Teenva AI & Digital Ventures builds transaction categorization AI for PFM apps, neobanks, account aggregators, and SME finance tools from Bangalore, India. Raw bank feeds arrive as cryptic narrations—UPI/123456789/ZOMATO, NEFT-ICIC0000123-SALARY—Teenva normalizes merchants, assigns spend categories, flags recurring bills, and learns from user corrections in your mobile app or custom fintech platform.
Clean categories power budget charts, tax summaries, and cash-flow features in credit risk models—not to be confused with approve/decline decisions or fraud scoring.
Build your AI solution with us
Transaction categorization AI—merchant enrichment and spend labels for PFM and accounting feeds.
Hub: AI solutions · Domain: Fintech
What gets labeled
| Input field | Example raw value | Enriched output |
|---|---|---|
| Narration / description | UPI-PAYTM-9876543210-SWIGGY | Merchant: Swiggy · Category: Food & Dining |
| Amount + direction | −₹499 debit | Expense (not transfer) |
| MCC (cards, scoped) | 5812 | Category: Restaurants |
| Counterparty (scoped) | IFSC + account hash | Transfer vs payment classification |
| User history | Same merchant last month | Confirm category; detect subscription |
| Geo (scoped) | ATM near home | Category: Cash withdrawal |
Taxonomy is your chart of categories—we map to standard PFM buckets or custom SME GL codes (scoped).
How categorization works
| Layer | Role |
|---|---|
| Parser | Extract merchant tokens from UPI/NEFT/IMPS/card strings |
| Merchant dictionary | Alias table + fuzzy match (Paytm* vs PAYTMQR) |
| ML classifier | Text + amount features → category when merchant unknown |
| Rules | Salary keywords, EMI to known lender, internal transfers |
| LLM assist (optional) | Ambiguous narrations only—bounded taxonomy output, not free text |
| User feedback | Correction API retrains merchant → category mapping |
| Confidence score | Low confidence → “Uncategorized” or ask user in app |
Batch and real-time API modes—PFM sync overnight or label at transaction ingest.
Features we implement
Categorization API
Single txn or bulk CSV/JSON ingestMerchant enrichment
Display name, logo URL (scoped), website guessRecurring detection
Rent, SIP, subscriptions, EMIsTransfer detection
Self-transfer, wallet top-up vs spendIncome vs expense
Salary, refund, cashback classificationMulti-account (scoped)
Same merchant across savings and credit cardSME mode (scoped)
Map to expense GL for accounting chatbot feedsAdmin console
Review low-confidence, edit taxonomy, export metricsAccuracy monitoring
Precision on holdout; user override ratePrivacy
Processing in your VPC; no resale of transaction data
SME mode maps to expense GL for accounting chatbot feeds.
Architecture
- 1.Bank / AA / card feed (batch or webhook)
- 2.Normalization pipeline (parse narration, amount, type)
- 3.Merchant match → dictionary + ML + rules
- 4.Category + subcategory + tags + confidence
- 5.Categorization API → PFM app / data warehouse
- 6.User correction events → feedback store → retrain
- India-specific parsers — UPI handle patterns, Paytm/PhonePe prefixes, NEFT remarks
- Latency — Bulk 1M+ txns/day batch; <50 ms single txn API (scoped)
Use cases by product type
| Product | Value |
|---|---|
| Consumer PFM app | Budget pie charts, “where did my money go?” |
| Neobank | In-app spend insights after KYC |
| Account aggregator (scoped) | Consent-based multi-bank categorize |
| SME bookkeeping | Auto-tag expenses for GST and reports |
| Lender bank statement analysis (scoped) | Features for credit risk—category aggregates |
| Corporate card (scoped) | Policy flags on category (travel, entertainment) |
Integrations
| System | Role |
|---|---|
| Custom PFM / neobank | Transaction ingest via custom software development |
| Mobile app | Category override UI in mobile app |
| Account Aggregator / bank APIs (scoped) | Connect via API integration services |
| Financial advisory chatbot | Dining spend tool via financial advisory chatbot (scoped) |
| Data warehouse | Snowflake/BigQuery category columns for analytics |
| Zoho Books / CRM | Expense export via Zoho integration (scoped) |
80% of UPI strings are unreadable to humans—your charts shouldn't stay "Unknown".
We combine India-specific parsers, merchant dictionaries, and ML so PFM insights actually drive retention.
Categorization vs other fintech AI
| Product | Focus |
|---|---|
| Transaction categorization AI | Label spends—merchant and category for PFM/books |
| AI credit risk analysis | Underwriting default risk |
| AI fraud detection | Block fraudulent transactions |
| Financial advisory chatbot | Answer product FAQ in chat |
| AI KYC verification | Identity at onboarding |
Category aggregates may feed credit models—they do not replace fraud scoring.
Why Teenva AI
PFM product + NLP
Mobile apps and categorization pipeline togetherIndia narrations
UPI/Bharat BillPay/IMPS patterns in training dataUser-in-the-loop
Corrections improve accuracy without full re-labeling projectOwn your taxonomy
Custom categories for SME GL, not one global bucket onlyOps
Managed IT support when feed formats change
Delivery process
Discovery
Feed sources, volume, taxonomy, accuracy target
Sample ingest
10k–100k anonymized txns for EDA
Taxonomy design
Category tree + transfer/income rules
Parser + dictionary
India UPI/NEFT patterns; top merchants
ML classifier v1
Train on labeled subset
API + batch jobs
Integrate with ingest pipeline
App override UX
Capture user corrections (scoped)
Pilot
Measure accuracy vs manual baseline
Retrain cadence
Weekly merchant updates; monitor drift
Frequently asked questions
We build or hybrid—custom taxonomy, India narrations, and data residency in your cloud.
85–95%+ top-1 category on mature books with dictionary + ML; improves with user feedback.
Optional for ambiguous rows only—bulk path uses parser + ML for cost and latency.
Yes—map to GL codes and GST expense types (scoped).
Yes—API per txn or micro-batch on ingest (scoped).
- Aggregates can input credit risk—not standalone approve/decline.
English/Hindi mixed narrations handled; regional bank formats in parser rules (scoped).
Processing under your consent framework; no cross-client merchant learning unless you opt in.
Yes—interval and amount pattern detection for bills and SIPs.
8–10 weeks MVP API with sample taxonomy and India UPI parser for one feed type.
Explore
Build your transaction categorization engine
Build your AI solution with us—merchant enrichment and spend labels for PFM and accounting feeds.
Build your AI solution with us
Ready to take your business to the next level?