Win and Deliver More Projects with Governed AI

The seal and the fee: how a design firm actually operates

The principle

An architecture or engineering firm sells something unusual: documents a licensed professional signs and answers for under statute. Strip the mystique and the operating model is a project machine. Fees are set early, at negotiation, before the design unknowns have surfaced, and spent late, in construction documents, where the hours actually live. Margin is decided by how cleanly a team moves from concept to coordinated, checked, sealed documents, and by how little of the fee is consumed doing work twice. Rework is margin leaving the building, and document quality that slips does not show up in the studio; it shows up months later as RFIs, change orders, and a client who remembers. The demand side makes all of this acute: engineering firm leaders reported a median backlog of 11 months in early 2026, with 88 percent of firms still carrying at least one open position (ACEC Research Institute, Q1 2026; a sentiment survey of 628 firm leaders, not a delivery metric). The work is there. The hands are not. And the binding constraint is not production seats; it is the attention of the licensed professionals who must direct, check, and seal what goes out.

The trap

The trap is chasing throughput with tools while leaving the operating model alone. A drafting add-on here, a rendering engine there, and a year later the sheets render faster but the firm still redraws details it has drawn fifty times, SOQs still consume principal weekends, and checking still queues behind the same two saturated professionals. The licenses were real. The leverage never moved, because nobody redesigned the workflow the tool was supposed to serve.

The checklist

• Name the constraint in each studio: winning work, producing it, coordinating it, or checking it. They are different problems with different fixes.
• Follow one week of principal and senior licensed hours and mark which of them only that professional could have spent.
• Count how often the firm re-solves something it has already solved: details redrawn, specifications rebuilt, narratives rewritten from memory.
• Trace last year's RFIs back to their documents, and put a number on what document quality is costing during construction.
• Choose workflows to improve, not tools to buy. A tool dropped on an unchanged workflow returns almost nothing.

Where judgment beats the tool

An hours analysis shows where licensed time goes. It cannot say which of those hours are the product. Some principal time spent slowly, on a massing study, a difficult foundation, a code interpretation, is exactly what the client is paying for, and automating it would cheapen the firm. Deciding which senior hours are the value and which are habit is a call only the principals can make.

Four outcomes that matter, and one that does not

The principle

Four outcomes justify this whole program. Win and deliver more projects without proportional hiring, so a long backlog converts to revenue instead of burnout. Increase delivery capacity without sacrificing document quality, so throughput never spends the firm's reputation or its insurance history. Turn the firm's accumulated design intelligence, the proven details, tuned specifications, calculation templates, and project histories, into controlled, reusable leverage, so the best work compounds instead of retiring with its authors. And adopt AI while protecting client confidentiality, responsible charge, and the trust a client places in a sealed set. Notice what is not on the list: adopt AI. Adoption is a means. The moment it becomes the goal, the program starts optimizing for usage instead of for the firm.

The trap

The trap is measuring the means. Licenses issued, seats active, prompts per designer per day: activity metrics reward the appearance of change while the operating outcomes sit unmeasured. A firm can hit every adoption target it sets and end the year with the same SOQ turnaround, the same redrawn details, and the same checking queue behind the same saturated professionals.

The checklist

• Write each of the four outcomes as an operating sentence with a named owner, not a slogan on a slide.
• Tie every initiative to exactly one outcome. An initiative that maps to none is a hobby.
• Baseline the outcome metrics before the first pilot, or the after will have no before.
• Retire activity metrics from leadership reporting. Keep them for operations, where they belong.

Where judgment beats the tool

Outcomes conflict at the margin: pushed far enough, throughput presses on checking capacity, and reuse presses on design fit. A dashboard will not arbitrate that tension. Where the firm sets each trade, project by project, is a leadership decision, and it is one that clients and contractors experience directly in the documents.

Who has to say yes: the six chairs in the room

The principle

Nothing durable happens in a design firm without the principals aligned, and this program touches every chair at the table. The managing principal asks whether it grows the firm without diluting what the firm's name means on a title block. The operations director asks whether delivery gets more reliable or just busier. Studio and discipline leaders ask what happens to the quality of the work that carries their seal. The digital practice leader asks whether this becomes one governed platform or another layer of tools on an estate already crowded with them. The risk manager and QA/QC director ask what it does to liability, document quality, and confidentiality. The CFO asks what it costs, what it returns, and who will stand behind the number. Six different questions, and the program has to hold a real answer to all of them.

The trap

The trap is the champion-led initiative that answers one chair. It moves fast on borrowed enthusiasm, usually the digital practice leader's, then dies in a principals' meeting the day the QA/QC director asks the liability question nobody prepared for, and the silence costs the champion a year of credibility. In a firm of licensed professionals, the unanswered chair is a veto waiting for its moment.

The checklist

• Map the six chairs to named people, including the ones who hold the role without the title.
• Write down each chair's question and the evidence that would satisfy it, before the program is proposed.
• Brief the skeptics privately before the principals' meeting, not at it.
• Give the QA/QC and risk chair a genuine design seat. Controls added at the end read as concessions; controls designed in read as competence.

Where judgment beats the tool

An org chart names the titles. It does not reveal whose no actually ends a program in your firm, or whose quiet yes moves the undecided. In a partnership of licensed professionals, authority follows the seal as much as the title, and reading that structure is judgment no tool has ever held.

Qualifications-to-Proposal: win more of the right pursuits

The principle

Qualifications-to-Proposal runs from the RFQ landing to a submitted SOQ and fee: research and go/no-go, project sheets and resumes, approach narratives, fee-proposal support, principal review, and the handoff into project setup. It is where growth is won and where enormous senior time goes to die, because most of it is spent assembling material the firm already has: project sheets scattered across old submittals, resumes six versions out of date, past-performance narratives rebuilt from memory on a deadline. The drafting layer is exactly where the causal evidence is strongest: in a preregistered experiment, access to a general AI assistant cut time on professional writing tasks by about 40 percent while raising judged quality (Noy and Zhang, Science, 2023; task-level writing results, not design work). The redesign moves assembly onto governed rails and returns principal attention to the three calls that decide everything: whether to chase, what approach to promise, and what fee the work actually needs.

The workflow, stage by stage

• Research and go/no-go: the client, the program, the competition, and the firm's relevant history arrive assembled with sources attached, so the pursuit decision takes principal minutes instead of principal evenings.
• SOQ assembly: project sheets, resumes, and past-performance material come from a curated library with provenance, current and approved, instead of hunted out of old submittals.
• Approach narrative: first drafts start from the firm's own prior approaches to similar work, edited upward by the people who will actually run the project.
• Fee support: historical effort by phase and project type arrives as evidence, and the fee judgment stays with the principal who will answer for it.
• Principal review and handoff: with assembly off the critical path, review is about pursuit, promise, team, and fee, and every win carries into project setup as a structured record of what was promised.

The trap

The trap is an SOQ machine that outruns the go/no-go. When assembly gets cheap, the temptation is to answer every RFQ, and pursuit discipline quietly dies. Submittal volume goes up, shortlist rate goes down, and unbilled pursuit cost climbs while the studios inherit projects the firm should never have chased. Faster assembly only pays when the pursuit gate holds.

The checklist

• Keep the go/no-go gate ahead of the assembly engine. No SOQ starts before the pursuit decision is made.
• Stand up the qualifications library with an owner, provenance on every artifact, and review dates, so nothing stale, misattributed, or confidential slips into a submittal.
• Hold principal review of promise, team, and fee on every submission. Speed is not a reason to skip the one review that matters.
• Hand off every win as a structured record of what was promised, to whom, by when, for how much.
• Track SOQ turnaround and shortlist rate together. Either one alone will lie to you.

Where judgment beats the tool

Retrieval can assemble everything the firm knows about a pursuit. It cannot decide whether the project fits the studio's strengths, whether the fee covers the coordination risk, or whether this client relationship is worth the unbilled hours. The most expensive sentence in this business is a scope promised well and judged badly.

Detail & Specification Reuse: stop re-solving what the firm already solved

The principle

A design firm's edge is what it has already solved: the detail that has kept water out of ten buildings, the spec section tuned across a decade of projects, the calculation template a principal actually trusts. Most firms cannot retrieve any of it, so teams redraw and respecify at production rates what the firm solved years ago. Detail & Specification Reuse turns that production memory into working leverage: retrieval of proven details and specifications with provenance, narrative and report drafting grounded in firm material, specification first drafts from the office master, and calculation support that a licensed professional verifies line by line. Done well, every project starts from the firm's best prior production instead of a blank sheet, and QA/QC reviews documents built from known-good components. One boundary is not negotiable: generated calculations and code-compliance assertions are exactly the kind of task where field evidence has been reported to show professional quality declining when AI is relied on beyond its capability (Dell'Acqua et al., 2023; a consulting field experiment, and the transfer to engineering is a caution, not a measurement). Drafting support is leverage. Unverified engineering is liability.

The workflow, stage by stage

• Detail retrieval: proven details surface by condition and performance history, with provenance and the projects they served, instead of living in whoever happens to remember them.
• Specification drafting: sections start from the office master and prior project specs, flagged where this project's conditions differ.
• Narrative and report drafting: basis-of-design narratives and report sections arrive as first drafts from firm material, for professional editing rather than professional assembly.
• Calculation support: assembly, formatting, and cross-referencing assist; every input, method, and result is verified by a qualified professional before it goes anywhere.
• Write-back: RFI patterns and post-construction lessons feed the library at close-out, so the corpus compounds instead of decaying.

The trap

The trap is retrieval without curation. Point a model at the whole project archive and it will faithfully surface the superseded detail, the spec section from a different code cycle, the assembly that caused the RFI storm two years ago, and one client's confidential project data inside another client's set. Garbage retrieved faster is garbage issued faster, and in sealed documents the cost of that speed arrives with a claim number attached.

The checklist

• Curate the library: approved details and master specifications with owners and review dates, not the whole archive.
• Wall reuse by client and project boundary, so confidential project material never crosses into another client's set.
• Attach provenance and performance history to every reused component, visible at the QA/QC checkpoint.
• Route calculations and code-compliance work to qualified professionals, always. AI may assemble; it never asserts.
• Make the write-back a closing deliverable with an owner: what the RFIs taught, filed where the next project will find it.

Where judgment beats the tool

Retrieval can find the detail from the building that resembles this one. It cannot know this site's soils, this climate's exposure, this jurisdiction's code edition, or this contractor market's habits. Knowing what transfers and what was circumstance is very nearly the definition of professional experience, and it is what the client is buying when they buy the seal.

Project-to-Principal Insight: see the fee before it is gone

The principle

Principals run the portfolio on synthesized truth: phase status, fee burn, schedule risk, staffing. In most firms that synthesis is assembled by hand on Friday afternoons from memory and optimism, which is why trouble so often arrives fully grown. Project-to-Principal Insight rebuilds the picture from live project artifacts instead: fee burn read against phase completion from timesheets and deliverable status, schedule risk read from submittal and review logs, staffing read from assignments rather than hallway recollection, construction-phase signals read from RFI and submittal streams while they can still be answered cheaply. Every number in the principals' pack drills to the artifact behind it. The fee is set early and spent late, which means trouble arrives late too, unless the instruments say it early.

The trap

The trap is synthesis that launders problems. A model summarizing project reports will happily smooth a project at 80 percent fee burn and 50 percent construction documents into confident amber prose, and a principals' meeting can sit on top of a failing project for a month without knowing it. The failure is not the summary. It is a reporting chain where no number can be challenged, because no number can be traced.

The checklist

• Give every metric in the pack exactly one source of record, and synthesize from sources, not from prior summaries.
• Read fee burn against phase completion, never against the calendar. Time passing is not progress.
• Make every principal-level number drillable to the artifact behind it, in one step.
• Escalate risk to a named owner with a date, never to a distribution list.
• Keep a principal read on the pack before it circulates. Synthesis prepares the story; it does not get to tell it.

Where judgment beats the tool

Synthesis can surface the anomaly. It cannot decide which project is worth a principal's Saturday, which client call gets made before the number is certain, or when a struggling project needs a leadership change rather than a recovery plan. Escalation is a judgment about consequences, and it belongs to the people whose names are on the title block.

QA/QC and the professional seal

The principle

In most industries quality review is good practice. In this one it is the product's warranty and a statutory duty. The seal on a document set says a licensed professional in responsible charge directed and controlled the work and answers for it, and responsible charge cannot be delegated: not to a junior, not to a subconsultant, and not to software. That is why this playbook gives QA/QC its own chapter instead of a paragraph inside governance. The profession's own ethics guidance has already drawn the line in the right place: AI-assisted report drafting can be ethical when a competent engineer thoroughly reviews and verifies the content under the engineer's direction and control (NSPE Board of Ethical Review, Case 24-2; a hypothetical ethics case interpreting the NSPE Code, not a court decision). What AI changes is what arrives at the checkpoint: drafts with provenance attached, assemblies built from proven components, calculations formatted for verification. What it never changes is who answers for the set.

The trap

The trap is review compression. Production throughput rises, checking capacity does not, and the QA/QC gate quietly thins. "The model is usually right" creeps in as an unwritten assumption, thorough verification becomes a spot check with a good conscience, and the checking discipline that took a generation to build erodes in two busy quarters. The failure surfaces later, in construction, wearing the firm's seal.

The checklist

• Write the QA/QC standard for AI-assisted content explicitly in the quality manual: same checking discipline, stated rather than assumed.
• Scale checking capacity with production throughput. A faster studio behind a fixed gate is first a queue, then a rubber stamp.
• Keep provenance visible at the checkpoint, so reviewers know what was reused, what was generated, and what was drawn fresh.
• Require qualified verification of every calculation and code-compliance assertion, whatever produced it.
• Document the review trail behind every seal, so the firm can show its discipline instead of asserting it.

Where judgment beats the tool

A checklist can enforce that review happened. Only a licensed professional can decide what thorough means for this document type, this risk, this project: where to look harder, what to recalculate from scratch, when a clean-looking sheet deserves suspicion. The judgment of how hard to look is the professional act the seal certifies, and no tool holds it.

Where the risk lives: liability, document quality, confidentiality, and client trust

The principle

A design firm's risk concentrates in four places, and AI touches all of them. Professional liability: fluent drafts that are wrong in ways a saturated checker will not catch, on documents people build from. The quality risk is documented rather than hypothetical: the same field experiment that reported in-frontier gains also reported professional quality falling when AI was relied on for tasks outside its capability frontier (Dell'Acqua et al., 2023; a field experiment at one elite consulting firm, and the transfer to engineering work is a caution, not a measurement), and in this industry the clearest outside-frontier examples are calculations and code-compliance assertions that no qualified professional has verified. Confidentiality: the same ethics case that blessed verified AI drafting also concluded that uploading client confidential information to an open AI interface without consent is unethical, treated as exposing it to the public domain (NSPE BER Case 24-2; a hypothetical ethics case interpreting the NSPE Code, not a court decision, and not legal advice). Deliverable IP: the U.S. Copyright Office's January 2025 analysis concludes that using AI to assist rather than replace human creativity does not, by itself, bar copyright in the original human expression of the output (an agency view, not a court holding, and not legal advice for your contracts). And client trust: some owners, public ones among them, now ask in RFQs and contracts how the firm uses AI (a practice observation; the research library holds no measurement of how common this has become), and the firm needs an answer it is proud of before the question arrives. None of this argues against the program. All of it argues for running it inside boundaries, with AI assisting production and documentation while licensed professionals retain responsible charge, every design decision, and the seal.

The trap

The trap is treating confidentiality as an IT setting instead of a project boundary. A tenant toggle does not know which project a document belongs to, what this owner's contract prohibits, or that a public-sector client's security clauses reach every tool that touches its data. General policy plus default settings is how a firm ends up technically compliant and actually exposed.

The checklist

• Set data boundaries at the project level, mirroring the project team, and check them against each client's actual contract.
• Keep an approved-tool list with a sanctioned path for new needs, so the shadow alternative never becomes easier.
• Write the disclosure answer before the RFQ asks, and make it one the managing principal would happily read aloud.
• Put AI-assisted content on the agenda with your counsel and your insurer deliberately, before a claim or a contract does it for you.
• Stand up an incident path that assumes a miss will eventually happen, and rehearses who says what to the client when it does.

Where judgment beats the tool

A control framework can set the floor. It cannot weigh whether this project, this client, this facility justifies more caution than the standard: a hospital, a school, and a warehouse do not carry the same consequences. Proportionality is a judgment about consequences and relationships, and it belongs to the professionals who answer for both.

The maturity path: baseline, AI-enabled, selectively AI-native

The principle

There are three honest stages, and they belong to workflows, not to firms. Baseline is where most work lives today: knowledge in people, drafting by hand, status by meeting. It is not failure; it is just a leverage ceiling. AI-enabled is the working middle: the same workflow, with drafting, retrieval, and synthesis assisting inside governed boundaries, and QA/QC exactly where it was. Most workflows should live here. Selectively AI-native is the far stage: a workflow redesigned around governed AI because its volume and structure justify it, with licensed professionals owning judgment, verification, and the seal by design; in this industry the seal makes that far stage narrower than the vendors imply, and that is fine. The profession has started building the on-ramp: the AIA's AI task force publishes a firm toolkit for responsible adoption, with maturity assessment, policy frameworks, and workflow patterns (AIA AI Firm Toolkit; practice guidance, not measured proof of outcomes). The guidance exists. The evidence, your firm has to generate for itself.

The trap

The trap is the firm-wide transformation that tries to make every studio native at once. It burns a year on platform debates, frightens the licensed professionals whose consent it needs, and usually dies in a principals' meeting. The opposite trap is subtler: staying enabled forever out of comfort, when two or three high-volume workflows have long since earned the redesign.

The checklist

• Place each significant workflow on the path separately. The firm does not have a maturity level; its workflows do.
• Earn native with evidence from enabled: volumes, error rates, and QA/QC outcomes, not enthusiasm.
• Keep judgment, verification, and the seal human at every stage, including native. What changes is where drafting and assembly happen, never who answers for the work.
• Revisit placements quarterly. Workflows earn promotion, and some earn demotion.

Where judgment beats the tool

A maturity model can place a workflow on the path. It cannot decide whether the redesign is worth the disruption this year, in this studio, with this bench and this backlog. Sequencing is strategy, and strategy is what the principals are for.

Governance that speeds the firm up, and measurement that keeps it honest

The principle

Governance here is a small set of operating patterns, each of which buys speed as well as safety. Project-level boundaries decide what AI can reach. Role-based access mirrors the project team. Approved libraries decide what drafts and details may be built from. Provenance rides along automatically, which is what makes checking fast. QA/QC review stands before anything issued, and named approval points say who releases a set. Retention rules say what happens to prompts and outputs at project close; reuse rules say what the firm may carry forward and what stays behind with the client. A checker who can see provenance checks in minutes; a designer with a sanctioned path never needs the shadow one. Then the measurement, on the same one page: SOQ and proposal turnaround, hours per sheet or deliverable on recurring project types, QA/QC findings caught internally versus by the client or contractor, RFI volume attributable to document quality, fee spent versus phase completed, and backlog months by studio. Baseline them before the first pilot, count program cost once across every workflow that shares it, expect the value to arrive after an adoption lag, and present ranges the principals can believe. And state the gap plainly: no measured architecture-specific AI productivity evidence exists in the public record. The toolkit exists, the writing experiments exist, the backlog is real; none of that is proof this program pays in your firm. Your own baseline is the only number that will ever settle it.

The trap

The trap comes in a matched pair: governance as a binder and value as a vendor slide. A policy document nobody opens, an approval queue that takes a week, and people route around both, leaving the firm with the risk and the bureaucracy at once. Meanwhile the business case borrows hours-saved arithmetic nobody measured, a principal hears one inflated number, and the whole program gets discounted, including the parts that were true. In a firm that sells checked work, the business case is a work sample.

The checklist

• Build controls into the path of work as defaults, not into a document as clauses.
• Price every control by what it buys: speed gained or risk retired. A control that buys neither is friction wearing a badge.
• Baseline every KPI before the pilot that is supposed to move it, with one owner and one source of record per metric.
• Count shared program cost once, then let each workflow carry only its own marginal cost.
• Name where freed hours go: backlog converted, checking capacity restored, pursuit coverage widened. Unassigned capacity evaporates.

Where judgment beats the tool

The measurement can prove capacity was freed. It cannot decide whether that capacity becomes converted backlog, restored margin, or breathing room for a bench that has been running hot for two years, and it cannot make the principals enforce the choice. Value realization is a leadership act. The instruments only make it visible.