E-commerce Recommendations
Best-performing model in a 3-vendor pilot – based on story similarity.
Recommendations based on narrative similarity, not just sales data.
#1 · of 3 vendors · The Problem
What needed to change
- Traditional recommendation systems typically rely on purchasing behavior (the "customers who bought X also bought Y" model).
- They do not consider the content of the books or subtle reader preferences (such as narrative style, story pacing, or mood).
- Martinus was looking for an AI solution that would recommend books based on story similarity, not just purchase data.
The solution was developed as part of a pilot competition between three vendors, aiming to identify the most effective approach.
The Solution
What we implemented
An AI model was developed that analyzes story descriptions and identifies thematic and plot similarities between titles.
- Identifies thematic and plot similarities between titles.
- Recommends books with a similar atmosphere, pacing, or setting.
- Now expanding to work with full book texts, not just publisher-provided descriptions.
- Extracts key elements of plot, characters, time, and setting that shape the reading experience.
The resulting recommendations operate with a much broader data context than traditional algorithms.
The Results
What we achieved
- The solution delivered the best recommendations according to evaluations from both the client and their customers.
- Development of a solution that offers reader-relevant recommendations, not just statistical matches.
- A step toward AI systems capable of truly "understanding" literature – including the ability to suggest similar stories based on complex elements like theme, tone, and writing style.
- Expected impact: increased conversion rates from recommendations, longer time spent on site, and higher customer satisfaction with book selection.
More case studies
Document Processing
Dôvera · Insurance
Enforcement document processing: 60% back-office time saved.
100% of documents processed automatically. Senior staff freed during peak periods.
60% back-office time saved
Receivables Processing
Dôvera · Insurance
Receivables write-off review: from 2 weeks to ~1 hour.
Senior staff were freed up during peak periods. Decisions became consistent and auditable.
2wk→1h processing time · Data Extraction · Dôvera · Insurance
Medical leaflet extraction: structured dosage data from unstructured text.
New data source created for prescription audits. Previously invisible information made operational.
New audit data source created
A similar process in your organization?
Tell us about your workflow. We'll tell you whether AI fits – honestly, even if the answer is no.