The firms getting the most from AI are not buying better tools, but building better systems for organizing institutional knowledge.
Your firm is about to waste money on artificial intelligence, and it will not be the software's fault.
Here's how it usually goes. Someone in leadership watches a demo at a conference, gets excited, signs a contract, and rolls out a shiny new chatbot assistant to the staff. Somebody asks it a question about the firm's QA/QC process. The tool answers with complete confidence and gets it wrong, or worse, stitches together an answer from a 2014 slide deck and three conflicting versions of the same checklist. Within a month, the staff has quietly gone back to asking the person in the cube next door, and the tool becomes another line item nobody wants to defend at budget time.
I have spent the last several months building a firmwide knowledge system at The Wooten Company, and I can tell you exactly where that failure starts. It starts in your folders. Not in the generative model, not in the license tier, not in the Copilot prompt training nobody attended. The tool you bought is a retrieval engine, and a retrieval engine can only retrieve what you gave it. If what you gave it is 20 years of digital sediment, you did not buy an assistant; you bought a very expensive way to search your own mess.
I want to be blunt about what most firms are sitting on, because Wooten was sitting on it too. The difference is that our leadership chose to attack the problem ahead of the trend, before buying anything new. We had 13 or more root folders with countless subfolders that grew organically as the firm grew, named by whoever created them, organized by whatever made sense that year. Multiple documents claiming to be the current version of the same procedure. Standards that live in one practice lead's shared OneDrive and nowhere else. Critical knowledge trapped in PowerPoint slide decks that were built to be presented once, not referenced forever.
A new hire cannot find anything in that environment without asking a veteran, and the learning machine you just bought is the newest hire you have. It knows nothing about your firm except what it can read, and it cannot walk down the hall.
How to tell if your firm is actually ready for AI
So before you spend a dollar on any of this, run one test. Pick a procedure your firm genuinely relies on, something like how you close out a project or when a QC review is required. Now ask whether a person who started yesterday could find the current, authoritative answer in under two minutes without asking anyone. If the answer is no, stop shopping. You have found your actual project, and it costs a lot less than the software does.
Fixing it is not glamorous work, which is probably why nobody writes about it. We collapsed our structure down to six root folders with very clear purposes. We established that every document has exactly one authoritative home, and every other place it appears is a link, not a copy. We separated the stable stuff, the principles and expectations that change rarely, from the living stuff, the how-to guidance, and lessons learned that change often, because those two kinds of knowledge age differently and a retrieval tool needs to know which is which. We gave folders an owner, because a folder without an owner is a folder that rots. And we rewrote fragmentary content into consistent, self-contained documents, because a machine reading a slide that says "Remember what happened on the Henderson job" retrieves exactly nothing useful, and frankly neither does a human three years from now.
None of that required buying anything. All of it was the difference between a tool that works and a tool that embarrasses itself and the person who pushed for it.
Why more data does not always make AI better
There is a second lesson buried in this, and it is the one that runs against the grain of every vendor pitch you will hear this year. The pitch is always to connect and integrate everything, to pour it all into some mysterious data lakehouse. Every drive, every inbox, every project file, and apologies to Tolkien, one tool to rule it all.
We deliberately did the opposite. Active project files stay out of the knowledge system entirely, because project data and firm knowledge are different animals with different owners, different lifespans, and different risks. Copyrighted material from standards bodies like ICC, AASHTO, and others stays out too, replaced by pointer pages that tell people where the licensed copy lives, because ingesting a code book you license is a legal problem waiting for a discovery request. Individual names stay out of procedures, because procedures should outlive the people who wrote them. Every one of those exclusions made the system better, not worse. Our restraint turned out to be one of its greatest features.
I will grant the objection I can already hear, which is that this sounds like a lot of unbillable time. It is, and it takes a leadership team willing to see it as an investment rather than overhead, which is exactly the bet Wooten made. So far, the entire effort has run on software licenses we already owned, which means the price of admission was structure and discipline, not dollars.
Compare that cost to fixing the errors that arise when a PM works from the outdated version of a template and to the claim that follows. As cliché as it sounds, well-meaning mentors tell young designers and engineers constantly that the quality of the output depends on the quality of the inputs. Then firms feed their own institutional knowledge into a machine in whatever condition it was left in last and act surprised when the answers are garbage. Ironically, the industry that preaches QA/QC on every deliverable should not need this explained.
Why buying new technology is not the same as making progress
After 35 years in this business, I have watched firms buy their way past CAD standards they never wrote, intranet hubs nobody ever opened twice, BIM execution plans they never followed, and project management systems nobody configured and almost nobody adopted.
The pattern is always the same. We purchase the tool because purchasing feels like progress, and we skip the structural work because structure can be overwhelming. The current trend of internal AI assistants is no different except in one respect: this time the tool will actually show you, in real time and in front of your entire staff, exactly how disorganized you are. It will answer every question with whatever you fed it. That is not a flaw. That is the most honest performance review your file system has ever received.
So here is your Monday morning assignment, and it does not involve a demo or a sales call. If your firm fails the two-minute test I suggested earlier, find out why. Open your on-prem or cloud server, the main one where the real work lives, and count the root folders and how many levels deep the subfolders go. Ask who owns each one. Check the last modified dates. Count how many versions of your most important procedure or template exists. Then decide whether you are ready to pay an expensive consultant or buy anything at all.
The firms getting real value from these tools right now are not the ones with the biggest budgets. They are the ones that fixed their folders first.
 |
Kraig Kern, CPSM is the innovation and integration lead at The Wooten Company. Contact him at kkern@thewootencompany.com. |
