely to buy a complementary product, and why.
Churn Prediction
Score every customer on the risk of leaving and see the variables behind it.
SEE FOR YOURSELF
Test it on your own data.
Run TIMi on your own data with a free trial, or talk to us about your case.
