Payroll parallel testing with SmartCompaire: questions and answers

Pay results from a source system and a target system, employee by employee and pay code by pay code. Both sides arrive as spreadsheet exports, paired through your employee and paycode mappings. Each run records source amount, target amount and difference, graded against your thresholds. SmartCompaire does not correct differences or write back to any payroll system.

The upload screens offer .xlsx, .xls and .csv; the server also accepts .xlsm and .ods. Only the first worksheet is read. CSV files can be UTF-8 or Windows-1252, separated by commas, semicolons, tabs or pipes. Header matching ignores case, spaces and punctuation, and leading zeros in IDs are kept. Rows can also be typed or pasted.

The default upload limit is 100 MB per file. Files over 2 MB, and pastes over 500 rows, go to a background import with progress tracking and a cancel option. It reads the file in 5,000-row batches; the design target is a few hundred thousand rows. On-screen edits are capped at 1,000 pay rows, 500 employee mappings or 5,000 paycode mappings per request.

Any system that can export pay results to a spreadsheet or CSV file, for example SAP, Workday, Oracle or ADP, once the export is arranged in the 15-column pay data template (only PayCode is required). There are no live connections, scheduled pulls, SFTP folders or public API.

An employee mapping pairs each source employee number with one target number. A paycode mapping links a source pay code to a target pay code under a Bucket and Bucket code type, such as Earnings/Gross. Mappings that share a source or target key (tax authority plus pay code) are grouped, so two source codes mapped to one target code are summed and compared as a total. Rows found on only one side still appear as result rows.

A pay row is matched first on exact tax authority plus pay code; then, if the row has no tax authority, on pay code alone, but only when that code is mapped exactly once; then to a mapping with a blank tax authority, which matches its pay code under any authority. Ambiguous rows are left unmatched, not guessed.

Each pay data import reports how many rows have no paycode mapping yet. Before a run, the readiness check shows how many rows will match, lists up to 25 unmatched pay codes with reasons, and counts unmapped employees. Unpaired result rows are labeled No employee mapping or No paycode mapping. Each run re-matches against current mappings, so after a fix you rerun without uploading again.

A threshold is an absolute amount, such as 0.01, set per Bucket and Bucket code type pair (a pay category and its type, such as Earnings/Gross). A row passes when its absolute Amount difference is at most the threshold; rows with no threshold get no grade. Percentage thresholds are not supported, and Number and Rate differences are not graded. Projects start with your organization’s defaults, editable per project. Status rules, such as Matched or Over Threshold, label each row, and you can add your own.

A project can be run as often as you need, and every run is kept in its history. Within a project, amounts are summed per employee and pay code across all loaded pay data, so several periods or check dates become one total. To compare periods separately, use one project per period. MTD, QTD and YTD amounts are stored but not compared.

A Time project uses the payroll file layout and comparison logic, with hours or units in the Number column. Number differences are recorded but not graded, and there is no logic for shifts, overtime or punches. Country is a project label: the comparison is the same for every country, with no country-specific tax rules.

Reviewers can set a status, status override and comment on any result row, and link it to a defect or a KD/AD (known or acceptable difference) record. Defects carry a priority, assignee, due date and status; KD/AD records hold a justification document link and approval status. Linking is manual, row by row, and does not carry over to later runs.

Hirra is the AI assistant inside SmartCompaire, labeled beta with a warning that it can make mistakes. It reads projects, pay data, results, defects and analytics, and can summarize a project with next steps. For members and above it can also create projects, raise and update defects and KD/AD (known or acceptable difference) records, update review fields on result rows, generate reports and cancel jobs; viewers can only read. Runs, imports, bulk status updates, threshold changes and archiving need a confirmation chat cannot give yet, so you do them on their own screens. Hirra cannot edit pay amounts, delete data or change mappings.

Questions go from your browser to the SmartCompaire backend, its internal AI service, then OpenRouter, a third-party model router. In chat, OpenRouter’s Auto router picks the model per turn by default; configured fallbacks include Claude Sonnet 4.5, GPT-4.1 and Gemini 2.5 Pro. Data Hirra reads, which can include pay rows, employee numbers and amounts, reaches the model unredacted, with no zero-retention or no-training option set. Hirra acts with your own login, so it sees only what you could. Each tool call goes to an append-only audit log the database blocks from edits and deletion. With the optional avatar on, reply text goes from your browser to HeyGen.

Organizations are separated by database row-level security, and the server checks every request against four roles (viewer, member, admin, owner) and per-person permissions. Sign-in is by Google or by email and password (12+ characters, bcrypt hashed, locked 15 minutes after 8 failed attempts). Uploaded files sit in AWS S3 with AES-256 server-side encryption. There is no multi-factor authentication, no single sign-on beyond Google, and no certification such as SOC 2 or ISO 27001. The platform operator’s superadmin staff can see data in every organization; switching into one is recorded in the audit log, but individual reads are not.

Pay data, results, uploaded files and Hirra chat history have no automatic expiry; generated reports expire after 14 days. Permanent deletion works only on an archived project, after you type its exact name. It removes the project’s mappings, pay data, runs, results, defects, KD/AD records, reports and jobs from the database. Audit and AI action logs, and Hirra conversations that mention the project, are kept, and the original uploaded files stay in file storage.

Not yet: there is no approved SmartCompaire Privacy Notice or Terms of Service. Incomplete drafts awaiting SLRIS Solutions, LLC legal review can be read at /privacy and /terms; they are not in force. Until they are approved, the site footer links to SLRIS’s company-wide Terms & Conditions and Privacy Policy on slris.com. For how those apply to SmartCompaire, contact SLRIS at slris.com/contact-us.

Pricing is not published; ask SLRIS Solutions through the Request a demo page. To start, create a project, load the employee mapping, paycode mapping, source pay data and target pay data, review the readiness check, then start a run.