Where this gets used.
The same five layers serve very different problems. Seven we have built or scoped, each with the situation we found, what we build, and what the result looks like.
Asset and site inventories from project records
Contractors and property owners with hundreds of sites and years of closeout documentation, and an asset register that is nearly empty. Nobody has time to key in thirty line items per site for three hundred sites, so the system of record stays thin and the truth stays in the PDF.
“What is installed at this property, and is it under warranty” answered in seconds with the page it came from, and a maintenance program that can finally plan by what is actually there. Organisations that reach a current, accurate register report 10 to 20 percent reductions in maintenance cost.
- Extract every line item from closeout packs, schedules and O&M manuals
- Connect each item to its site, location, supplier and warranty
- Write assets into the CRM under the right location with a stable key
- Cite the document and page on every record
Which sites have chillers installed before 2018 that are out of warranty?
Field service intelligence
A technician on site with a fault, a phone, and no easy way to find out what unit is in front of them, which part it needs, or whether the repair is covered. Second trips and missed warranty claims are the daily cost.
Second trips drop. Warranty claims that used to be missed get filed. Every answer carries the citation, so the technician can check it and the office can audit it.
- An assistant that takes a photo and a sentence from the technician
- Identifies the installed unit from the register and the document corpus
- Returns the replacement part, the warranty position, and the source
- Drafts the work order for approval
Photo attached. Unit at 4 Kensington, plant room. What part and is it covered?
Supplier and market monitoring
Public catalogues, price lists, regulatory notices and tender portals change weekly. Someone checks them by hand when they remember, and procurement finds out about a price change after the order is placed.
Price intelligence remains the most widely adopted use of web data. Procurement sees changes the day they happen, with the source page and timestamp, instead of at invoice time.
- Collect public sources daily, within each site’s rules and the law
- Detect changes and structure the results into typed records
- Compare to last week and compute the difference
- Push a note to the people who need to act
What changed in our top 20 suppliers’ price lists this week?
Contract and obligation tracking
Agreements sit in a shared drive as PDFs. The lead times, penalties, renewal dates and duties inside them are enforced by memory, if at all.
“What did we agree with this supplier about lead times” answered with the clause, the page, and the signed version it came from. Deadlines surface before they pass.
- Extract agreements into parties, terms, dates and duties
- Connect obligations to the systems that should enforce them
- Alerts before deadlines and renewals
- An assistant that answers with the clause on screen
What did we agree with Supplier B about lead times and late delivery?
Document intake and routing
Invoices, certificates, reports and forms arrive by email and portal all day. Someone reads each one, works out what it is and who it belongs to, and files it. Mistakes are found months later.
Routine documents are filed correctly without a person touching them. The exceptions arrive with an explanation, not a mystery. Every decision is logged.
- Read and classify each incoming document
- Match it to the right customer, project or case by content
- File it and update the record it belongs to
- Route anything uncertain to a person with a note explaining why
Inbox: 212 documents since Friday. What needs a human?
Migration and consolidation
Two companies merge, or a system is retired. The old data is in a format nobody fully understands, the new system has different fields, and the same customer exists three times across both.
One system, with counts you can defend and an audit trail of what was merged and why. Every merge can be undone.
- Read the old system in place
- Reconcile it with the new: the crosswalk, the disagreements, the fixes
- Resolve duplicates by content and record every merge
- Load with a dry run first and verification from the target after
How many customers will we have after the merge, and where do the two systems disagree?
Research corpora
Regulatory filings, standards, scientific literature and internal reports. Thousands of documents that a team needs to know before it decides, and no way to ask a question across all of them.
Teams find what is known before they decide, and can verify every claim in one click. When the corpus is silent, the answer is “not stated,” not a guess.
- Index every page for meaning, with structural chunking
- Connect entities across documents into a graph
- Route each question to search or graph, whichever answers it best
- Cite document and page on every result
Which standards reference corrosion allowances for buried steel pipe, and do they agree?
Tell us the question you cannot answer today.
An assessment inventories your sources, samples them, and tells you in writing whether your data can answer it, with what confidence, and what it would take.