← case studies
Published
July 2026
How One PE Firm Clawed Back 40+ Hours a Month With AI
A PE firm's associates rebuilt 14 P&Ls by hand every month. An AI now reconciles any format into one source of truth — reclaiming ~42 hours a month across the firm and its portcos, and catching trends early.
~42 hrs/mo
Reclaimed from reconciliation
< 4 weeks
Implementation Time
Not disclosed
Project Cost
the challenge
A mid-market private equity firm with a portfolio of 14 operating companies was losing a senior associate's week, every month, to portfolio reporting — and the GPs still didn't have a clean view of trends across the book. Every portfolio company submitted financials in a different format: different chart of accounts, different periods, different definitions of EBITDA, different file types. The 'reconciliation' work was actually an associate manually rebuilding 14 P&Ls into one comparable view, every month. GPs got trends late, and the trends were often wrong because the underlying reconciliation had errors nobody had time to catch. There was no single source of truth for any portco, let alone the portfolio — every conversation about a company started by re-establishing what the numbers even were. Timely portfolio visibility drives everything from add-on capital decisions to LP communications.
what they built
DealSage layered an AI reconciliation workflow onto the firm's existing process — no rip-and-replace. The firm had first engaged DealSage for new-deal triage (financials arriving as PDFs and Excels across different years and formats); the platform already had reconciliation functionality, which was tuned for the firm, and a custom Excel plugin was co-developed so users could drag-select years from spreadsheets or PDFs and reconcile them. The technical core marries the AI's strength at extracting and structuring messy data with a deterministic P&L schema: Google Gemini parses the documents, a GPT model (GPT-5.4) structures the output into a standardized P&L 'asset' with every line item tagged (COGS, OPEX, gross profit), periods standardized to ISO formats, and totals validated so errors surface for human review. Once it worked for triage, it was extended to monthly portfolio reporting across the firm's 14 portfolio companies: portcos email their latest financials straight to the DealSage agent — no login required — and the platform appends each month to a source-of-truth P&L per asset, so per-company trends accrue automatically with no firm template to fill in. A self-hosted OpenAI open-weight model triages the portco emails so the querying model (Claude Opus 4.8 or GPT-5.4, user's choice) can answer questions with the context the portco team already gave. On the roadmap: cross-portfolio benchmarking (comparing a cost spike at one company against similar holdings) and direct integration with portco ERP systems.
The firm first came to DealSage with a deal-triage problem: financials arrived as PDFs and Excels across different years, formats, and definitions, and associates were rebuilding them by hand to evaluate new opportunities. DealSage already had reconciliation functionality and tuned it for the firm, then co-developed an Excel plugin that connects to the platform so users can drag-select years — from PDFs or spreadsheets — and reconcile them. The technical core marries the AI's strength at extracting and structuring messy data with a deterministic P&L schema: Google Gemini parses the documents, a GPT model structures them into the P&L 'asset,' every line item is tagged (COGS, OPEX, gross profit), periods are standardized to ISO formats, and totals are validated so errors surface for human review. Querying runs on the user's model of choice — Claude Opus 4.8 or GPT-5.4 — and a separately self-hosted OpenAI open-weight model triages the portco team's emails so that context travels with the numbers. Once triage worked, the firm realized the same painful reconciliation ran every month across its 14 portfolio companies. DealSage extended it: portcos email financials directly to the agent (no login), the platform appends each month to a source-of-truth P&L per asset, and trends accrue automatically. The time saved — roughly 42 hours a month across the investment team's juniors and the portcos' finance teams — mattered less than the shift it enabled: the firm moved from a reactive posture, only getting involved when something was on fire, to a proactive one with no surprises, freeing the deal team to actually talk with portcos and guide them where needed. Data security was addressed via SOC 2 certification and a human review step kept in the loop. Time to live use was about three weeks.
best fit for
Lower-middle to mid-market PE firms (~$100M–$2B AUM, 5–25 investment professionals) running platform or roll-up strategies, drowning in monthly portfolio reconciliation across 10+ portcos that submit financials in inconsistent formats — plus deal teams doing manual financial extraction during triage.
Ai ROLE
Google Gemini parses inbound PDFs and Excels; a GPT model (GPT-5.4) structures the extracted data into a standardized P&L 'asset' — every line item tagged to schema (COGS, OPEX, gross profit), periods ISO-standardized, totals validated so errors surface for review; and the user's model of choice (Claude Opus 4.8 or GPT-5.4) answers natural-language questions across the single source-of-truth P&L. A separately self-hosted OpenAI open-weight model triages the portfolio companies' emails, so when a GP asks why a number moved, the answer already carries the context the portco team explained.
impact
~42 Hrs/Month
Monthly reconciliation time reclaimed across 14 portfolio companies — roughly 42 hours a month, split between the investment team's juniors doing the reconciling (~1.5 hrs per portco) and the portcos' own finance teams prepping data into templates (~1.5 hrs per portco).
Live in ~3 Weeks
From kickoff to live use on real monthly reporting cycles.
Reactive → Proactive
GPs moved from reactive cleanup — stepping in only when something was already on fire — to proactively managing the book with no surprises, freeing time to actually talk with portcos and guide them.
implementation complexity
Largely productized — reconciliation functionality already existed in DealSage and was tuned for the firm; the one custom build was an Excel plugin connecting spreadsheets to the platform. AI extraction (Google Gemini) is paired with GPT-5.4 structuring into a deterministic P&L schema (tagged line items, ISO-standardized periods, validated totals) and a human review step, with querying on the user's model of choice (Claude Opus 4.8 or GPT-5.4) and a self-hosted OpenAI open-weight model for email triage. Live in ~3 weeks; SOC 2 certified.
Harry Ratcliff
Co-Founder & CEO of DealSage, an AI-native deal-intelligence platform for middle-market PE, corporate development teams, VCs, and family offices. 10+ years in middle-market M&A at J.P. Morgan.
Get an intro
Talk to this teamindustry
Financial Services
business organization
Finance & Accounting
Executive & Strategy
AI TYpe
Document Processing & Extraction
Data Synthesis & Reporting
Process Automation (RPA + AI)
value type
Time Savings
Cost Reduction
frequently asked questions
How did a mid-market PE firm cut monthly portfolio reconciliation time with AI?
The team layered an AI reconciliation workflow onto the firm's existing process with no rip-and-replace: portfolio companies email their financials straight to an AI agent (no login), which extracts and structures them into a standardized, source-of-truth P&L per company, with trends accruing automatically. That reclaimed roughly 42 hours a month across the investment team's juniors and the portfolio companies' finance teams, across 14 portfolio companies.
What AI models and tools were used?
A multi-model platform paired with a deterministic P&L schema: Google Gemini parses the documents, a GPT model (GPT-5.4) structures them into a tagged P&L asset (COGS, OPEX, gross profit) with periods standardized and totals validated for human review, and querying runs on the user's choice of Claude Opus 4.8 or GPT-5.4. A separately self-hosted open-weight model triages portfolio-company emails, and a custom Excel plugin lets users drag-select and reconcile years from PDFs or spreadsheets. The platform is SOC 2 certified with a human review step in the loop.
What results did the firm achieve?
About 42 hours of manual reconciliation reclaimed each month across 14 portfolio companies, and a single source-of-truth P&L per company with trends accruing automatically. More importantly, the firm shifted from reactive cleanup — stepping in only when something was already on fire — to proactively managing the book with no surprises, freeing the deal team to actually talk with and guide portfolio companies.
How long did the implementation take?
About three weeks from kickoff to live use on real monthly reporting cycles.
Who is this approach best for?
Lower-middle to mid-market PE firms (about $100M–$2B AUM, 5–25 investment professionals) running platform or roll-up strategies, drowning in monthly reconciliation across 10+ portfolio companies that submit financials in inconsistent formats — plus deal teams doing manual financial extraction during triage.
Have a similar challenge?
A network of pre-vetted applied AI experts and AI-native service organizations.
©2026. All rights reserved
