Clearpoint Financial Services processes tens of thousands of card and ACH transactions per month. Rachel Torres sits at the intersection of accounting and risk — certifying that posted debits match authorized amounts, that balance checks passed at clearing, and that no duplicate charges slipped through. The team handles the full reconciliation lifecycle internally, including ownership of exception definitions and audit defensibility.
Legacy thresholds and manual filter passes could not scale to 50,000 records
The team's reconciliation workflow relied on a downloaded bank export, pivot tables, and hard-coded filter thresholds inherited from a previous analyst. Four distinct exception categories required separate manual passes through the data. The high-value cutoff was a flat dollar figure set two years prior — no statistical grounding, no update mechanism. Authentication-risk joins ran poorly at this record volume. Ledger discrepancy checks required identifying debit transactions that cleared despite insufficient account balances. Duplicate detection demanded row-level deduplication logic the spreadsheet could not reliably execute at scale. Compounding the problem, over 87 percent of records lacked a time component in the transaction timestamp, defaulting to midnight (00:00) and completely blocking off-hours fraud analysis. An internal audit review formalized the pressure: the committee flagged the flat-dollar threshold as statistically unjustified and requested documented derivation for every exception category.
Energent.ai became the statistical reconciliation engine
Torres uploaded the 50,000-record CSV directly to Energent.ai. Within a single session, the agent:
- Inspected the schema and confirmed column types before writing any analysis code
- Computed the mean ($297.87) and standard deviation of transaction amounts across the full dataset
- Derived the high-value cutoff at $1,176.33 using a Z-score threshold of 3 standard deviations — a figure recalcutable each cycle from the new batch
- Cross-referenced login attempt counts to isolate authentication-risk transactions at the top 1 percent of friction events
- Matched debit rows against recorded account balances to flag cleared-but-insufficient records
- Detected the 87-percent missing-timestamp gap mid-session and documented it as a concrete defect report rather than silently skewing results
- Generated an interactive HTML exception dashboard sorted by severity, ready to distribute as a standalone file
No data pipeline. No BI tool configuration. No handoff between systems.
Threshold derivation, not just cleaner reporting
- Audit-defensible logic. The $1,176.33 cutoff was derived from the live dataset and explained step-by-step, satisfying the audit committee's documentation requirement directly.
- Data-quality surfaced explicitly. The agent flagged the 87-percent missing-timestamp finding rather than defaulting silently — giving the team a quantified defect report to bring to the ingestion pipeline owner.
- Four queues, one pass. Amount exceptions, authentication risks, ledger discrepancies, and duplicate charges were enumerated in a single session, replacing four separate manual filter runs.
- Reproducible by design. The same statistical logic re-executes against each new batch export without rebuilding spreadsheet formulas.
4,004 exceptions isolated, ranked, and documented in one session
- 4,004 total exceptions surfaced — approximately 8 percent of the 50,000-record batch
- 989 amount exceptions flagged above $1,176.33, representing the top ~2 percent of transaction volume
- 1,305 authentication-risk transactions isolated at 4 or more login attempts per transaction
- 1,708 ledger discrepancies identified where debits cleared despite insufficient account balances — the largest single exception queue
- 2 confirmed duplicate charges escalated immediately as the highest-priority refund candidates
- 87-percent timestamp gap documented, unblocking the roadmap for off-hours fraud detection pending an upstream ingestion fix
"The ledger discrepancy count was something we'd never isolated cleanly at this scale before. Now we have a number we can defend — and a process we can run again next quarter without touching the formulas." — Rachel Torres, Reconciliation Analyst at Clearpoint Financial Services