Meridian Components is a mid-market manufacturer operating three branch offices across multiple legal entities. A team of two staff analysts and one controller runs the monthly bank reconciliation process across payroll, accounts payable, and intercompany accounts. Each close package feeds directly into external audit review at quarter-end.
Excel was silently truncating the ledger — and the workpaper never showed it
Multi-year GL exports from three branches pushed row counts past Excel's 1,048,576-row hard limit. Excel opened truncated files without warning. The analyst ran a gap check on an incomplete population, and the resulting workpaper reflected only a fraction of the ledger. There was no sign anything was wrong.
Two more failure modes compounded the problem. The GL used alpha prefixes — CHK- for physical checks, EFT- for electronic transfers. Running one numeric sequence across both series generated false positives wherever they interleaved. The bank feed stripped leading zeros before delivery, turning GL identifier CHK-0045 into CHK-45; VLOOKUP missed the match and logged it as a gap.
Parallel branch allocations added structural complexity: Branch A (1000–1999), Branch B (5000–5999), Branch C (2000–2999, inserted between the first two after the third office opened). Every close cycle required manual documentation of the 3,000-number inter-range gap to satisfy the audit committee — a recurring task with no analytical value that could not be skipped. When the external audit window compressed from ten to six business days, the existing workflow had no margin left.
Energent.ai replaced the pre-processing, normalization, and note-drafting steps
The team uploads three source files at session start — GL export (CSV), bank statement feed (CSV), void log (PDF) — with no pre-transformation. Energent.ai then:
- Partitioned transactions by alpha prefix, isolating CHK- and EFT- sequences before any sequence-diff analysis
- Normalized leading zeros, aligning GL and bank-feed identifiers before matching so CHK-0045 and CHK-45 resolve as the same transaction
- Applied expected-range parameters for all three branch allocations, filtering structural inter-range gaps from the anomaly list
- Classified each genuine gap into one of three states: outstanding item (GL present, bank absent), potential unrecorded disbursement (bank present, GL absent), or voided transaction (absent from both, pending void-log verification)
- Generated structured draft workpaper notes for confirmed anomalies, citing the specific check range, root cause, and corroborating evidence
No custom code. No BI dashboard. No Excel pre-processing pass.
Partitioning upstream, not filtering downstream
- Prefix partitioning eliminates the false-positive class at the source. Separating CHK- and EFT- before analysis means interleaving noise never enters the anomaly list.
- Expected-range parameters are registered once and carry forward. Branch allocations persist period to period as a standing approved-range register the audit committee can inspect directly — replacing recurring manual dismissals.
- Three-state classification assigns resolution paths at triage. Each gap arrives pre-labeled so the team routes anomalies rather than categorizing them from scratch.
- Full row-count logging provides a population certificate. The agent records the count of rows processed at each stage, giving the workpaper an auditable confirmation that analysis covered the complete ledger.

Workpaper drafting cut from two hours to fifteen minutes
- Silent truncation eliminated. Full ledger exports processed at any row count; record totals logged and auditable at every stage.
- False positives dropped to near zero. The team previously dismissed 15–30 apparent gaps per close cycle as prefix-interleaving or leading-zero artifacts; partitioning and normalization remove that category automatically.
- Note drafting: approximately two hours → fifteen-minute structured review. Four confirmed anomalies at the most recent close — checks 1041–1044, traced to a Branch A printer jam — produced complete, audit-ready notes from the agent's own triage output.
- Compressed audit window absorbed. When the external audit committee moved preliminary fieldwork two weeks earlier, cutting available triage time from ten to six business days, the team met the new schedule without adding temporary staff.
"We were writing the same note structure every month with different check numbers. Now I spend fifteen minutes reviewing what the agent already drafted. That time went back into close work that actually requires judgment." — Rachel Okafor, Financial Controller at Meridian Components
