Integration Guide

From raw feed to forecast-ready data. Under 200ms.

A single POST endpoint transforms miscategorized transaction batches into accurate, forecast-ready data. Here's exactly how the pipeline works.

Quickstart Guide API Reference

Pipeline stages

1. Ingest — normalize bank format

2. Classify — ML reclassification

3. Score — confidence assignment

4. Forecast — 90-day signal generation

5. Return — corrected JSON response

Total: <200ms median

Pipeline

The full data path

API Details

The API request and response

POST your transaction batch, get corrected categories and forecast signal back synchronously.

POST /v1/classify — Request

POST https://api.spendaqo.com/v1/classify
Authorization: Bearer spq_live_...
Content-Type: application/json

{
  "account_id": "acct_kst_8841",
  "transactions": [
    {
      "id": "txn_001",
      "amount_cents": 8499,
      "date": "2026-06-10",
      "merchant_name": "AMZN*MKTP",
      "raw_category": "DEBIT_MISC"
    },
    {
      "id": "txn_002",
      "amount_cents": 29900,
      "date": "2026-06-11",
      "merchant_name": "NOTION HQ",
      "raw_category": "PAYMENT_OTHER"
    },
    {
      "id": "txn_003",
      "amount_cents": 41200,
      "date": "2026-06-12",
      "merchant_name": "DELTA AIR",
      "raw_category": "UNCATEGORIZED"
    }
  ]
}

200 OK — Response (142ms)

{
  "status": 200,
  "latency_ms": 142,
  "results": [
    {
      "id": "txn_001",
      "corrected_category": "Office Supplies",
      "confidence_score": 0.97,
      "original_category": "DEBIT_MISC"
    },
    {
      "id": "txn_002",
      "corrected_category": "Software Subscription",
      "confidence_score": 0.96,
      "original_category": "PAYMENT_OTHER"
    },
    {
      "id": "txn_003",
      "corrected_category": "Business Travel",
      "confidence_score": 0.94,
      "original_category": "UNCATEGORIZED"
    }
  ],
  "forecast_signal": "stable_positive",
  "forecast_horizon_days": 90,
  "forecast_confidence": 0.83
}

Latency

Why sub-200ms matters for banking apps

Banking apps make synchronous API calls during transaction loading. If your classification layer is slow, your app is slow. Spendaq is designed as a synchronous enrichment layer — not an async batch job.

In-region compute

Classification runs in US-East ephemeral containers. No cross-region hops in the critical path.
In-memory model serving

The classification model is pre-loaded in memory. No cold-start penalty on classification calls.
Zero retention after response

Transaction data is not persisted. Container destroyed after response returned.

Latency breakdown (p50)

API Gateway 8ms

Normalization 14ms

ML Classification 88ms

Forecast Signal 22ms

Response serialization 10ms

Total (p50) 142ms

Start your integration today.

Follow the quickstart guide — most integrations go live in under a sprint.

Quickstart Guide Talk to the team