The workflow before#
Exactera helps companies document and defend their transfer pricing: the prices that related companies inside one group charge each other. Every new client arrives with a pile of raw material, from PDFs and Excel workbooks to scanned pages describing legal entities and their financials. Before this project, Exactera's subject matter experts read through all of it by hand, found each entity and figure, and mapped it into Exactera's models. One client could take days. Results varied from analyst to analyst, and the reasoning behind each mapping lived in notes, if it was written down at all. That's a real gap in a field where the analysis may have to hold up in an audit years later.
What we built#
We built a document agent that does the first read for Exactera's experts. It takes the client's PDFs and workbooks, finds the legal entities and financial figures, and proposes where each one belongs in Exactera's transfer pricing model. The experts stay in charge; they just stop starting from a blank page.
A first read in minutes
Documents arrive in every format, including scanned pages. The agent runs them through OCR, then extracts each entity and figure with a pointer back to the page it came from. An analyst opens a client file and finds a proposed mapping already waiting.
Review built around the expert
Every suggestion sits beside its source passage on one screen, and the expert confirms, edits or rejects it with one click. We put the human checkpoint on each mapping, not at the end of the file, because in transfer pricing a single wrong entity can unravel an audit defense.
An audit trail for every mapping
Each decision is written to a relational database with the source location, the agent's suggestion, the reviewer's call, a timestamp and the reviewer's name. If a tax authority asks why an entity was treated a certain way, the answer is one lookup away. The same log doubles as labeled data, so the team can measure extraction accuracy and improve it over time.
How we delivered it
We built the agent in a four-week sprint, with Exactera's experts involved from the first week. The sprint covered the full path from document ingestion and OCR through extraction, the review screen and the decision log. Experts tested it on real client files as it came together, and their corrections shaped how suggestions appear on screen. The result is a working foundation the team can extend into the rest of onboarding.
Before and after#
| Measure | Before | After |
|---|---|---|
| Mapping a client's financial entities | 2 to 3 days of manual review | Under 1 day, AI draft plus expert review |
| Starting point for the expert | Raw documents and a blank model | Proposed mappings beside their source passages |
| Audit trail for mapping decisions | Partial, in analyst notes | Every decision logged with source and reviewer |
What's next#
The next step is to extend the agent from entity mapping into the rest of onboarding. It could flag missing documents as soon as a client uploads, and draft a first pass of the comparables analysis. The experts would still approve the exceptions, and every decision would land in the same audit log.
