The Knowledge Graph that knows 'SFDC' & 'Salesforse' is actually just Salesforce

In part four of our series "Closing the IT Data Trust Gap: From Raw Records to Decision‑Grade Intelligence" we take a look at how Asato’s asset knowledge graph handles entity resolution, enrichment, and how it powers continuous learning of the intelligence layer in IT asset optimization.

In the previous posts in this series, we described two problems: inflated user counts that don’t match headcount, and inflated app counts full of infrastructure noise. Both problems share a root cause: raw data that hasn’t been resolved and enriched against a curated knowledge base. And both problems block the thing IT leaders actually want: the ability to optimize their technology assets for cost, risk, and business value.

This post goes inside the engine that does that work: Asato’s Asset Knowledge Graph — the Think layer in our Link → Think → Execute architecture.

The Naming Problem

Consider a simple question: how many organizations use Microsoft 365? To answer it, you first need to recognize that “Microsoft Office 365,” “M365,” “Office365,” and “office.com” all refer to the same product. You need to know that “Salesforse” in a procurement spreadsheet is a typo for Salesforce, and that “MSFT Teams” means Microsoft Teams.

This is entity resolution, and it’s far harder than it sounds. Simple string matching fails on abbreviations, typos, internal product codes, and ambiguous names. “Slack” could mean the Salesforce product or a generic English word.

Multi-Signal Entity Resolution

Asato’s MKG resolves entities using four complementary techniques:

• Fuzzy logic and phonetic search; pattern matching that handles typos, abbreviations, and phonetic similarity (“Salesforse” → Salesforce)

• Vector-based semantic similarity; embedding-based matching that understands conceptual relationships (“Jira Service Management” matched to Atlassian’s product catalog)

• Knowledge base embeddings; purpose-built embeddings trained on the MKG’s entity corpus for high-precision disambiguation (“Slack” the product vs. “Slack” the generic term)

• Normalized scoring and aggregation; multiple resolution signals combined into a confidence-weighted match score (a high-confidence match from 3 techniques vs. a weak match from 1)

This multi-signal approach means entity resolution works reliably even when source systems use inconsistent naming conventions, abbreviations, or internal product codes.

From Resolution to Enrichment

Entity resolution tells the platform what an entity is. Enrichment tells the platform everything else about it.

Once resolved to a canonical record, the MKG immediately attaches rich metadata: application type (user-facing, CDN, auth endpoint, infrastructure), business function (Collaboration, IT & Security, Engineering), vendor and product family relationships, compliance profiles (SOC 2, HIPAA, PCI-DSS), and risk and security postures aggregated from multiple providers.

This is what makes the identity and application funnels from Posts 2 and 3 possible. Without enrichment, you can’t classify. Without classification, you can’t filter meaningfully. And without meaningful filtering, every dashboard shows noise.

Provenance: Not All Signals Are Equal

One capability that sets the MKG apart is provenance tracking. When the same application appears in an ERP contract, CASB logs, and a credit card statement, most platforms treat all three signals equally. The MKG doesn’t. It prioritizes authoritative sources, signed contracts and direct integrations over circumstantial evidence like network traffic and browser extensions. Without this hierarchy of trust, data quality is a black box.

A Knowledge Graph That Gets Smarter Over Time

The MKG is not a static database. It’s continuously updated through multiple automated mechanisms:

• Automated sweepers scan vendor websites, app marketplaces, and regulatory databases to identify newm applications, product changes, and updated compliance certifications

• Augmentation pipelines research entities discovered in customer environments that don’t match existing records, expanding coverage for all customers

• AI-driven classification validates existing entries and classifies new ones, combining broad language models with purpose-built models for high-speed, high-precision tasks

• Entity embeddings are regenerated as the knowledge base grows, improving vector-based similarity matching over time

Every new customer environment that Asato processes makes the MKG smarter, improving entity resolution and enrichment accuracy for every customer on the platform.