# Ten Questions That Reveal Your Firm's Hidden Indexes **By Dan Cumberland** · Published May 7, 2026 · Categories: AI Strategy > A hidden index is a repeatable judgment your firm depends on but has never written down— a pattern that lives in a person, not a process. It is the lookup the... ## What a Hidden Index Actually Is A hidden index is a repeatable judgment your firm depends on but has never written down— a pattern that lives in a person, not a process\. It is the lookup the senior person performs without consulting anything\. No checklist\. No SOP\. Just a glance and a verdict\. Process documents capture steps\. Indexes capture the judgment that decides which steps apply, in what order, and when to deviate\. Every firm has indexes\. Few firms know which ones they're running on\. A short list of common ones: - The fee\-estimating instinct that adjusts a number by 12% based on the client name - The RFI risk smell that flags a problem in the first paragraph - The client\-fit read that knows by the second meeting whether the project will be miserable - The detail\-reuse pattern that picks the right wall section for a given typology In professional services firms, the more tacit a firm's productive knowledge is, the harder it is for competitors to imitate it[1](/blog/blog-architecture-questions#ref-1)\. Indexes are the moat\. And the moat is undocumented\. The reason this matters now is not that knowledge management suddenly got urgent\. It's that AI changed what an unsurfaced index costs you\. ## Why This Matters Now AI deployment in a professional services firm is gated by the structure of its tacit knowledge, not by the quality of the model\. Firms that have surfaced their judgment can deploy AI usefully\. Firms that haven't, can't\. According to California Management Review[2](/blog/blog-architecture-questions#ref-2), the real differentiator in the agentic\-AI era is the tacit knowledge embedded in human judgment— not the data and not the models\. The same Berkeley analysis[2](/blog/blog-architecture-questions#ref-2) argues that operationalizing that judgment requires a *semantic layer*— the relationships among the conceptual entities your experts rely on\. Translation: AI agents can mimic your senior architect's first\-pass review only if the patterns she recognizes have names, and the names have relationships, and the relationships are written somewhere a model can read\. Larger firms are already doing this\. RIBA Journal[3](/blog/blog-architecture-questions#ref-3) reports that larger architecture practices are codifying project\-based tacit knowledge by building bespoke LLM\-based systems that index, retrieve, and synthesize practice knowledge\. That is the work\. Mid\-market firms— the $20M to $100M band— have a window\. The structural gap between firms with surfaced indexes and firms without is widening every quarter\. A solid [AI strategy for founder\-led firms](/services/ai-strategy/) is not buying a tool\. It is naming the judgment your firm already runs on, then deciding which parts of it should scale\. Pre\-AI, the cost of a missing index was rework\. Post\-AI, it is structural disadvantage\. ## How to Use the Ten Questions Run these ten questions in a single afternoon with your senior team\. Each question targets a class of decision your firm makes routinely without consulting a document\. The goal is not to answer them perfectly— it is to notice which ones the room can answer, which ones produce silence, and which ones produce three different answers\. Track three signals: - **Confident answer in the room** — the index is at least partially surfaced\. Codifying it is mostly transcription work\. - **Silence** — the index is fully tacit and at risk\. Whoever can answer it is the holder, and the firm is exposed\. - **Three different answers** — the index has drifted\. Different senior people are running on different versions, and the firm is paying for the inconsistency without noticing\. Silence is information\. So is three different answers\. These questions are diagnostic, not surveillance— the point is to make the firm survive a retirement, not to evaluate any individual\. And honestly? You can't read the label from inside the bottle\. Run it together\. See [how founders evaluate AI readiness](/for-founders/) for the broader frame this fits inside\. Question one\. ## The Ten Architecture Questions These questions are diagnostic, not prescriptive\. They are arranged from the most common indexes \(project intake, fee estimation\) to the most easily ignored \(knowledge\-loss exposure\)\. If the room goes silent on a question, you've found an index\. If three people answer differently, you've found an index that's already drifting\. ### 1\. How does your most experienced person decide which prospective projects to pursue— and what would they look for in the first ten minutes that a less experienced person would miss? **What it reveals:** The go/no\-go judgment— the project\-intake index that decides which RFPs are worth chasing and which will burn the next six months\. **If the room can't answer it:** Junior staff pursue work the seniors would have killed, and the firm wins projects it shouldn't have bid on\. **AI\-readiness signal:** If unsurfaced, AI cannot pre\-screen RFPs\. If surfaced, intake screening is one of the highest\-leverage early agents you can deploy\. ### 2\. When you estimate a fee on a project type you've done before, what are the three or four numbers your gut adjusts based on, and where do those adjustments live? **What it reveals:** The fee\-estimation heuristics— the multipliers and risk premiums senior PMs apply without consulting a model\. **If the room can't answer it:** Fees drift\. Margin varies wildly by who priced the project, and no one can explain why\. **AI\-readiness signal:** Fee modeling is one of the first deployable indexes, but only if the heuristics can be named\. An AI agent cannot infer your gut\. ### 3\. Which standard details get reused on which project types, and who decides— and is that decision happening in someone's head or in a library everyone trusts? **What it reveals:** The detail\-reuse and design\-library decisions— the rules that govern which wall section, which connection, which spec applies where\. **If the room can't answer it:** Teams reinvent details every project, and small inconsistencies show up at CA\. **AI\-readiness signal:** Per Knowledge Architecture[4](/blog/blog-architecture-questions#ref-4), AEC firms hold this knowledge in artifacts \(drawings, specifications, RFIs, meeting minutes\) rather than in documents written for transfer\. AI can index those artifacts only if the reuse logic is named\. ### 4\. When an RFI comes in, who can read it and tell within a minute whether it predicts a problem— and what are they actually pattern\-matching on? **What it reveals:** RFI risk recognition— the smell test that distinguishes routine clarification from impending change\-order\. **If the room can't answer it:** Problems show up at the punch list that should have been caught during construction\. **AI\-readiness signal:** Strong knowledge\-agent candidate\. Knowledge agents are most effective when scoped to a single workflow, discipline, or role[4](/blog/blog-architecture-questions#ref-4)— RFI triage is exactly that scope\. ### 5\. What are the early signals that a client is going to be a problem, and which of those signals do you wish you'd named before you signed the last bad contract? **What it reveals:** Client\-fit reads and red\-flag patterns— the index that decides whether a relationship will be profitable, miserable, or both\. **If the room can't answer it:** The firm keeps signing the same kind of bad contract every two years and acting surprised\. **AI\-readiness signal:** This index is often the most tacit and the most expensive to lose\. Surfacing it is a culture decision before it is a tooling one\. ### 6\. How does your firm decide which subconsultants to bring in for which project types— and what happens if the person who knows that retires next year? **What it reveals:** Subconsultant selection and vendor judgment— who plays well with whom, who delivers on a tight schedule, who needs to be managed closely\. **If the room can't answer it:** New partners default to whoever was on the last project, and the firm's bench narrows\. **AI\-readiness signal:** Documentable as a structured database before it needs to be an agent\. Start there\. ### 7\. When you staff a project, what makes a particular project architect right for a particular project— and is that judgment written down anywhere a new partner could read it? **What it reveals:** The internal staffing\-fit index— the talent\-deployment judgment that decides who runs what\. **If the room can't answer it:** Projects get the architect who is available, not the architect who is right\. Margin and morale both degrade\. **AI\-readiness signal:** Lower priority for AI; higher priority for partner\-level documentation\. This is a "document the holder" candidate\. ### 8\. What does a senior reviewer actually look at first when they pick up a set, and what catches their eye in the first thirty seconds that the team missed? **What it reveals:** Quality\-review checkpoints— the visual index a senior PA runs through before signing\. **If the room can't answer it:** Review quality varies by reviewer, and the same mistakes ship to the GC repeatedly\. **AI\-readiness signal:** Per CMR Berkeley[2](/blog/blog-architecture-questions#ref-2), this is exactly the kind of expert relationship a semantic layer makes deployable\. Name what the senior eye is looking at, and an [AI implementation roadmap](/services/ai-implementation/) for QC review becomes tractable\. ### 9\. What patterns show up in your closeouts and punch lists that, if anyone had named them at SD, would have prevented half the rework? **What it reveals:** The closeout and lessons\-learned index— the patterns the firm sees every project but never feeds back upstream\. **If the room can't answer it:** The firm is paying for the same mistakes annually\. No one tracks the cost\. **AI\-readiness signal:** Surfacing this index produces the self\-reinforcing improvement loop Friedman & Partners[5](/blog/blog-architecture-questions#ref-5) documents— use of the firm's knowledge system surfaces gaps and contradictions, which improves the system, which surfaces more\. ### 10\. If your three most senior people all left the firm in the next eighteen months, which decisions would you no longer know how to make— and have you written any of those decisions down yet? **What it reveals:** Knowledge\-loss exposure and succession risk— the index that names what the firm cannot survive losing\. **If the room can't answer it:** You're not running a firm; you're renting one from the people who hold its judgment\. **AI\-readiness signal:** This is the diagnostic itself\. The answer maps directly to which indexes need codification first\. You now have a list\. The work is what you do with it\. ## What to Do With the Answers Once you've surfaced your firm's hidden indexes, every one of them falls into one of three categories: **codify, protect, or document the holder**\. Which category is the right one depends on whether the judgment scales when written down— or whether writing it down is the thing that breaks it\. ```html-table
CategoryWhat belongs hereAction
CodifyFee models, intake heuristics, detail libraries, RFI patterns, subconsultant dataWrite it down. Feed it to a scoped knowledge agent4. Make it the firm's source of truth.
ProtectDesign judgment, client-fit reads, quality-review instinctPreserve it with deliberate friction6. Codifying these can degrade the very judgment you're trying to keep.
Document the holderIndexes that are too tacit to codify yet, but too valuable to loseCapture who carries it, what triggers its use, and what would have to change for the firm to survive their departure.
``` Not every index should be codified\. Some judgments survive only because a person makes them\. The Frontiers BEACON research[6](/blog/blog-architecture-questions#ref-6) is direct about this: over\-reliance on generative models risks skill atrophy, and preserving professional judgment sometimes requires manual sessions a tool cannot replace\. Both are true\. The triage frame is the resolution\. The first index to codify is the one whose holder is closest to retirement\. And don't wait for a perfect map before starting\. Per Friedman & Partners[5](/blog/blog-architecture-questions#ref-5), the diagnostic effect runs upstream of the productivity effect— using a knowledge system, even an imperfect one, is what surfaces the gaps that tell you what to build next\. AEC firms already hold the raw material in artifacts[4](/blog/blog-architecture-questions#ref-4)\. The question is whether you've named what the artifacts are evidence of\. Most firms won't run this diagnostic\. The ones that do are not necessarily the biggest\. They're the ones whose principals decided that the judgment in three or four heads was worth the afternoon\. ## FAQ ### What is a hidden index in an architecture firm? A hidden index is a repeatable judgment a firm depends on that lives in a senior person, not a documented process\. Examples include the fifteen\-second profitability read on a prospective project, the RFI risk smell, and the senior reviewer's first\-glance pattern catch\. Every firm has them\. Most firms haven't named them\. ### Why is tacit knowledge becoming more valuable, not less, in the AI era? Because AI amplifies whatever structure already exists\. Firms that have surfaced their judgment can deploy AI usefully; firms that haven't, can't[2](/blog/blog-architecture-questions#ref-2)\. The model quality is not the constraint\. The constraint is whether your firm's judgment is in a form a model can read\. ### How is a hidden index different from a process document or SOP? Process documents capture steps\. Indexes capture the judgment that decides which steps apply, in what order, and when to deviate\. An SOP tells you how to draft an RFI response\. An index tells you which RFI to worry about in the first place\. ### Should a 30\-person firm do this, or does it require enterprise resources? Yes— a 30\-person firm can absolutely do this\. The diagnostic itself runs in an afternoon with the senior team and requires no platform\. The implementation that follows can scale to firm size\. You don't need an internal AI team to surface what you already know\. ### What's the cost of not surfacing these indexes? Rework, repeated mistakes, unprofitable projects accepted on instinct, and a structural disadvantage as competitors deploy AI on top of codified judgment[2](/blog/blog-architecture-questions#ref-2)[3](/blog/blog-architecture-questions#ref-3)\. Pre\-AI, the cost was operational\. Post\-AI, it is competitive\. If running this diagnostic surfaces more than your firm can act on— which is the most common outcome— that's exactly the place where outside help earns its keep\. ## The Afternoon Worth Spending The principals running this diagnostic well are not the ones with the most resources\. They're the ones who decided that the judgment in three or four senior heads was worth a structured afternoon— and a quarter of structured follow\-through\. AI doesn't capture institutional memory\. It amplifies whatever structure you've already built\. Most firms won't do this work, and that's fine\. The ones that do gain a structural advantage that compounds every quarter the gap widens\. If the answers reveal more than your firm can act on alone, an outside partner who has run this diagnostic with other AEC firms can shorten the path\. Dan Cumberland Labs runs AI\-readiness diagnostics and [fractional AI leadership](/service/) engagements with $20M–$100M AEC firms— principal\-to\-principal, no platform pitch, no transformation theater\. Just the work of naming what your firm runs on, and deciding which parts of it should scale\. ## References 1. Mokhetho et al\., "Retention Management as a Means of Protecting Tacit Knowledge in an Organisation: A Conceptual Framework for Professional Services Firms" \(2023\) — [https://www\.researchgate\.net/publication/368898971\_Retention\_management\_as\_a\_means\_of\_protecting\_tacit\_knowledge\_in\_an\_organisation\_a\_conceptual\_framework\_for\_professional\_services\_firms](https://www.researchgate.net/publication/368898971_Retention_management_as_a_means_of_protecting_tacit_knowledge_in_an_organisation_a_conceptual_framework_for_professional_services_firms) 2. California Management Review \(UC Berkeley Haas\), "Tacit Knowledge Is Your Next Competitive Moat" \(2026\) — [https://cmr\.berkeley\.edu/2026/03/tacit\-knowledge\-is\-your\-next\-competitive\-moat/](https://cmr.berkeley.edu/2026/03/tacit-knowledge-is-your-next-competitive-moat/) 3. RIBA Journal, "How architects use and will use AI in 2026 and beyond" \(2026\) — [https://www\.ribaj\.com/intelligence/how\-architects\-use\-and\-will\-use\-ai\-in\-2026\-and\-beyond/](https://www.ribaj.com/intelligence/how-architects-use-and-will-use-ai-in-2026-and-beyond/) 4. Knowledge Architecture, "How Knowledge Agents Will Change the Way AEC Firms Scale Expertise and Accelerate Learning" \(2024\) — [https://www\.knowledge\-architecture\.com/blog/how\-knowledge\-agents\-will\-change\-the\-way\-aec\-firms\-scale\-expertise\-and\-accelerate\-learning](https://www.knowledge-architecture.com/blog/how-knowledge-agents-will-change-the-way-aec-firms-scale-expertise-and-accelerate-learning) 5. Friedman & Partners, "The Rise of AI Search in AEC Knowledge Management" \(2025\) — [https://friedmanpartners\.com/the\-rise\-of\-ai\-search\-in\-aec\-knowledge\-management/](https://friedmanpartners.com/the-rise-of-ai-search-in-aec-knowledge-management/) 6. Frontiers in Built Environment, "A BEACON through the walls: AI\-assisted tacit knowledge extraction from built environments" \(2025\) — [https://www\.frontiersin\.org/journals/built\-environment/articles/10\.3389/fbuil\.2025\.1674307/full](https://www.frontiersin.org/journals/built-environment/articles/10.3389/fbuil.2025.1674307/full) --- Source: https://dancumberlandlabs.com/blog/architecture-questions/