DATA STRUCTURE SPECIFICATIONS
Format your data. Upload it. Get your report.
You need one file: your reconciled transactions in the Workbench schema, with each internal (A) record and its external (B) match on the same row, plus the match outcome. One CSV, one upload.
The schema is standardised - no custom scripting or field remapping required for your accounting system.
.png)
%20(2)%202.png)
Three steps from raw data to Uplift Report.
1. Format your transaction data
Provide one CSV file using the 27 fields below. Each row pairs an internal (A-side) record with its external (B-side) match. For unmatched items, leave the other side blank. The match columns (matchId, matchRule, matchedBy, wasPreviouslyMismatched, targetAllocation) tell Matchimus how your team matched historically - it's what the engine calibrates against.
Field Name
Type
Description
Why it matters
Record_ID
String
A unique identifier for every row.
Prevents duplicate processing.
Trans_Date
Date
The date the transaction occurred (YYYY-MM-DD).
Used for "fuzzy matching" on timing differences.
Value_Date
Date
The date the funds actually cleared.
Critical for bank-to-ledger lag analysis.
Amount
Decimal
The absolute value of the transaction.
Core matching criteria.
Currency
String
ISO Code (e.g., USD, EUR, GBP).
Identifies FX reconciliation needs.
Direction
String
DEBIT or CREDIT (or +/-).
Essential for basic rule filtering.
Reference
String
The primary txn reference (e.g., Check #, Wire ID).
Best for 1:1 Rules-based matching.
Description
Text
The transaction narrative or bank commentary - this is what Matchimus parses
The ML Value Prop: ML parses this better than rules
Trans_Type
String
Category (e.g., DIVIDEND, FEE, SWEEP).
Helps segment the data for specific rule sets.
Field Name
Description
matchId
Unique identifier for the match record pairing an A-side and a B-side transaction. (Blank in the case of a unmatched transactions only file)
matchDate
Date the match was made.
matchRule
The name of the matching rule or rule set that produced the match. If a manual match, this should be populated with 'MANUAL'
matchedBy
Who or what made the match, e.g. automatic engine or a named user (manual).
wasPreviouslyMismatched
Flag (1/0) showing whether this transaction pair was previously matched incorrectly (used to baseline the error rate).
A_transactionType
"A" if an internal/ A type transaction, blank otherwise
A_id
Unique identifier of the A-side transaction.
A_allocation
Allocation of the A-side transaction. This is usually composed of the currencyCode, account, valueDate and transactionAttributes. The allocation field is taken as the primary business characteristics of the A type transaction and any A-type transactions with the same allocation will be considered equivalent from a business perspective (and equivalent in terms of match correctness)
A_importDate
Date the A-side transaction was imported/loaded into the system.
A_debitOrCredit
Whether the A-side transaction is a debit (DR) or a credit (CR)
A_amount
Monetary amount of the A-side transaction.
A_valueDate
Value date of the A-side transaction (the date it takes effect / settles).
A_currencyCode
Currency of the A-side transaction (ISO 4217 code, e.g. GBP, USD).
A_account
Account the A-side transaction belongs to.
A_transactionReferences
Reference fields on the A-side transaction (e.g. invoice number, payment reference) used for matching.
A_transactionAttributes
Additional descriptive attributes of the A-side transaction (e.g. narrative, counterparty, other metadata).
B_transactionType
"B" if an external/ B type transaction, blank otherwise
B_id
Unique identifier of the B-side transaction.
B_importDate
Date the B-side transaction was imported/loaded into the system.
B_debitOrCredit
Whether the B-side transaction is a debit (DR) or a credit (CR)
B_amount
Monetary amount of the B-side transaction.
B_valueDate
Value date of the B-side transaction (the date it takes effect / settles).
B_currencyCode
Currency of the B-side transaction (ISO 4217 code).
B_account
Account the B-side transaction belongs to.
B_transactionReferences
Reference fields on the B-side transaction used for matching.
B_transactionAttributes
Additional descriptive attributes of the B-side transaction.
targetAllocation
The correct/expected A_allocation for the match - the target value the A-side allocation should be (e.g. ground-truth label). If a single match contains multiple A transactions with A_allocations, this consists of a comma separated list of these allocations, enclosed by brackets
Field Name
Type
Description
Match_Group_ID
String
A unique ID that links matching rows together.
Internal_Record_ID
String
The Record_ID from the Internal/GL file.
External_Record_ID
String
The Record_ID from the Bank/External file.
Match_Type
Enum
1:1, 1:M (Many), M:M, or Manual.
Confidence_Label
Boolean
Was this a "System Match" or a "Manual User Match"?
Anonymise before you upload
2. Anonymise your data locally
Run these utilities on your files before upload. Everything happens within your network - nothing is transmitted during the masking process.
Strip personal identifiers
How it works: Strips name and address patterns from narrative fields. Replaces them with anonymised tokens - Client_Alpha, Client_Beta, and so on.
Hash account identifiers
How it works: Converts routing numbers, account numbers and unique identifiers to SHA-256 strings. Matching logic is preserved - raw values are not.
Scale numeric values
How it works: Applies a uniform multiplier across all balance and amount fields. Exact values are hidden - variance relationships that Matchimus needs are preserved.
3. Upload your data
Once your files match the schema and your masking is applied, log in and upload. Your data goes into an isolated partition scoped to your organisation.
01
Confirm no production identifiers remain
Check that all account names, addresses and account numbers have been masked or hashed using the utilities above.
02
Your data stays in your partition
Your files are held in a single-tenant repository restricted to your organisation. No data is shared with or accessible by other clients.
03
Deleted 14 days after your report is delivered
Every file and derived metadata is deleted within 14 days of your report being delivered.
What you get back.
Once your data is processed, you’ll receive your Uplift Report within 5 business days. It contains three specific outputs:
Where your rules break down
The exact transaction types and narrative patterns your current logic can't handle — and how many of them there are.
How much Matchimus improves on your current rate
The percentage of your manual exceptions that Matchimus resolves automatically, expressed as FTE time and cost.
How long your reconciliation takes vs. the industry
How your resolution time and manual intervention rate compare to the Tier 1 BenchRec benchmark.
Four things to do before you upload.
Step 1: Choose your dataset
Pick the reconciliation process with the highest manual intervention — that's where the benchmark will be most useful.
Choose a high-friction dataset
Bank-to-ledger and intercompany reconciliations work best. The higher the manual rate, the clearer the benchmark result.
Check your volume
You need at least 50,000 rows for statistically significant results. Up to 250,000 gives the strongest confidence calibration.
Include your historical match outcomes
Fill in the match columns in your file (matchId, matchRule, matchedBy). They tell Matchimus which records your team matched historically. Without them, the benchmark can't run.
Step 2: Anonymise your data
No PII should leave your environment. Run the Local Masking Utility on your files before upload - it takes minutes and runs entirely within your network.
Hash account identifiers
Apply SHA-256 hashing to account numbers, client names and any national identifiers. Matching logic is preserved - raw values are not.
Scale financial values (optional)
If exact amounts are sensitive, apply a uniform multiplier across all amount fields. Variance relationships - which Matchimus needs - are preserved.
Step 3: Format your file
Make sure your file matches the Workbench schema before upload. Three things to check:
File Format
.CSV only.
Schema Alignment
Your column headers must match the 27 fields in the Workbench Schema Template exactly, including capitalisation (e.g. matchId, A_amount, B_valueDate).
Step 4: Upload to the portal
Upload your anonymised files via the secure Workbench portal. Encrypted in transit, Zero-PII. Do not send data files by email.
Get internal sign-off first
Share the anonymised file with your Data Privacy or Compliance officer before uploading. The Compliance FAQ and Zero-PII Security Document are available to support that review.
Upload your file
Log in to the Workbench and upload your anonymised .csv file to your secure data partition. If you don't have an account yet, create one - a corporate email is required.
What happens next
Once your file passes the automated schema check, you'll receive a confirmation email. Your Uplift Report will be ready within 5 business days.
Technical Toolkit
