Multicloud AI & ERP / Cloud AI: Machine Learning & Generative AI
Why it matters
Personalized results lift conversion, basket size and engagement. Generic "most popular" lists leave value on the table, while poor recommendations annoy customers. The difference is usually data freshness and business rules, not the algorithm.
How it works
From events to results
User events such as views, clicks and purchases flow in, ideally in real time. A candidate model finds items a user may like, a ranking model orders them by predicted value, and business rules apply stock, margin and policy before results appear.
Measure with experiments
Offline metrics guide development, but only A/B tests prove business impact. Every change to models or rules should be tested on live traffic.
AWS, Azure and Google Cloud compared
| AWS | Azure | Google Cloud | |
|---|---|---|---|
| Managed recommender | Amazon Personalize | Azure ML recommendation models | Vertex AI Search for commerce |
| Real-time events | Personalize event tracker | Event Hubs | User events API |
| Search ranking | OpenSearch with Personalize | AI Search scoring profiles | Personalized search for commerce |
| Business rules | Personalize filters | Scoring profiles, custom logic | Serving controls |
| SQL option | Redshift ML | Fabric data science | BigQuery ML matrix factorization |
What makes each cloud different
AWS
Amazon Personalize packages the technology behind Amazon.com recommendations as managed recipes, with real-time event tracking, filters for business rules and re-ranking of OpenSearch results.
Azure
Azure builds recommendations with Azure ML on data in Fabric, serves them through managed endpoints and uses AI Search scoring profiles to boost results by freshness, margin or popularity.
Google Cloud
Vertex AI Search for commerce brings Google-quality product search and recommendations for retailers, with serving controls that let merchandisers boost, filter and redirect.
Design decisions that matter
- 01Stream events in real time
Fresh signals beat nightly batches.
- 02Keep business rules explicit
Stock and policy filters outside the model.
- 03A/B test every change
Measure revenue, not just click-through.
- 04Start managed, then customize
Your own signals add the edge.
Common mistakes we see
- Recommending out-of-stock items. Frustrates customers and wastes slots.
- Optimizing clicks only. Clickbait recommendations hurt revenue.
- No cold-start plan. New users and items need sensible defaults.
A practical 30-day plan
- Week 1
Instrument real-time user events.
- Week 2
Launch a managed recommender on one page.
- Week 3
Add business rules and filters.
- Week 4
Run an A/B test against the current experience.
How NTS helps
NTS delivers recommendation and search personalization on AWS, Azure and Google Cloud, from event tracking and model setup to business rules and A/B testing tied to revenue.
What you get
- A recommendation strategy for your products and users
- Candidate generation and ranking models
- A/B testing to prove impact
- Real-time personalization pipelines
Typical outcomes
- Higher conversion and engagement
- Personalized experiences at scale
- Clear evidence of business impact
How we work with you
- 1Assess
A focused review of your recommendations and ranking across AWS, Azure and Google Cloud. We look for the gaps we find most often: recommending out-of-stock items, optimizing clicks only and no cold-start plan.
- 2Design
A target architecture and roadmap built on the decisions that matter most: stream events in real time, keep business rules explicit, a/B test every change and start managed, then customize.
- 3Build
Hands-on delivery alongside your team. We instrument real-time user events, launch a managed recommender on one page, add business rules and filters, and finally run an A/B test against the current experience.
- 4Run and improve
We retrain models, run ongoing experiments and tune ranking as catalogs and users change.
Why teams choose NTS for recommendations and ranking
- Revenue-focused
We measure recommendations by conversion and engagement.
- Proven by experiments
A/B tests show real impact before full rollout.
- Managed or custom
We choose managed services or custom models based on your needs.
Book a free 30-minute review
Talk to an NTS architect about your recommendations and ranking on AWS, Azure and Google Cloud.