Skip to main content
Transaction enrichment is the core capability of Triqai. It takes raw, unstructured transaction strings from bank statements and transforms them into clean, structured data that’s easy to understand and use.

​
The Problem

Bank transaction data is notoriously messy. A simple coffee purchase might appear as: 
POS 4392 STARBUCKS STORE #1234 NEW YORK NY 10001
This string contains useful information: merchant name, store number, location, but it’s buried in noise and inconsistent formatting. Every bank formats transactions differently, making it impossible to reliably extract meaning without specialized processing.

​
The Solution

Triqai analyzes transaction strings using a combination of:
  • Pattern matching against known merchant signatures
  • Natural language processing to extract entities
  • Extensive merchant databases (150M+ companies)
  • Location intelligence (10M+ places globally)
  • Machine learning models for classification and confidence scoring
The result is structured data you can immediately use in your application.

​
Enrichment Pipeline

When you submit a transaction for enrichment, it goes through several stages:
1

Parsing

The raw transaction string is tokenized and analyzed for structure.
2

Entity Detection

Potential merchants, locations, intermediaries, and other entities are identified.
3

Matching

Detected entities are matched against our databases of known merchants, locations, and intermediaries.
4

Classification

The transaction is categorized and additional metadata (channel, subscription status) is determined.
5

Scoring

Confidence scores with explanatory reason tags are calculated for each entity and field.

​
What Gets Enriched

Each enrichment request returns two main sections: transaction metadata and an entities array.

​
Transaction Metadata

Classification and signals about the transaction itself:
  • Category — Hierarchical spending categories (primary, secondary, tertiary) with MCC/SIC/NAICS codes
  • Channel — How the transaction occurred (online, in-store, mobile app, ATM, bank transfer)
  • Subscription — Whether this is a recurring payment and its type
  • Confidence — Overall enrichment confidence with reason tags

​
Entities Array

An array of identified real-world entities, each with a type, role, confidence, and data:
  • Merchant — The business behind the transaction (name, logo, website, domain)
  • Location — The physical place where it occurred (address, coordinates, timezone)
  • Intermediary — Payment processors, delivery platforms, wallets, or P2P services (Stripe, Venmo, DoorDash, etc.)
  • Person — Recipient information for P2P transfers (display name)
Only entities that are actually identified appear in the array. If no location is found, no location entity is present. There are no "status": "no_match" entries.

​
Request Format

A basic enrichment request requires three fields: 
{
  "title": "AMAZON MKTPLACE PMTS AMZN.COM/BILL WA",
  "country": "US",
  "type": "expense"
}
FieldTypeRequiredDescription
titlestringYesRaw transaction description from bank statement
countrystringYesISO 3166-1 alpha-2 country code (e.g., “US”, “NL”, “GB”)
typestringYesTransaction direction: expense or income

​
Response Structure

The enrichment response uses an entities array pattern. Only found entities are included: 
{
  "success": true,
  "partial": false,
  "data": {
    "transaction": {
      "category": {
        "primary": {
          "name": "Coffee Shops",
          "code": { "mcc": 5814, "sic": 5812, "naics": 722515 }
        },
        "secondary": {
          "name": "Food & Dining",
          "code": { "mcc": 5812, "sic": 5812, "naics": 722511 }
        },
        "tertiary": null,
        "confidence": { "value": 95, "reasons": ["merchant_category_match"] }
      },
      "subscription": { "recurring": false, "type": null },
      "channel": "in_store",
      "confidence": { "value": 92, "reasons": [] }
    },
    "entities": [
      {
        "type": "merchant",
        "role": "organization",
        "confidence": {
          "value": 98,
          "reasons": ["name_closely_matched", "results_consensus"]
        },
        "data": {
          "id": "...",
          "name": "Starbucks",
          "alias": ["SBUX"],
          "icon": "https://logos.triqai.com/images/starbuckscom",
          "website": "https://www.starbucks.com",
          "domain": "starbucks.com"
        }
      },
      {
        "type": "location",
        "role": "store_location",
        "confidence": {
          "value": 85,
          "reasons": ["city_match", "store_id_match"]
        },
        "data": {
          "id": "...",
          "name": "Starbucks - Times Square",
          "formatted": "1530 Broadway, New York, NY 10036, USA",
          "structured": { "city": "New York", "state": "NY", "country": "US" }
        }
      }
    ]
  },
  "meta": {
    "generatedAt": "2026-01-15T10:30:00Z",
    "requestId": "3c90c3cc-0d44-4b50-8888-8dd25736052a123",
    "version": "1.1.9",
    "categoryVersion": "triqai-2026.01"
  }
}
Only identified entities appear in the entities array. If no intermediary or location was found, they are simply absent not present with a null status.

​
Partial Results

Sometimes not all enrichment modules succeed. For example, a transaction might have a recognized merchant but no identifiable location. In these cases:
  • The response includes partial: true
  • Successful enrichments are returned normally in the entities array
  • The meta.errors array lists which enrichers failed
  • No credits are deducted for partial results
{
  "success": true,
  "partial": true,
  "data": {
    "transaction": { "confidence": { "value": 75, "reasons": [] } },
    "entities": [
      {
        "type": "merchant",
        "role": "organization",
        "confidence": { "value": 95, "reasons": ["name_closely_matched"] },
        "data": { "name": "Acme Corp" }
      }
    ]
  },
  "meta": {
    "errors": ["location_timeout"]
  }
}

​
Best Practices

The country code helps Triqai narrow down merchant and location matching. An incorrect country code may lead to less accurate results.
Don’t pre-process or truncate transaction titles. The full string often contains valuable signals like store numbers and location hints.
The type field (expense vs income) helps classify refunds, transfers, and income sources correctly.
Check confidence scores and their reason tags before displaying data to users. Low-confidence results may need manual review. Reason tags explain why the score is what it is.

​
Next Steps

Categories

Learn about the category taxonomy

Entities

Understand merchant, location, intermediary, and person data

Confidence Scores

How to interpret confidence scores and reason tags

API Reference

See the full API documentation