AI Workflow & Prompt Engineering Project

This project demonstrates how I use AI to solve financial analysis problems through prompt engineering, iteration, and real world application.

PROJECT EXAMPLE: RECONCILIATION DISCREPANCY ANALYSIS

Problem:

HQ Finance and Corporate Accounting teams identified multiple reconciliation discrepancies across departmental forecast submissions. Inconsistent reporting structures were creating variances between departmental submissions and consolidated reporting, affecting financial visibility and audit readiness.

The AI Workflow — What Triggered It:

Q4 forecast submissions revealed reconciliation gaps between departments and consolidated totals that needed quick diagnosis before monthly close and quarterly reporting cycles.

The Prompt Structure:

The HQ Finance and Corporate Accounting teams have identified multiple reconciliation discrepancies across departmental forecast submissions ahead of Q4 planning and monthly financial reporting reviews. Several departments are using inconsistent reporting structures, resulting in forecast totals that do not align across the monthly management reporting packages and financial consolidations. Finance Operations requested support from the Corporate Accounting team after multiple reporting inconsistencies were identified during monthly close activities.

The team is trying to better understand the financial impact and reporting exposure associated with these inconsistencies, including whether differences in reporting hierarchies, cost center mappings, and forecast classifications are creating material variances across departmental reporting and consolidated financial results. The review is intended to help determine whether the inconsistencies represent isolated reporting differences or broader reconciliation risks that could affect financial visibility, management reporting accuracy, and audit readiness during quarterly close activities.

You have four source documents available for the assessment: HQ Actual Financials (HQ_actuals.xlsx), Department Forecast Submissions (Forecast_Submissions.xlsx), Cost Center Reporting Mapping (Cost_Center_Map.csv), and Anaplan Consolidated Reporting Export (Anaplan_export.xlsx). Using these reporting files, review the current forecast submissions and consolidated reporting totals to identify where reporting inconsistencies, mapping variances, duplicate reporting structures, or classification differences may be contributing to reconciliation gaps across departments.

The final deliverable should be a finance reporting assessment package that summarizes the key reconciliation findings, identifies any material reporting variances or inconsistencies discovered during the review, quantifies the financial impact where applicable, and highlights any reporting or consolidation risks that may require additional review from Finance leadership before Q4 planning and monthly close reporting cycles continue.

Iterations & Learning:

Iteration 0 (Initial Output — Static Format):
Claude delivered comprehensive analysis but output was delivered as unformatted text. The analysis was insightful but lacked structure—no headers, no tables, no visual hierarchy. The deliverable was not dynamic; it required manual formatting to become presentation-ready.
→ Fix: Requested Claude output in structured Excel format with tables, headers, and conditional formatting. Specified exact worksheet tabs (Executive Summary, Reconciliation Assessment, Hierarchy and Mapping, Department Hierarchy, Forecast Submissions, Cost Center Map, Anaplan Export).

Iteration 1 (What Failed):
Claude surfaced every variance regardless of size. Output was overwhelming; leadership could not prioritize what to fix.
→ Fix: Added filtering logic to flag only material variances.

Iteration 2 (What Improved):
Better filtering, but explanations were generic. Claude identified gaps but could not explain why they existed.
→ Fix: Fed prompt business context,

recent cost center changes, reporting line mergers, historical patterns.

Iteration 3 (Final):
Prompt returned actionable insights with root cause explanations grounded in business reality.

Result:

Delivered a comprehensive finance reporting assessment identifying $477,000 in material variance exposure across departments. The executive summary supported a leadership meeting to discuss reconciliation issues and prioritize fixes. The detailed analysis file then guided the correction of entries, providing a roadmap for which entries needed adjustment and which cost center mappings required updating.

Disclaimer: The provided documents are samples. Confidential data has been redacted and replaced with fictionalized information.

KEY LEARNINGS

What I have Learned About Effective AI Prompts in Finance:

Context is everything. AI outputs are only as good as the inputs and business context you provide. Generic prompts produce generic answers
Iteration matters. First-version prompts rarely nail the problem. Test, validate, refine.
Filtering and thresholds are critical. Without filtering, AI surfaces noise. Add business logic to surface only material insights.
Validation is non-negotiable. I never ship AI output without reviewing it against actual data and business reality.
Combine AI with human judgment. AI is a tool for synthesis and pattern-finding. Your analytical judgment, business knowledge, and decision-making are irreplaceable.