40 Years of Geotech Reports, Zero Way to Search Them

"Have we drilled near this site before?"

A junior engineer asks it on a Tuesday. The honest answer is: probably, sometime in the last forty years, and the only person who knows for sure is the principal who has been here since the Clinton administration. So the question routes to him. He thinks for a minute, names a project from 2009, and points at a shelf of binders. Twenty minutes of someone's afternoon disappears into a filing cabinet.

That exchange is so normal in AEC firms that nobody counts it as a problem. But it is the problem. The firm is sitting on four decades of subsurface data, borings, lab results, foundation recommendations, settlement observations, and the only working index to it is a human being who is going to retire.

The cost is not the search. It is the senior hours.

When firms think about their document archive, they picture retrieval time: the minutes spent hunting for a file. That is the small cost. The large one is that the archive is queried by interrupting the most expensive people in the building. Every "have we seen this before?" that can only be answered by a principal is a senior hour spent doing the work of a search box.

Geotechnical reports make this worse than most domains. They are dense scientific documents that are genuinely hard to interpret even for experienced engineers. The value is buried in tables, cross-sections, and caveats, not in a title you can grep for. So the knowledge does not live in the file. It lives in the person who read the file and remembers what it meant. The archive is just the part you can photocopy.

The retirement cliff is an indexing cliff

Here is the part that should worry every firm owner. Roughly a third of the AEC workforce is over 55. The people who function as the firm's living search index are the same people heading for the door. And the industry is doing almost nothing about it: 41% of organizations rarely or never even attempt to capture expertise from retiring employees.

Think about what that actually means. The reports stay on the drive. The person who knew which reports mattered, what the soft-clay surprise on that one river-district job taught them, and why the firm stopped recommending a certain foundation type after 2011, that person leaves, and the index leaves with them. You keep the documents and lose the ability to use them. The most valuable asset the firm owns quietly converts from knowledge back into storage.

Why SharePoint never fixed this

Every firm has tried to solve it with a drive, a naming convention, and good intentions. It does not hold, for reasons that are structural, not disciplinary:

• Half the archive is not searchable text. Scanned reports, photographed logs, and old PDFs are images. A keyword search cannot read them.
• Folder trees encode one person's logic. The structure that made sense to whoever built it in 2008 is a maze to everyone else, and re-filing forty years of history never makes it to the top of anyone's list.
• Keyword search requires you to already know the word. To find the right boring you have to guess the project name, the client, or the exact term the author used. If you knew that, you would not be searching.
• The meaning is implicit. "Soft compressible soils below eight meters near the old channel" is the thing you want to find. No file is named that.

So search degrades back to its only reliable mode: ask the person who remembers. And we are back to the principal and the binders.

What actually changed: searching by meaning

The shift that makes this solvable is semantic search: retrieval based on meaning rather than exact keywords. Under the hood, a retrieval-augmented (RAG) system turns every document into numerical embeddings, stores them so they can be compared by similarity, and enriches them with metadata like author, project, and date. Practically, that means an engineer can ask, in plain language, "where have we encountered soft clay below 8 meters near a river," and get the three borings that actually match, not a list of every file with the word "clay" in the name.

Two things make this real for AEC specifically rather than a generic demo:

1. It reads the documents you could not. Modern ingestion runs OCR and layout parsing over scanned reports, so the forty years of image-only PDFs finally become queryable.
2. It is becoming purpose-built. Tools like Knowledge Architecture's Synthesis are built for AEC firms, with the Deltek and Revit integrations the rest of the stack lives in. The broader KM trend for 2026 is the move from enterprise search boxes to conversational retrieval: you ask, it answers, with the source report cited.

The result is the junior engineer's Tuesday question getting answered in fifteen seconds, by the engineer, without spending the principal at all.

The honest caveats

This is not magic, and two things keep it grounded.

Garbage in, garbage out: literally. If the archive is un-OCR'd scans with no structure, the AI has little to retrieve and will hand back confident mush. The unglamorous work of ingesting, OCR-ing, and tagging the back catalog is the project. The chat interface is the easy 10% on top.

Retrieval assists judgment; it does not replace it. The system can surface the three relevant borings and summarize them. It cannot decide what they mean for this foundation on this site. A geotech report still has to be read by someone qualified to read it. The win is that they start from the right three documents instead of from the principal's memory.

Your archive is a moat or a cost center

Step back and the strategic picture is stark. A firm's forty years of local project data is either dead weight, storage you pay your most senior people to navigate by hand, or it is the single thing no competitor can replicate: a record of what is actually under the ground in your region, learned the expensive way, one project at a time.

AI is what flips the archive from the first thing to the second. The firms that pull ahead locally will be the ones that can answer "have we seen this before?" in seconds, on every desk, long after the principal who used to answer it has retired. That is not a search feature. It is the firm's institutional memory, finally written down in a form that survives the people who built it.