Feature auto-discovery

Overview

Beyond applications, Kabeen AI identifies the business features actually being used in every application: order management, invoice tracking, planning, ticket handling and so on.

This analysis moves you from a purely declarative inventory to a real map of the functional capabilities covered by your information system.

How it works

1. Kabeen agents aggregate the navigation patterns observed on each application: URL paths, page titles, visit frequency, number of unique users.
2. These patterns are forwarded to a language model hosted by Mistral, our French AI provider.
3. The model matches these usage paths against your organization's functional capability catalog (HR, finance, supply chain, marketing and so on) and proposes the features most likely to be in use.
4. Each feature comes with a confidence score (0 to 100%) computed from the number of converging signals (URL families, usage volume, teams involved).

The agent consistently favors business capabilities (Order-to-Cash, Procure-to-Pay, leave management...) and excludes technical modules (authentication, administration, settings).

Reviewing discovered features

From an application's page:

1. Open the Features tab
2. Identified features are displayed with:
   • The name of the functional capability linked to your catalog
   • The confidence score
   • The associated teams (inferred from the users who executed those paths)
3. You can validate (✓) or reject (✗) each proposal.

Validated features automatically feed your Functional Blueprint and the mapping reports.

Analysis frequency

Analysis is scheduled automatically for every eligible application as soon as it has sufficiently representative URL patterns. Proposals get richer over time as real usage is confirmed.

Best practices

• Populate your organization's functional capability catalog before the first analysis: proposal quality depends directly on the richness of the reference framework.
• Validate or reject proposals as you go: this anchors the Blueprint and improves the precision of later analyses.
• Trust the confidence score: a confidence > 80% means several converging URL families; a confidence < 50% warrants a human review.

Privacy

Only navigation metadata (URLs and page titles) are sent to the AI model, after strict filtering of potentially sensitive strings. No page content, no user-entered data and no personal information leaves your environment. Processing is operated by Mistral, our French AI provider, on a GDPR-compliant infrastructure.