i6 Decision Suite — home

Relevance

Every interaction, the right decision.

Relevance decides what to present in every interaction based on your catalog, observed behavior and the context of the moment. The decision comes out ready to be executed by your channels.

Subscribe nowBook a demoSee how it works

Model

A model that understands your business, tailored to your catalog

Before recommending anything, Relevance needs to know which market your operation works in. That choice defines the vocabulary, the categories and how the catalog is interpreted — the same product reads a fashion operation one way and a consumer goods operation another.
  • Each domain, a different vocabulary
  • One choice, shared across the entire company
  • The same catalog, read a different way
Model choice
  • Fashioncollection, size grid, styleselected
  • Consumer goodsbrand, pack, turnover
  • Electronicsspecification, generation
  • Grocerybasket, recurrence
Product illustration. It does not represent data from any operation.

Data intake

From catalog to first recommendation, without friction

The catalog arrives by file or integration and is validated field by field before it counts for decisions. You see coverage per field, quality warnings and a preview of how Relevance interpreted the content you sent.
  • no history required
  • file is not stored
  • Validated before it becomes a recommendation
  • See how it was interpreted before confirming
Catalog validation
  • 12.480records received
  • 12.206valid
  • 274with warning
Illustrative coverage per catalog field.
fieldcoverage%
title99%
category94%
description81%
attributes.color72%
Product illustration. It does not represent data from any operation.

Integration

One API call, results in production

The operation consumes decisions through API, SDK or server to server. The implementation journey is described step by step, with call examples and published technical reference.
  • One step at a time, with no guesswork
  • One call, a production-ready response
Implementation journey
  1. Credentialskey per environment, no secret in the client
  2. First calla decision in one request
  3. Result feedbackobserved events flow back into the model
# call example
POST /v1/decisions
{ "session": "s_9281", "placement": "pdp" }
# → 200 · decision ready for the channel
Product illustration. It does not represent data from any operation.

Measurement

You don't have to trust — you measure

The product shows what is happening to the decisions it delivers: volume, response, how much of the catalog is reached and how the model behaves technically. No black box.
  • Revenue, orders, ticket — always in view
  • Where the journey loses momentum
  • Technical rigor, not just outcomes
Measurement dashboard
  • R$ 1,8Minfluenced revenue+12%
  • 4.214influenced orders+9%
  • R$ 427influenced average ticket+3%
  • 68%catalog coverage+5 p.p.
  • exposed · 100%
  • clicks · 35%
  • cart · 18%
  • purchase · 10%

Observed difference

Difference between the site's normal pace and the engagement pace with Relevance.

  • Click+34.6%
  • Add to cart+18.1%
  • Purchase+22.7%
Product illustration. It does not represent data from any operation.

Insights

The part of the business nobody is looking at

Beyond deciding, Relevance gives back what it observed: prioritized behavior patterns, heavily exposed items with weak response, nearly invisible items with strong response and parts of the catalog that almost never appear. Each finding describes measured behavior, not a cause.
  • One finding, backed by evidence
  • An opportunity nobody had asked for
Prioritized findings
  • High exposure, low responsehigh

    A group of items shows up frequently and converts below what the category would suggest.

  • Affinity between itemsmedium

    Two groups appear together in the same session more often than average.

  • Nearly invisible catalogmedium

    Part of the catalog is rarely presented, even though response is positive when it appears.

Product illustration. It does not represent data from any operation.

Learning

The model never stops learning

Behind every recommendation is a proprietary model pre-trained on billions of behavioral signals and calibrated to your catalog. The architecture combines product representation, journey sequencing and context signals — all continuously evolving.

Every interaction and every observed result flows back into the model. Tomorrow's decision already accounts for what happened today, with no manual reprocessing cycle.

Learning cycle
  • control interactions
  • interactions with the model
  • result with the model
  • control result
  • 4.7×result vs. control
  • +54%additional interactions
  • 23.8%Long-tail Contribution
Product illustration. It does not represent data from any operation.

Stop recommending in the dark

See Relevance deciding over a real catalog, with measurement and insights from the first week.

Subscribe nowBook a demo