01 the business
Commercial property owner
35 staff, 140 leases and 600 supplier contracts
Leases, supplier contracts and building manuals pile up. When someone needs one clause, such as a break date or a repair obligation, they open file after file and search for words that may not match the wording used.
02 before
The problem
Long documents pile up: supplier contracts, equipment manuals, tender responses, leases and product specs. When someone needs one clause or one setting, they open file after file and search for keywords that may not match the wording used.
- Finding a clause takes hours.
- Older scanned files are not searchable.
- Comparing terms across documents is a project.
03 the build
What the AI does
- Step 1: Loads the document sets you choose from SharePoint, Google Drive, Dropbox or your document management system, including scanned PDFs.
- Step 2: Reads tables, headings and scanned pages so older files become searchable too.
- Step 3: Answers each question with a short reply and quotes the passages it relied on, linked to the page.
- Step 4: Compares documents on request, for example the payment terms across every supplier contract.
- Step 5: Keeps each document set behind the same access rules it had in the source system.
paused: waiting for a person to approve
How it was put together
- Loaded chosen document sets, including scanned PDFs.
- Read tables, headings and scanned pages.
- Answered questions with quoted passages linked to the page.
- Compared terms across documents on request, such as payment terms across all suppliers.
04 human in the loop
Where a person stays in charge
The person using an answer checks the quoted passage before relying on it in advice, a quote or a contract. An administrator controls which document sets each team can search.
- Users check the quoted passage before relying on it.
- Administrators control which sets each team can search.
- Source access rules are kept.
05 rollout
From kickoff to live in 4 weeks
- Weeks 1–2Choose and load the first document set.
- Week 3Test against questions the team already knows the answers to.
- Week 4Open to the team and add more sets.
06 after
What changes
Finding a clause, spec or past answer takes a question instead of an afternoon. Teams reuse what the company already knows, and decisions rest on the actual wording in the document.
Assume 60 document lookups a month at 40 minutes each by hand, falling to 5 minutes.
60 × 35 minutes = 2,100 minutes, or 35 hours a month saved.
What to measure
- Time to find a clause.
- Answers verified against the quoted passage.
- Cross-document comparisons produced.
- Questions with no answer found.
07 evidence
What the research says
A US federal judge fined two lawyers and their firm $5,000 for filing a brief that cited cases ChatGPT had invented.
ABA Journal, Lawyers who defended ChatGPT's fake cases must pay $5K, judge says (opens in a new tab) · 2023
Knowledge workers spend roughly a fifth of their working week searching for and gathering information.
McKinsey Global Institute, Capturing business value with social technologies (opens in a new tab) · 2012
08 lessons
Mistakes to avoid
- Relying on answers without reading the quote.
- Poor scans. Re-scan the worst ones.
- Opening every set to everyone.
09 questions
Common questions
- Can it make things up?
- It must quote the passage it used, and you check that quote. That is the safeguard against invented answers.
- Does it work with iManage or NetDocuments?
- Yes, along with SharePoint, Google Drive, Dropbox and Box.