Recommend relevant products and content to your customers.
Connect your catalog and interactions across your channels to offer recommendations based on popularity, trends and affinity. With Anteater, our own technology, your team configures business rules and measures the performance of each carousel.
Better recommendations need context, visibility and measurement.
- 01
Suggestions with limited context
A general list can be useful, but it does not account for what someone viewed, the product they are exploring or the channel they use. Combining these signals makes it possible to evaluate more relevant recommendations.
- 02
New arrivals need visibility
New products and content do not yet have a usage history. Their attributes, descriptions and images make it possible to assess similarities and find places to recommend them.
- 03
Results need tracking
Recording impressions, clicks and conversions by carousel helps your team understand how recommendations perform and compare them with the current experience.
Catalog data, behavior and business rules guide what to recommend.
The engine combines your catalog information with events such as views, clicks and purchases. Depending on the available data, it uses popularity, trends, similarity and affinity to suggest products or content. We agree on business rules and measurement criteria with your team.
- 1
Connect catalog and events
Receive products or content from your database, PIM or file, and interactions from integrated channels. We define the sources and event schema before starting.
- 2
Prepare the catalog for recommendations
Use attributes, text and images to represent similarities between products or content. AI enrichment is defined according to the information available in the catalog.
- 3
Calculate usage signals
Recalculate metrics such as reach, usage, retention and momentum daily over 7-, 14- and 28-day windows, depending on the selected methods.
- 4
Combine methods and rules
Use the methods your data supports and apply priorities for inventory, seasonality, margin, private labels or exclusions. We evaluate their effect on recommendation relevance.
- 5
Publish to integrated channels
Deliver recommendations through an API to the agreed placements: home page, product page, cart, app, email or other channels included in the project.
- 6
Configure and measure
Your team manages carousels, algorithms, positions and available filters from a console. The dashboard provides clicks, conversion and catalog coverage by carousel.
Our technology · Anteater
Anteater is Amarello's search and recommendation technology. For recommendations, it combines product and user representations, precomputed metrics and business rules. Intelligent search can be added as a separate solution on the same platform.
One engine, many ways to recommend.
We select methods based on the purpose of each placement and the available data. Together, we define how they are combined, which rules apply and which metrics will be used to evaluate them.
Popularity and trends
Identify frequently used products or content and changes in demand from aggregated events. Individual user history is not required, but there must be enough data to calculate the selected metrics.
- Home page
- Most viewed
- Aggregated events
Similar items and new arrivals
Compare attributes, text and images to recommend alternatives and new arrivals related to a product, content item or available interest profile. New items can be evaluated without their own interaction history.
- Product page
- New arrivals
- Alternatives to out-of-stock products
Affinity between users
Use interaction patterns from users with similar histories to generate suggestions such as “also bought” and “also viewed.” This method is introduced when enough data is available.
- Cart
- Also bought
- Bundles
Recent activity and context
Use recent interactions and the channel to guide suggestions. The scope depends on the events received and the update frequency defined for the integration.
- Session
- Recently viewed
- App
Next relevant offer
Prioritize product, coverage or plan offers based on the profile, available behavior and business rules defined for each use case.
- Banking
- Insurance
- Cross-sell
Audiences by affinity
Describe audiences in natural language and combine them with recency, frequency and value criteria when that data is available. Your team reviews segments before using them in campaigns.
- Campaigns
- CRM
- Dynamic segments
Recommendations shaped by the channel and the business goal.
These are examples of possible applications. For each project, we define what to recommend, where to display it, which data is needed and how to measure performance. Channels and metrics are selected according to the available integration.
Retail and e-commerce
- Goal
- Help customers discover products and evaluate cross-selling opportunities.
- What it recommends
- Similar products, “also bought,” bundles and suggestions based on affinity.
- Possible channels
- Home page, product page, cart and email.
- Metrics to define
- Clicks and conversions from recommendations; average order value when purchase information is available.
Banking
- Goal
- Present offers and content related to the customer's profile and activity.
- What it recommends
- Commercial offers prioritized according to the bank's rules, and related educational content.
- Possible channels
- App, online banking, contact center and WhatsApp.
- Metrics to define
- Offer clicks and uptake rate when that outcome can be tracked.
Insurance
- Goal
- Surface relevant complementary coverage, prevention content and benefits.
- What it recommends
- Suggestions for coverage, prevention and benefits based on available policy information and history.
- Possible channels
- Policyholder portal, agent channel and email.
- Metrics to define
- Cross-sell, renewal and benefit usage, according to the integrated data.
Entertainment
- Goal
- Help users discover content and evaluate continued usage.
- What it recommends
- Popular content, trends, new arrivals by affinity and items similar to those explored.
- Possible channels
- Home screen, section carousels and app.
- Metrics to define
- Usage per session, weekly retention and reach of new releases.
Media and content
- Goal
- Surface articles and videos related to the reader's interests and activity.
- What it recommends
- Related content based on topic, affinity and session context.
- Possible channels
- Website, app and newsletter.
- Metrics to define
- Time on site, pages per session and return visits.
Marketing and CRM
- Goal
- Build audiences by affinity and evaluate campaign response.
- What it recommends
- Segments defined in natural language and combined with available signals.
- Possible channels
- Campaign tools and integrated workflows.
- Metrics to define
- Clicks, conversion, unsubscribes and complaints when the campaign platform provides that data.
Methods defined with your team, configurable rules and measurable results.
- Integration with your channels
- API connections to your catalog and recommendation placements, with a review of the interfaces and adaptations needed for your platform.
- Methods and evaluation
- Definitions, formulas and weights agreed with your team. Evaluation against the current carousel through an A/B test with defined traffic and metrics.
- Control for your team
- Configure carousels, algorithms, positions and available filters from a console, without code changes for those settings.
- Operations and support
- Daily recalculation, owners for the catalog and events, and operations under agreed service levels.
- Continuity and handover
- Export of the enriched catalog, vectors, rules and dashboards, with documentation and handover conditions defined in the contract.
Anteater in operation.
popularity, trends, similarity and affinity, according to the available data
metrics over 7-, 14- and 28-day windows, depending on the selected method
comparison with the current carousel to evaluate performance with real traffic
we have developed recommendations on the same technology as our search solution
Recommendations and search · Retail
Supermarket chain in Mexico
Recommendations and search across the online store's full catalog, with business rules configurable by the chain. Anteater also ran on the marketplace of a retail group.
Search with Anteater · Banking
Two of the largest banks in Mexico
Anteater runs on the digital channels of two of the country's largest banks, providing semantic and exact search connected to their content and rules. This case concerns the search solution built on the shared technology.
Companies Amarello has worked with
We move phase by phase, with verified results.
We start with one channel and one recommendation placement. We review the catalog and events, select the methods the data can support, and agree on how to compare results before expanding the scope.
- 1
Scope and data
- What we do
- Select the channel and carousel, review catalog and events, and define the comparison with the current experience.
- Your team's role
- Access to a sample of catalog data and events, available analytics, and business and technical owners.
- Criteria to proceed
- Agreed sources, goal, metrics and initial scope.
- 2
Catalog and methods
- What we do
- Integrate the catalog, define enrichment and configure methods, formulas, weights and rules. Verify which events each method requires.
- Your team's role
- Validate definitions and data, and coordinate with the platform provider where needed.
- Criteria to proceed
- Integrated catalog, verified events and recommendations reviewed against the defined criteria.
- 3
Pilot and comparison
- What we do
- Activate a share of traffic and compare with the current carousel through an A/B test. Use the methods supported by the available data.
- Your team's role
- Pilot traffic and a business owner to review results and priorities.
- Criteria to proceed
- Verified measurement and results reviewed over the agreed traffic and period to decide whether to expand, adjust or continue evaluating.
- 4
Operations
- What we do
- Expand to the agreed channels and traffic, with daily recalculation, metric tracking and defined service levels.
- Your team's role
- Owners for the catalog, events and business rules.
- Criteria to proceed
- Agreed responsibilities, support terms and operating documentation.
Engine, integrations, data and deployment.
Integration
An asynchronous catalog ingestion API accepts CSV or JSON with a tracking ID. Events are received through a queue, API or daily file. The recommendation API connects to the agreed placements on your website, app or other channels. The technical assessment determines the interfaces and adaptations needed. Search has its own API when that solution is included.
Engine
Catalog enrichment with language models, vector representations of products and users, time-window aggregations and scoring by algorithm. Methods are combined using weights and rules defined for the use case, according to the available data.
Data
Analytical calculations run in daily batches, with precomputed results available to integrated channels. Windows of 7, 14 and 28 days are used according to the selected metrics. On Google Cloud, the architecture uses BigQuery and Firestore. Event intake frequency and response time are verified during integration and the pilot.
Identifiers and events
The minimum schema has five fields: user identifier, product identifier, event type, date and channel. The client retains the mapping between the user identifier and their identity. The minimum schema does not require names, email addresses or purchase amounts; additional data needed for other measurement goals is defined separately.
Configuration
The console manages carousels, algorithms, positions and available filters, together with the business rules supported by the configuration. These settings can be changed without modifying code; new behaviors or integrations are assessed within the project scope.
Measurement
Impression, click and conversion events are recorded by carousel to populate a dashboard with clicks, conversion and catalog coverage. The definition of conversion and its attribution are agreed before the pilot. Monetary metrics and other business indicators require the corresponding additional sources.
Deployment
Google Cloud Platform (GCP) is our primary runtime environment. Availability on Amazon Web Services (AWS) or private infrastructure is confirmed during the technical assessment, based on the environments validated for the solution.
Data export and handover
The contract defines export of the enriched catalog, vectors, rules and dashboards, the documentation to be delivered and the conditions for handover to the client's team.
What we get asked before we start.
Do we need user history to start?
It depends on the method. Similarity can use catalog attributes without each product having its own history. Popularity and trends require enough aggregated events, but not individual user history. Behavior-based personalization is introduced when enough data is available. At kickoff, we define which methods your data supports.
Which events do we need to send?
The minimum schema includes user identifier, product identifier, event type, date and channel. Your team retains the mapping between the identifier and the user's identity. We provide the specification and receive events through a queue, API or daily file. For average order value or other monetary metrics, we review the required information separately.
Can we decide what gets recommended first?
Your team can configure inventory, seasonality, margin, private-label and exclusion rules from the console. We define how they combine with each method and evaluate their effect on relevance and the agreed metrics.
Does it work with our e-commerce platform?
The solution connects through APIs to your catalog and the placements where recommendations appear. We review available interfaces, events and necessary adaptations with your team or provider to define the integration with your current platform.
How often are recommendations updated?
Analytical calculations run daily. Event intake and availability to each method depend on the defined integration. We agree on these frequencies during the project, particularly when recent interactions or session context are used.
How do we evaluate whether recommendations perform better?
We define the primary metric, verify measurement and compare the new carousel with the current one through an A/B test. We review clicks, conversion and catalog coverage, together with any additional indicators your data supports. Traffic and the evaluation period are agreed before interpreting the results.
Is this the same as Intelligent search?
They are two solutions built on Anteater. Search responds to a query; recommendations suggest products or content based on the catalog, interactions and rules, without requiring an explicit query. They can be implemented separately or together and share an enriched catalog, signals and dashboard, according to the project scope.
How does Amarello support continuity and handover?
The contract defines export of the enriched catalog, vectors, rules and dashboards, the documentation to be delivered and the conditions for handover to the client's team.
We also solve
- Intelligent searchCombine semantic search and exact matching to find products and content, with ranking rules and usage metrics. A separate solution powered by Anteater.See how we work on it
- AI assistantsAssistants on WhatsApp, web and voice that consult your sources and take actions in your systems, according to the defined permissions and scope.See how we work on it
Let's test which recommendations work for your customers.
We start with one channel, one carousel and the available data. We define what to recommend, how to measure it and how to compare results with your current experience before expanding the scope.