Use case
A similarity engine matches each new case with the firm’s previous ones, and every conclusion traces back to a named rule that a lender’s reviewer can read.
The ontology
The ontology has three layers.
1 The vocabulary: what exists
The classes of objects in the firm’s business:
- Assets: site, gas turbine, steam turbine, heat recovery steam generator, generator, equipment model, manufacturer.
- Engagements: due diligence engagement, independent engineer’s report (IER), portfolio.
- Parties: lender, sponsor, operator.
2 The links: how it holds together
Typed relations, which are never confused with one another:
- “GT1 feeds HRSG1” is a physical link.
- “GT1 is a 7F model” is a manufacturer link.
- “The engagement covers the site” and “the lender finances the project” are contractual and financial links.
These links are what lead from an inspection to its effect on the debt.
3 The rules: what a senior engineer checks
For example, “every planned outage duration is compared with similar outages already completed”, or “every maintenance deferral requires a signed approval”. Each rule produces findings that an engineer validates.
The example that makes it concrete
A line in the maintenance plan (“GT1 inspection, 12 days”) says nothing on its own. The ontology knows that GT1 feeds a boiler that feeds the shared steam turbine. It deduces that the unit loses 306 MW, not a third of its output, and knows which lender carries that risk. No document in the data room contains this path; the ontology does.
What it is not
- Not a document store: the reports stay where they are.
- Not a fixed IT data model: the classes are data, which the firm evolves itself.
- Not an AI that guesses: everything is declared, versioned and verifiable.
The actors
The operator, the due diligence firm, the bank and the trader look at the same unit, but each needs different data.
| Actor | The data they need |
|---|---|
| Operator | Hour counters, last inspection, manufacturer interval |
| Due diligence firm | Planned schedule, completed inspections, actual durations and costs |
| Bank | Unit capacity, 2×1 configuration, production plan |
| Trader | Outage calendar, lost capacity, forward sales, price curve over the period |
The questions
Each actor asks their own question. Brought together in one model, the data gives four answers in a few clicks.
| Actor | Their question | What they find at Alder Creek |
|---|---|---|
| Operator | Are my turbines on schedule? | Gas turbine 2 has run 8% past its inspection interval, while the plan schedules it for 2027. |
| Due diligence firm | Is the seller’s maintenance plan credible? | The 2027 inspection is planned at 12 days and 420 kUSD; the same one in 2025 took 15 days and cost 465 kUSD. |
| Bank | What does maintenance cost the output that repays the debt? | 306 MW lost during the outage: 2.4% of annual output according to the plan, 3% according to history. |
| Trader | Which output have I sold that I will not be able to deliver, and when? | About 22,000 MWh to buy back if the inspection takes 15 days instead of 12. |