Why Mid-Size Structural Firms Are Bottlenecked on Partner Attention, Not Demand
More than half of ACEC member firms turned down projects in 2025 due to insufficient staffing, and the median engineering firm now reports an 11-month backlog1. The structural engineering consulting firm doesn't need more demand— it needs senior engineer time back.
The Q4 2025 ACEC sentiment data tells the story plainly:
- 92 percent of member firms report at least one open position1
- More than half have turned down projects in 2025 because they could not staff them1
- 48 percent report a year or more of work on hand, with a median backlog of eleven months1
Macro tailwinds keep coming. ENR's 2026 Top 500 Design Firms report points to AI infrastructure and data-center demand as the dominant revenue driver for design firms this year2. Demand is growing into a sector that already has more work than it can staff.
That math has a consequence most "AI for AEC" content misses. Mid-size structural firms are demand-saturated and staffing-constrained at the same time. More leads don't help. Better handling of the leads that already arrive does.
92 percent of ACEC member firms have at least one open position. The constraint is people, not pipeline.
Where does senior engineer time go when it isn't on billable work? Reading RFPs. Scoping fit. Writing one-line responses to intake forms that should have been declined at the gate. Reviewing pursuit decisions that other partners already flagged as borderline. The work is necessary— and it consumes the exact hours the firm's bill rate is designed around.
If your firm already turns down work because you can't staff it, every hour a senior engineer spends qualifying a wrong-fit RFP is an hour stolen from the work that justifies the bill rate.
If senior engineer attention is the constraint, then lead triage is where the leverage lives. So what does lead triage actually mean inside a structural engineering consulting firm?
What "Lead Triage" Actually Means at a Structural Engineering Consulting Firm
Lead triage at a structural engineering consulting firm is the work of receiving inbound RFPs, RFQs, web inquiries, and repeat-client referrals— and deciding which ones are worth the proposal hours. Every firm does it. Most do it expensively.
The mechanical workflow looks the same at every firm in the 50–250-staff range: intake capture → scope-fit assessment → jurisdiction and code check → client-history check → go/no-go decision → proposal commitment or polite decline. The variance isn't in the steps. It's in who does each step, and how long it takes them.
Here's the math that makes this an operations problem rather than a marketing problem. HSO's industry analysis puts the average AEC RFP response at 33 hours of work3. Help Everybody Everyday's data corroborates the same number and adds a contributor count: nine people involved per response, on average4. At typical net multipliers of 2.94 to 3.54 times, every 100 hours wasted on a losing proposal requires 300 to 350 billable hours to recover the overhead impact3.
| What the average AEC RFP response costs | |
|---|---|
| Hours of work | 33 |
| Contributors | 9 |
| Loaded labor per bad pursuit | $1,600–$3,2003 |
| Billable hours required to recover 100 wasted hours | 300–3503 |
The hit-rate baseline puts this in starker light. Engineering firms have the highest average proposal hit rate in the AEC industry at 44.2 percent, with construction firms at the lowest at 37.9 percent, per SMPS Foundation data via Building Design + Construction5. Unanet's 2025 AEC Inspire Report puts the broader AEC win rate at roughly 50 percent6.
That means the structural engineering consulting firm loses the majority of pursuits it commits to. And the Help Everybody Everyday analysis adds the figure that decides the math: 67 percent of AEC RFPs cost more in response time than the winning firm earns in fees4.
HSO's go/no-go whitepaper frames the operational reality directly: "Undisciplined project chasing is likely causing great angst among staff and costing firms hundreds of thousands of dollars."7 Operators sometimes call this what it is— chasing pennies when the firm could be chasing dollars.
Sixty-seven percent of AEC RFPs cost more in response time than the winning firm earns in fees.
If the math is this stark, the obvious question is why most structural firms still triage manually. Part of the answer is cultural. Part is mechanical. The mechanical part is where the 180-person firm started.
The Redesigned Workflow— Intake to Handoff
The 180-person firm's redesigned lead-triage workflow runs as a five-stage pipeline: intake capture → AI classification → structured fit summary → partner review → go/no-go decision. AI does the first three. Humans own the last two.
Here is each stage in operational detail.
1. Intake capture. Every inbound (web form, RFP attachment, referral email, repeat-client inquiry) lands in the firm's existing AEC CRM— Unanet Cosential, Deltek Vantagepoint, or BST10, depending on the firm. Unanet's CRM has been built around the AEC pursuit-to-invoice lifecycle for more than twenty years8, and its position in the 50–250-staff segment is dominant. The CRM stays the system of record. Replacing it is not part of this workflow.
2. AI classification. A workflow orchestrator (Zapier, Make, or Microsoft Power Automate) routes each intake to an LLM classifier— an OpenAI or Anthropic API call configured against the firm's specific go/no-go criteria. The classifier extracts project type, structural system, jurisdiction and code edition, estimated scope, retrofit vs. new construction signals, client-history flags, and payment-history flags. For background on the orchestration pattern, see our guide to AI workflow automation.
3. Structured fit summary. The classifier outputs a 200-word fit summary in a fixed format: project headline, scope-fit score, jurisdiction-fit score, scope-clarity flag, client-history flag, red flags, and a suggested go/no-go recommendation with one paragraph of reasoning. The summary attaches to the CRM record automatically.
4. Partner review. A senior partner reviews the structured summary during a daily or twice-daily review pass. Reading a structured 200-word fit summary takes about 90 seconds. Reading the underlying raw RFP packet takes 12 minutes or more. That delta, multiplied across the volume of weekly intake, is where the partner time gets reclaimed.
5. Go/no-go decision and handoff. The partner approves, declines, or flags for deeper review. An approval triggers the proposal team handoff with the AI summary as kickoff brief. A decline triggers a templated, courteous response. Either way, the decision is the partner's, and the CRM logs the partner as decision-maker.
AI doesn't make the go/no-go call. AI surfaces, summarizes, and tags— so the partner walks into review with a structured view, not a wall of unstructured email.
Speed-to-lead data from broader B2B research suggests faster first-touch correlates with substantially higher qualification rates— cross-industry studies find dramatic conversion gains inside the first few minutes9. That mechanism translates to AEC in modified form: speed-to-lead in engineering services is about respect and competence signaling at first touch, not racing competitors to a close. ZoomInfo's broader sales research notes that B2B teams typically waste a third of their time on unqualified leads10— a tax this workflow is designed to remove.
The five-stage workflow makes one architectural decision explicit. AI handles the volume. Humans handle the judgment. That line isn't a convenience. It's a hard requirement set by professional licensure.
Where AI Stops and Humans Start— The PE/SE Judgment Boundary
In a licensed engineering practice, professional liability rests with the Professional Engineer (PE) of record. AI cannot make a go/no-go decision because AI cannot be sued for malpractice. AI flags, summarizes, and surfaces— humans always decide.
The right frame here isn't artificial intelligence replacing engineering judgment. It's intellectual augmentation: expanding the partner's effective capacity to review more pursuits, faster, with structured information at hand. Domain expertise stays where it belongs— with the licensee whose stamp ends up on the calcs.
What AI can do inside this workflow, without crossing the licensure line:
| What AI does | What humans do |
|---|---|
| Classify intake by project type, structural system, jurisdiction | Decide whether the firm's expertise matches the scope at the level the PE will stamp |
| Extract structured data from unstructured RFP documents | Assess relationship risk with long-tenured clients |
| Score scope fit against documented go/no-go criteria | Override a fit score with a strategic-pursuit flag |
| Surface relevant past projects and qualifications | Make the go/no-go call and own the liability |
| Draft initial fit summaries and decline templates | Sign the proposal, stamp the work |
The confidentiality wrinkle deserves attention. Client RFP data is sensitive— often under NDA. The workable pattern uses enterprise-tier API access (OpenAI's Zero Data Retention tier, Anthropic's enterprise terms) with PII redaction at intake and classifier prompts hosted on customer-controlled infrastructure where required. Most firms get to a defensible answer here inside the first week of focused work.
AI doesn't decline projects. AI doesn't accept projects. AI prepares the partner to decide faster, with better information.
With that boundary established, the question is what actually changed at the 180-person firm once the workflow was in production.
What Changed at the 180-Person Firm
After three months in production, the 180-person structural engineering consulting firm reclaimed roughly two full days a week of senior partner time previously spent on intake triage, while compressing response time on qualified inbound from days to hours.
Three things shifted, directionally:
- Partner time reclaimed. Roughly two days per week, in aggregate across the senior partners doing daily triage. That time didn't get redirected at more proposals.
- Response time compression. Qualified inbound got partner-reviewed within hours of arrival, down from a "later this week" rhythm that had crept in as backlog grew.
- Pursuit selection. Not a higher absolute hit rate. A smaller pool of better-fit pursuits, with higher confidence on each one the firm chose to commit to. Fewer proposals, more careful ones.
The Bluebeam 2025 AEC Technology Outlook places this case in industry context. Among early AI adopters in AEC, 68 percent saved at least $50,000 and 46 percent reclaimed 500 to 1,000 hours on critical tasks11. The 180-person firm is one example of that pattern— not an outlier. For more on the metrics that matter when you're evaluating outcomes like these, see how to measure AI success.
The recovered time didn't go to more proposals. It went to the work only partners can do: technical review, client relationships, mentoring junior engineers.
Every implementation has things that didn't work. Pretending otherwise is one of the AI-content patterns AEC operators learn to discount immediately.
What Didn't Work— Honest Notes from Implementation
Two parts of the workflow had to be rebuilt before production. The first was the LLM classifier's confidence. Early versions over-flagged "fit" on borderline projects because the training examples skewed toward optimism.
Here are the three failure modes that showed up consistently:
- LLM classifier calibration. Early versions over-classified projects as "go" because training examples were drawn from past pursued projects— a "yes" bias baked into the firm's own data. Recalibration required deliberately including no-fit examples and adversarial cases. An LLM trained on a firm's past "yes" decisions will replicate the firm's past bias toward "yes."
- CRM data hygiene. The classifier was only as good as the structured intake data. The firm rebuilt its intake form to capture the fields the classifier actually needed— jurisdiction, structural system, project type— rather than treating intake as a free-form notes field.
- Review cadence. The initial design had partners reviewing summaries in real time as they arrived. That cadence got switched to a twice-daily batch review. Better context-switching economics, no measurable delay impact on the qualified inbound.
An LLM trained on a firm's past "yes" decisions will replicate the firm's past bias toward "yes." Calibration requires deliberately including the no-fit examples.
The 180-person firm is one operational scale. The pattern scales down with adjustments— and scales further than most firms expect.
What This Looks Like at Smaller Scale (And the Implementation Playbook)
Below 50 staff, the AI lead-triage stack collapses to a single custom GPT trained on the firm's go/no-go criteria, fed RFP excerpts manually. Above 50 staff, CRM integration starts to pay back. Above 100 staff, full workflow orchestration becomes the default.
Three scales:
- 30–50-person firm (no AEC-specific CRM yet). A custom GPT trained on the firm's go/no-go criteria and past decisions. RFP excerpts pasted in manually for the fit summary. No workflow tool, no integration. Implementation in 2–3 weeks; ongoing platform cost under $100/month. Benefit: standardizes assessment language and shortens partner reading time.
- 50–100-person firm (Unanet, Deltek, or BST10 in place). LLM classifier integrated to the CRM via a Zapier or Make connection. Structured fit summary attaches to the CRM record automatically. Daily partner review pass. Build runs 4–6 weeks; ongoing platform cost under $1,500/month. For context on build-vs-buy on this, our AI consultant vs in-house comparison lays out the tradeoffs.
- 100–250-person firm (the 180-person pattern). Full workflow: intake → CRM → LLM classifier → structured summary → partner review interface → go/no-go handoff. Multiple classifier types (RFP, RFQ, referral, repeat-client). Build runs 6–10 weeks; ongoing platform cost under $3,000/month.
The implementation pattern itself is straightforward:
- Discovery (1–2 weeks): Document current go/no-go criteria, intake sources, and decision authority.
- Build (3–6 weeks): Classifier design, CRM integration, review interface.
- Calibration (2–4 weeks in production): Adjust classifier weights against real decisions.
- Ownership: Client owns the stack— no vendor lock-in.
Bluebeam's 2025 data confirms this is still early-mover territory: only 27 percent of AEC firms use AI for automation, problem-solving, or decision-making12, and 94 percent of those that do plan to expand investment in the next year13. Cost planning matters here. Most of the surprises live in the calibration phase, where firm-specific judgment has to be translated into classifier weights. Our writeup on the hidden costs of AI projects covers the line items most engagements underestimate.
If the math in your firm points the same direction— staffing-constrained, partner-attention-bottlenecked, more proposals than the team can responsibly pursue— the implementation gap usually isn't tooling. It's translating your specific go/no-go criteria into a calibrated AI workflow that respects the PE-of-record line. Dan Cumberland Labs helps mid-size structural and AEC firms with exactly this build, as part of our broader AI implementation services for mid-size firms.
Frequently Asked Questions
What does a structural engineering consulting firm do?
A structural engineering consulting firm employs licensed Professional Engineers (PE) and Structural Engineers (SE) who design and analyze structural systems for buildings, bridges, retrofits, and infrastructure. Clients include owners, architects, developers, and contractors. The firm carries professional liability for the engineering judgment its licensees stamp.
What percentage of AEC firms use AI?
About 27 percent of AEC firms use AI for automation, problem-solving, or decision-making, according to Bluebeam's 2025 AEC Technology Outlook12, which surveyed 1,000+ AEC decision-makers. Among AEC firms already using AI, 94 percent plan to expand AI investment in the next year13.
What is the average AEC proposal hit rate?
Engineering firms win on average 44.2 percent of the proposals they pursue— the highest in the AEC industry— while construction firms win 37.9 percent, the lowest, per SMPS Foundation data published via Building Design + Construction5.
How long does a typical RFP response take?
The average AEC RFP response takes 33 hours of work and involves nine contributors, per HSO industry analysis3 and Help Everybody Everyday's proposal hit-rate research4. At typical net multipliers of 2.94 to 3.54 times, every 100 hours wasted on a losing proposal requires 300 to 350 billable hours to recover3.
Can AI replace business development at an engineering firm?
No. AI can automate intake classification, fit assessment, and routing— but the go/no-go decision, relationship management, and proposal strategy stay with humans. Professional licensure means the PE of record owns the engineering judgment, and AI cannot carry that liability.
The Real Constraint Is Partner Attention
The 180-person structural engineering consulting firm didn't automate lead triage to chase a productivity metric. It did it because the partners needed their time back for the work only partners can do.
In a market where over half of AEC firms turn down work for staffing reasons, the most valuable thing AI does isn't generate more work. It protects the time of the people you can't replace. Technical review. Client relationships. Mentoring engineers who will carry the firm's stamp in fifteen years.
The firms that build operational discipline around partner attention now— in a 2026 market that is demand-saturated, staffing-constrained, and uneven in its AI adoption— will have a structural advantage that compounds. The math is in their favor. The pattern is becoming legible. And the constraint, finally, is the right one to be solving for.
References
- American Council of Engineering Companies (ACEC), "Engineering Business Sentiment Q4 2025" (2025)— https://www.acec.org/resource/engineering-business-sentiment-q4-2025/
- Engineering News-Record, "ENR 2026 Top 500 Design Firms: AI Boom Buoys Design Revenue" (2026)— https://www.enr.com/articles/62878-enr-2026-top-500-design-firms-ai-boom-buoys-design-revenue
- HSO, "Maximizing Proposal Win Rates for AEC Firms" (2024)— https://www.hso.com/blog/maximizing-proposal-win-rates-for-aec-firms
- Help Everybody Everyday, "Bid to Win: Proposal Hit Rate Analysis" (2025)— https://www.helpeverybodyeveryday.com/proposal-development/2505-proposal-hit-rate
- Building Design + Construction / SMPS Foundation, "How Does Your Firm's Hit Rate Stack Up to the AEC Competition?" (2024)— https://www.bdcnetwork.com/how-does-your-firms-hit-rate-stack-aec-competition
- Unanet, "Half the Battle: Why AEC Firms Are Only Winning 50% of Bids" (2025)— https://unanet.com/blog/half-the-battle-why-aec-firms-are-only-winning-50-of-bids
- HSO, "Automating Go/No-Go Processes to Optimize Pursuit Decisions" (2024)— https://www.hso.com/whitepaper/automating-gono-go-processes-to-optimize-pursuit-decisions/automating-gono-go-process-to-optimize-pursuit-decisions
- Unanet, "Unanet CRM by Cosential— AEC CRM Overview" (2026)— https://unanet.com/crm-aec/crm-overview
- GreetNow, "Lead Response Time Statistics 2026: 47 Data Points" (2026)— https://greetnow.com/blog/lead-response-time-statistics
- ZoomInfo, "Lead Qualification: How to Better Convert Inbound Leads" (2024)— https://pipeline.zoominfo.com/sales/lead-qualification
- Bluebeam, "New Bluebeam Report Shows Early AI Adopters in AEC Seeing Significant ROI Despite Uneven Adoption" (2025)— https://press.bluebeam.com/2025/10/new-bluebeam-report-shows-early-ai-adopters-in-aec-seeing-significant-roi-despite-uneven-adoption/
- Bluebeam, "AEC Technology Outlook 2025: AI Adoption Findings" (2025)— https://press.bluebeam.com/2025/10/new-bluebeam-report-shows-early-ai-adopters-in-aec-seeing-significant-roi-despite-uneven-adoption/
- Bluebeam, "AEC Technology Outlook 2025: AI Investment Plans" (2025)— https://press.bluebeam.com/2025/10/new-bluebeam-report-shows-early-ai-adopters-in-aec-seeing-significant-roi-despite-uneven-adoption/
