The 5,000-Project Memory Problem

The Part of AEC Project Management Nobody Draws on the Org Chart

Walk into most mid-sized AEC firms and the real project management system isn't the software on the server. It's a person. Every firm has him: the principal or VP who can tell you what you charged on the Henderson job, why the lift-station design got value-engineered, and which subconsultant to never call again, faster than any database can. And he's three years from retirement.

That person is the part of AEC project management nobody draws on the org chart. He looks like healthy seniority. He's actually a single point of failure: the firm's primary index for thirty years of project knowledge, held in one head. When the most load-bearing layer of your project management is someone's memory, you don't have a system. You have a countdown.

Before we talk about software, it's worth being precise about what AEC project management actually includes, because the part that matters most rarely shows up in a tool comparison, and there's a specific order to protecting it.

AEC Project Management Is Also Memory Management

AEC project management is the discipline of planning, coordinating, and delivering architecture, engineering, and construction projects across their full lifecycle: scope, schedule, budget, RFIs, submittals, QA/QC, closeout. That's the visible layer. There's an invisible one the software category ignores: every new project gets priced, scoped, and de-risked using the firm's accumulated memory of the projects that came before it. And that memory mostly lives in people.

Search "AEC project management software" and you'll get ten platforms: Deltek, Procore, BQE Core, the usual list. Every one of them runs the visible layer well. None of them manage the asset that actually de-risks your next proposal: the institutional memory of what worked, what blew up, and what you'd never bid that way again. The tool roundups treat project management as a software-selection problem. The hardest part was never the software.

So AEC project management has two layers: the visible one (schedules, RFIs, submittals) and the invisible one, which is the firm's memory of every project it has ever delivered. The second layer is the one quietly running on a person.

The Math That Makes One Head Load-Bearing

AEC firms aren't short on money. Operating margins recently hit a 10-year high of 21.4%.¹ They're short on people: 59% of architecture and engineering firms name the availability of qualified candidates as their top talent challenge.¹ When you can't easily replace experience, every senior who holds it becomes load-bearing.

Meanwhile, the rush to AI is already on. 53% of A&E firms now use AI tools, up from 38% the year before.¹ Firms are buying the technology faster than they're organizing the data underneath it, and the order matters more than anyone admits.

Deltek Clarity 2025, A&E firms:

• Operating profit margin: 21.4% (a 10-year high)
• Naming qualified-candidate availability as the top talent challenge: 59% of firms
• Now using AI tools: 53% (up from 38% the year before)

Now layer in age. The median age of the U.S. construction labor force is 42, a year older than the typical American worker,² and about 22.7% of construction workers are over 55.³ Those are trades-inclusive numbers, so read them as direction rather than precision. But the direction points at the senior tier, and the senior tier holds the project memory.

There's a hopeful counter-current here. Gen Z entry has been climbing, which helps the pipeline at the bottom. That's real. It just doesn't help you at the top, where the judgment lives. The risk concentrates in the few people you can least afford to lose— not in total headcount.

"But We Have All the Files": Why the Archive Doesn't Save You

Having the files is not the same as having the knowledge. A complete project archive tells you what was delivered. It almost never tells you why. And the "why"— the reasoning behind a value-engineering call, the assumption that made a fast-tracked schedule safe— is exactly what walks out the door.

Research has found that roughly 42% of institutional knowledge is unique to the individual who holds it: never written down, never filed, never searchable.⁴ Project files alone don't survive a retirement.

So why does "we have all the files" feel like such solid ground? Because your server really does have the drawings, the specs, the closeout documents. What it doesn't have is the index: the human who knew which of the four "final" folders was actually final, and why the third one got abandoned halfway through. An archive is storage. Knowledge is retrieval. Most firms have confused the two for years.

Seen clearly, this stops being an IT housekeeping chore and starts looking like treating knowledge as a governance risk: a continuity exposure that belongs in a board conversation, not a backlog.

What Actually Walks Out the Door (Tacit vs. Explicit)

What leaves with a retiring senior is rarely the explicit knowledge. The specs, templates, and CAD standards are already in the system. What leaves is the tacit knowledge: when to overrule the calc, how to read a client's silence on fee, which assumption on a fast-tracked job is safe to make. That judgment was never a document.

Explicit knowledge is what a firm writes down: procedures, specs, details, checklists. Tacit knowledge is the judgment behind it. One is capturable. The other was earned across thousands of projects and lives in instinct.

Explicit knowledge lives in the system and is capturable:

• Spec sections and CAD standards
• Project templates and proposal boilerplate
• Closeout documents and as-builts
• QA/QC checklists

Tacit knowledge lives in the person and is hard to capture:

• When to overrule the calc and trust the field
• Which fee wins the work and still holds margin
• How to read a client's silence on a change order
• Which reviewer to route a tricky calc to

Here's the part most succession plans miss. Knowledge loss doesn't start on the retirement date. It starts years earlier, the day the senior quietly stops mentoring and starts coasting toward the exit.

This is also why you can't simply hand the problem to AI. A modern AI model is like a brilliant but inexperienced intern: enormous capability, zero institutional knowledge. It can draft, summarize, and search at speed. It can make words. It can't make meaning out of judgment it was never given. The intern needs the senior's context before it's worth anything on your projects.

The Wrong Fix: Pointing AI at a Messy Repository

Pointing an AI search tool at a disorganized file server doesn't preserve knowledge. It produces fast, confident, wrong answers. Microsoft Copilot, Glean, and custom retrieval systems all sit on top of what you already have (SharePoint, Panzura, the network drive), and their answers are only as good as the metadata and data hygiene beneath them.

It's garbage in, garbage out, and quality in, quality out. Feed AI a folder structure nobody has trusted since 2014 and it will confidently cite the wrong precedent on a live proposal. A confident-wrong answer is more dangerous than no answer at all, because someone acts on it.

Why bolting AI onto bad data backfires:

• Retrieval tools inherit your repository's mess; they don't clean it up.
• Bad metadata produces plausible answers pointed at the wrong source.
• The faster the tool, the faster a wrong answer reaches a client.

The vendor pitch usually skips this step. One AEC-focused AI vendor's widely shared post⁵ names the aging-workforce problem correctly, then jumps straight to a five-point "buy our AI" shopping list, with no mention of whether your data can support any of it. That reflex is how firms discover the hidden costs of rushing an AI rollout: you pay twice, once for the tool and again to fix the foundation you skipped.

The Right Sequence: Capture → Metadata → Retrieval → AI

The firms that actually preserve project knowledge follow an order. Capture the senior's reasoning before they leave. Organize it with metadata and a taxonomy. Make it retrievable. And only then layer AI on top. Reverse that order— AI first, data later— and you've built a fast way to retrieve knowledge you never captured.

1. Capture before exit. Run structured debriefs of the senior's reasoning, not file dumps. Record the why behind the landmark jobs: the calls, the close ones, the never-agains.
2. Metadata and taxonomy. Tag and structure the archive so a human can find the right precedent in seconds. Findable beats complete.
3. Retrieval. Get clean search working on organized data before any AI sits on top of it.
4. AI. Now layer retrieval AI onto inputs you trust. On clean data it's leverage. On messy data it's a liability.

This feels slow. It's meant to. Going fast with AI creates technical debt in human systems, and you slow down to speed up. If you want a structured way to decide where AI investment pays off first, a decision framework for when to invest in AI beats buying the loudest tool in the demo.

And here's the both/and that matters: documentation doesn't replace mentorship. It makes mentorship shorter. Capture the procedures so AI can serve them back, and keep the human relationship for the judgment AI can't hold. People are the answer. AI amplifies them. Give someone with deep domain expertise an AI tool built on clean data and it's real leverage; skip the expert and you've only automated the average.

Put a Number on It (and the First Move)

There's no clean industry statistic for what one retirement costs an AEC firm. So don't borrow one. Build your own. Estimate the rework, the slower proposals, and the win-rate dip in the 18 months after a key senior leaves, and you'll have a defensible figure your board can act on.

An illustrative model (not a statistic): Say losing your top precedent-holder adds 60 rework hours across next year's proposals, stretches each major pursuit by a week, and shaves two points off your win rate on competitive work. Put your own rates and pipeline against those three lines and the number gets real fast. The figure matters less than the fact that you built it— that's what makes it defensible to a board.

Once you've invested in capture and retrieval, those same three lines tell you whether it worked. That's how to measure whether the investment paid off without hand-waving at "productivity."

The first move isn't buying software. It's identifying the one person you can least afford to lose and running a structured capture before the calendar makes the decision for you. If mapping that path— capture, then data foundation, then the right AI— feels like one more thing on an already full plate, that's the kind of foundation work a data-first AI implementation partner handles, so the tooling finally lands on ground that can hold it.

FAQ

What is AEC project management?

AEC project management is the discipline of planning, coordinating, and delivering architecture, engineering, and construction projects across their lifecycle: scope, schedule, budget, and quality. Less obviously, it includes managing the firm's accumulated memory of past projects, which informs every new bid and design. The software category covers the first part. The second usually lives in people.

What happens to project knowledge when a senior engineer retires?

Without a deliberate capture process, the tacit knowledge (why decisions were made, not just what was delivered) leaves with them. Research has found that roughly 42% of institutional knowledge is unique to the individual who holds it.⁴ The files remain. The judgment doesn't.

How old is the AEC and construction workforce?

The median age of the U.S. construction labor force is 42, a year older than the typical American worker,² and about 22.7% are over 55.³ Gen Z entry is improving the pipeline at the entry level. The senior knowledge tier is the part that's exiting.

Can AI preserve a retiring engineer's knowledge?

AI can capture and retrieve explicit, documented knowledge well, but only if the underlying data is organized, and it struggles to hold tacit judgment. Point it at a messy repository and it returns confident, wrong answers. Fix the data foundation first. Deploy AI search after, not before.

References

1. Deltek, "What the 46th Annual Deltek Clarity A&E Study Reveals About the Architecture and Engineering Industry" (2025) — https://www.deltek.com/en/about/media-center/press-releases/2025/what-the-46th-annual-deltek-clarity-ae-study-reveals-about-the-industry
2. National Association of Home Builders (Eye on Housing), "Median Age of Construction Labor Force Holds at 42" (2025) — https://eyeonhousing.org/2025/10/median-age-of-construction-labor-force-holds-at-42/
3. U.S. Bureau of Labor Statistics, reported via Fieldwire by Hilti, "Tackling Labor Gaps: The Aging Construction Workforce" (2024) — https://www.fieldwire.com/blog/aging-construction-workforce/
4. Panopto (with YouGov), "Inefficient Knowledge Sharing Costs Large Businesses $47 Million Per Year" (2018) — https://www.panopto.com/company/news/inefficient-knowledge-sharing-costs-large-businesses-47-million-per-year/
5. Joist.ai, "An Aging Workforce: A Growing Concern for Companies and the Role of AI in Preserving Institutional Knowledge" (2024) — https://www.joist.ai/post/an-aging-workforce-a-growing-concern-for-companies-and-the-role-of-ai-in-preserving-institutional-knowledge