3.5× more documents filed per person per day
Client documents read, classified and filed into the practice system, with low-confidence items queued for a person.
210
documents filed per person per day
60
→
210
3
keying errors per 1,000 docs
14
→
3
6 days
month-end close
9 days
→
6 days
Illustrative case study for a fictional client. Figures show how this layout presents results; they are not real outcomes. Results vary.
The problem
Penhallow & Reyes, a 30-person accounting practice, receives tens of thousands of client documents a year by email and upload. Staff renamed, classified and keyed them by hand — about 60 documents per person per day in busy season, with a keying error in roughly one document in seventy.
What we built
A document pipeline now classifies each upload, extracts the fields the practice needs and files it in the right client folder. Anything below the confidence bar lands in a review queue with the document shown next to the extracted fields. An internal copilot answers procedure questions from the firm’s filing guide, with sources.
The workflow
Hover or focus a step. The amber step is where a person signs off.
Penhallow & Reyes · workflow (illustrative)
- Documents arrive through the client portal or by email.
- Each document is classified and the needed fields extracted, each with a confidence score.
- Confident fields go through; anything else is queued for review.
- The reviewer sees the document next to the fields and fixes what's flagged.
- The document is filed in the client folder and the onboarding checklist updated.
01
Before
Every document opened, renamed and keyed by hand
Errors found weeks later at review
Procedures lived in one senior person’s head
02
After
Routine documents filed without anyone opening them
Low-confidence fields reviewed side by side, with the reason
Procedure questions answered with a link to the source
Trend after launch
Documents filed per person per day (illustrative)
Stack (named as text only)
Client portal uploads, Microsoft 365, the practice management system's API, Python on Azure in the firm's tenant, a hosted LLM under a no-training business agreement
"It's careful in the right places. Anything it isn't sure about lands on someone's desk with the reason why."
Dana Reyes
·
Partner, Penhallow & Reyes
· demo testimonial
Next case study
Formlark
Trial-to-active up from 31% to 47%
Ready when you are
Your first agent.
Four weeks.
Start with a free 30-minute automation audit. You leave with a shortlist, a cost estimate and no obligation.