Creditcrest Enrich

Raw bank descriptions in; categories and payees out, each with its reason.

Part of Creditcrest Statement, in Creditcrest Decisioning Suite.

Every match has a reason.

A category and a payee for every transaction, with the rule that matched.

What you do here. Paste a list of transactions and get a category and payee for each, with the rule that matched.

Enrich transactions Try it end to end Take the tour

You bring: A list of transactions
You get: Category, third party, confidence, and the rule that matched
Holds the line: An unknown payee is flagged, never guessed

The fourth matches no rule and is returned uncategorised, not guessed.

The transactions

One a line: a description and an amount, separated by a comma or a tab. Credits positive, debits negative. A CSV a bank exported reads too — the layout is worked out from the data. Up to 20,000 rows in one go.

Or a file

A PDF statement, a CSV, an Excel workbook, or an OFX or QIF file. Four megabytes, which is a great many years of transactions. It is read in this request and kept nowhere.

Setting

Strict: The rules alone. Every answer is a name or a kind of business a rule read. Balanced: The rules, the bank’s merchant code where a connection sent one, a register of Australian places, and a modelled reading where it keeps its category at 95% precision or better. Broad: More answers. The bank’s merchant code where a connection sent one, a register of Australian places, and a modelled reading where it keeps its category at 90% precision or better. Complete: Every line gets a category. Read as broad reads, and where nothing read a line, the model’s best guess, marked as one: money out as a spending line, money in as a transfer. Never income, a commitment or gambling on a guess, and not held to a precision floor.

Show me the mixed example Show me a whole invented statement

Nothing is kept: what you paste is read, then gone

What happens to what you paste. It is read while the request is open and then it is gone. Nothing is written to disk, nothing is written to a log, nothing is put in a database — there is no database here to put it in. The transactions travel in the request body rather than the address bar, so this reading has no link, which means there is no link to leak.

The cost of that is honest too. Reload and the reading is gone, and the form re-sends what you gave it from your own browser, because there is nothing on this end to refer back to.

No account, no email address and no tracking cookie; visits are counted without one. Nothing here calls a hosted model: the rules are a text file in this repository and the whole prediction happens in this process.

What it is for

What this is for. Categories and third parties are what let you club transactions together, and clubbing them is what turns a list of payments into ongoing commitments and liabilities. That is why a lender wants it, and it is why the output of this is the input to a decision rather than the decision.

What it will not do. It will not decide, it will not score anybody, and it will not fill a gap. A payee it does not recognise is unknown, not other — and a third party is the words a rule recognised in a bank's description rather than a fact about who was paid.