The 50% Problem — Why Pursuit Selection Is the Highest-Leverage BD Decision
AEC firms win roughly half of the bids they pursue, and only 40% of firms use a formal go/no-go process1. That makes pursuit selection— deciding what not to chase— the single highest-leverage business-development decision an architecture firm makes each week. Using AI in architecture firms to pre-score that decision is, right now, the use case that pays back in the first quarter.
The numbers behind the problem:
- ~50% AEC win rate across the industry1
- 40% of firms use a formal go/no-go process1
- 71% of AEC firms report new-client win rates in the 25%–49% band; only 21% win above 50%2
- 11–20 contributors over 6–10 working days in a typical AEC pursuit2
Run that math against $20M–$100M firm scale and one losing pursuit consumes 200 to 400 staff-hours. Half of those pursuits lose. Most firms have no structured way to decide which ones to skip— the meeting is held, the deck is drafted, the partner says yes out of momentum. Pursuit selection is the lever. Proposal writing is the tax.
The fix isn't another meeting. It's making the meeting unnecessary by scoring the opportunity before anyone walks in the room— which is where AI earns its keep in an architecture firm.
Reframing "Using AI in Architecture" — From Design Tool to Pursuit Decision Engine
Using AI in architecture usually means design tools, rendering, or BIM plug-ins. The boring, more profitable application is in business development: feeding an incoming RFP to ChatGPT or Claude, having it score the opportunity against your firm's go/no-go criteria, and using that structured first read to decide what to chase.
Adoption inside the profession is real but uneven. RIBA's 2025 survey found 59% of UK architects now use AI in some form, up from 41% the year before3. US adoption is reportedly lower— though the primary survey data is thinner— and the design-side conversation has consumed most of the oxygen.
The most profitable use of AI in an architecture firm right now isn't designing buildings— it's deciding which buildings not to bid on. That's the territory this piece claims. And before the AI scores anything, the firm has to tell it what to score against. The criteria are where the methodology lives.
The Criteria — What Your AI Should Actually Score
The standard AEC go/no-go evaluation scores opportunities against eight criteria: strategic fit, demonstrated experience, capacity, client relationship, timeline, competitive position, financial fit, and future-opportunity potential4. The criteria most firms miss— and the ones AI is uniquely good at flagging— are legal-risk and payment-risk dimensions5.
The Treble Hook framework adds a ninth: strategic-vision alignment6. Use whichever framework matches how partners already talk about pursuits. What matters is that the criteria are weighted to the firm's strategic plan, not pasted from a vendor's PDF.
Scoring methods come in two flavors: a green/yellow/red traffic-light system or a 1–5 numerical scale measuring strength or risk per criterion7. Either works. Pick the one the partners will actually read.
Here's what the AI is doing on each criterion when the RFP lands:
| Criterion | What the AI looks for in the RFP | Risk if missed |
|---|---|---|
| Strategic fit | Building type, geography, client sector vs. firm's strategic plan | Drift into shrinking markets |
| Demonstrated experience | Required project examples and certifications | Misrepresenting capability |
| Capacity | Submittal date, project schedule, staffing demands | Overcommitting the bench |
| Client relationship | Prior work, named contacts, debriefs | Bidding against incumbents blind |
| Financial fit | Fee structure, payment terms, indemnification language | Pursuing low-margin or at-risk work |
| Legal/contractual risk | Indemnification scope, change-order language, dispute terms | Hidden liability |
| Payment risk | Solvency signals, payment cadence, retainage | Working at risk |
| Timeline/competitive | Turnaround window, competitor likelihood | Wasted effort on long-shot bids |
The criteria most firms miss are legal-risk and payment-risk dimensions— exactly the categories AI can flag from an RFP in minutes. Kill criteria— auto-disqualifiers that override every other score— are where the legal and financial dimensions earn their keep. Three to start with:
- Client litigation history against designers or contractors5
- Payment-at-risk language or known solvency concerns5
- Sub-30-day proposal turnaround on a building type the firm hasn't built before (firm-discretion add — not in the legal-risk literature)
Any one of these triggers a No-Go regardless of how green the rest of the scorecard runs. Criteria without a workflow is just a list. Here's how to run the pre-score before the pursuit meeting.
The Pre-Scoring Workflow — 48 Hours, 8–10 Criteria, 30 Minutes
Best practice is to complete the AI pre-score within 48 hours of RFP receipt, evaluate against 8 to 10 criteria, and limit the follow-up meeting to 30 minutes8. Anything more elaborate gets abandoned within a quarter.
The workflow in five steps:
- Drop the RFP into Claude or ChatGPT. Modern models ingest 1,000-plus page documents in minutes9. Use an enterprise tier (Claude for Work, ChatGPT Enterprise) for confidentiality— not the consumer free version.
- Apply the firm's criteria prompt. This is the firm's go/no-go methodology written as a prompt. It lives as a markdown file and gets reused on every pursuit.
- Receive a structured recommendation with a score per criterion, a one- or two-line evidence quote pulled directly from the RFP for each score, and an overall recommendation: Strong Go, Go with Caution, Conditional, or No-Go.
- Principal reviews and overrides any line where the AI is wrong— and logs the override with reasoning.
- A 15-minute meeting confirms Go/No-Go and, if Go, names a win theme for the proposal team.
The prompt is the methodology. Everything depends on requiring the AI to quote-and-cite from the RFP for every score it assigns— so a human can verify each line in under five minutes. The skeletal structure looks like this:
ROLE: You are a senior business development analyst at an architecture firm.
CRITERIA: [Insert firm's 8–10 weighted criteria here, with definitions.]
KILL CRITERIA: [Insert auto-disqualifiers, e.g., client litigation history,
payment-at-risk language, sub-30-day turnaround on unfamiliar building type.]
TASK: Score this RFP against each criterion (1–5 OR green/yellow/red).
For EVERY score, quote the exact line or paragraph from the RFP that
justifies it. If the RFP does not contain evidence for a criterion,
mark it as "insufficient evidence." Do not infer.
OUTPUT: Scorecard table + overall recommendation (Strong Go / Go with
Caution / Conditional / No-Go) + named win theme if Go.Every prompt has to include four elements: the firm's criteria with weighting, a quote-and-cite mandate, an output schema, and the four recommendation categories. Unanet offers a free 17-point Go/No-Go assessment tool that's a useful reference for firms not ready to draft their own prompt10. Use it to calibrate, then replace it with the firm's own.
A structured score isn't a decision. Here's what the AI cannot see— and what only a principal can.
Where AI Fails and What Only the Principal Can Decide
AI is reliable for the structured pass through an RFP and dangerous for the unstructured judgment behind it. Relationship intangibles, political timing, the firm's strategic discomfort with adjacent work, and hallucinated reads of unclear contract language are the four places where a principal's override is non-negotiable.
| Failure mode | Why it matters | Principal's job |
|---|---|---|
| Relationship intangibles | "We owe this client one" doesn't appear in the RFP | Name the relationship debt and decide |
| Political/timing signals | Incumbent vulnerability, internal champions, election cycles | Read the room around the document |
| Strategic appetite for adjacent work | The AI optimizes for past wins, not future direction | Decide what the firm wants to become |
| Hallucination on unclear language | The AI fills gaps it shouldn't | Override the score; require evidence |
The hallucination risk is real— 42% of proposal professionals name it as their top concern with AI in proposal work11. Which is why the quote-and-cite requirement isn't optional. If the AI can't quote the RFP, it can't score the RFP.
The bias risk runs deeper. An AI trained on past wins will systematically favor projects resembling those wins, which over time can entrench a firm in the markets it's trying to leave. Log every principal override with a reason. Over a year, those overrides become the firm's counter-dataset— and the bias-correction signal for the next prompt revision.
This pattern shows up outside AEC. Federal grant consultant Fielding Jezreel built a suite of AI tools for grant-writing work and is direct about the limit: "It doesn't replace a grant writer. The magic is when you've got someone with deep content expertise and you pair that with AI. Neither one of those things, I think, are as strong alone." AI scores the document. The principal scores the relationship— and that's still where most pursuit decisions actually live.
The honest framing on AI failures leads to the honest framing on AI ROI.
The Honest ROI — Avoided Cost, Not Win-Rate Magic
AI does not yet produce measurable win-rate lift in AEC proposals— only 35% of firms link AI adoption to higher win rates per the 2026 QorusDocs survey12. The ROI is in avoided cost: every losing pursuit you don't chase saves 200 to 400 staff-hours, which more than pays for a year of any AI tooling.
Vendor decks like to quote 45% win-rate lifts and 90% time savings. The independent data doesn't back those numbers yet. What it does back is throughput: AI shreds a 1,000-page RFP in minutes that used to take a proposal team two days.
The math an architecture firm CFO can actually defend:
200 staff-hours × $175 loaded rate (industry-standard senior-staff loaded cost at $20M+ AEC firms) = $35,000 saved per avoided losing pursuit. One per year covers any AI tooling spend, with margin. Run two and you've funded the program for three years.
This is the first AI use case that pays in week one. Not the speculative-future-productivity pitch. For founders thinking through how to measure this, our AI decision framework for founders and the companion piece on measuring AI success cover the metrics that hold up in a partners' meeting.
One avoided losing pursuit at $20M-plus firm scale covers a year of any AI tooling. That's the ROI conversation principals can actually defend in a partners' meeting.
Crawl, Walk, Run — Start with ChatGPT or Claude First
Start with ChatGPT or Claude and a structured prompt for the first 90 days13. Move to a specialized AEC platform— ContraVault, Loopio, Responsive (formerly RFP.io), Unanet/Cosential, or QorusDocs— only when pursuit volume exceeds five proposals per month or multi-user version control becomes a real constraint14.
The platforms aren't wrong. They're just premature for most firms— version control isn't a problem until you have a volume problem. Three triggers for upgrading:
- Pursuit volume above five proposals per month
- More than two contributors editing the same scorecard simultaneously
- Audit-trail requirements that ChatGPT export can't satisfy
For the firms still standing up the workflow, our AI implementation services help principals match the right tool to the actual constraint instead of the loudest vendor. Once the firm says "Go" and needs to draft the response, how engineering firms are using AI to write better proposals is the natural next read.
The last shift is the one most firms forget— the pursuit meeting itself has to change.
How the Pursuit Meeting Itself Changes
With a pre-score in the room, the pursuit meeting is no longer about discovering whether the RFP is a fit— that work is already done. The meeting becomes a 15-minute decision review with two outputs: a Go/No-Go ruling and, if Go, a named win theme the proposal team will lead with.
| Element | Before pre-scoring | After pre-scoring |
|---|---|---|
| Meeting length | 45–60 minutes | 15 minutes |
| Agenda | Discover, score, debate, decide | Confirm score, override if needed, name win theme |
| Principal's role | Run the discovery | Land the decision |
| Outputs | Maybe a yes, maybe a meeting-to-decide | Go/No-Go + win theme |
The meeting isn't the decision anymore. The meeting is the confirmation. Every principal override gets logged with reasoning— that becomes the firm's institutional memory and, over time, the bias-correction dataset that improves the next quarter's prompt.
Below are the questions principals and BD directors ask most often when they first run this workflow.
FAQ
These are the questions principals and BD directors raise most often when they pilot AI-assisted go/no-go scoring.
What is a go/no-go decision in an architecture firm?
A go/no-go decision is the formal pursuit-qualification step in which an architecture firm evaluates an incoming RFP against weighted criteria— typically strategic fit, demonstrated experience, capacity, client relationship, timeline, competitive position, financial fit, and future opportunity— to decide whether to commit resources to a full proposal4. Adding legal-risk dimensions (client litigation history, payment-at-risk language, solvency signals) catches threats most frameworks miss5.
What is the average win rate for architecture firms?
The AEC industry average is roughly 40–50%. The 2025 Unanet AEC Inspire Report puts the figure at ~50%1, the 2026 QorusDocs survey finds 71% of AEC firms in the 25–49% range with only 21% above 50%2, and HSO cites an industry-average proposal hit rate of around 40%15.
Can ChatGPT or Claude actually run a go/no-go evaluation?
Yes. A general-purpose model can ingest an RFP— even one over 1,000 pages— and produce a structured scoring output against firm-defined criteria within minutes9. The non-negotiable design requirement is that the prompt force the AI to quote-and-cite directly from the RFP for every score, so a principal can verify each line in under five minutes8 11.
What's the ROI of using AI for pursuit scoring?
The ROI is avoided cost, not win-rate lift. A typical AEC pursuit consumes 11–20 contributors over 6–10 working days2— at $20M-plus firm scale that's 200–400 staff-hours per pursuit— and only 35% of firms link AI adoption to higher win rates, per the 2026 QorusDocs survey12. One avoided losing pursuit per year covers the cost of any AI tooling.
How long should an AI pre-score take?
Best practice is to complete the AI pre-score within 48 hours of RFP receipt, evaluate against 8–10 criteria, and limit the follow-up review meeting to 30 minutes8. More elaborate processes get abandoned within a quarter.
Should we buy a specialized AEC platform or use ChatGPT?
Start with a general-purpose model (ChatGPT or Claude) and a structured prompt for the first 90 days13. Move to specialized AEC platforms such as ContraVault, Loopio, Responsive, Unanet/Cosential, or QorusDocs only when pursuit volume exceeds five proposals per month or multi-user version control becomes a real workflow constraint14. Unanet also offers a free 17-point Go/No-Go assessment tool worth referencing14.
Conclusion
The math is straightforward: a firm winning half its bids and running pursuits without a formal go/no-go process is leaving 200 to 400 staff-hours on the table every time it chases the wrong RFP. AI doesn't fix that with magic— it fixes it with discipline a principal already knows how to enforce.
Three changes deliver the result. Codify the criteria— eight standard, plus the legal and payment-risk additions. Run the pre-score on ChatGPT or Claude inside 48 hours of receipt, with a quote-and-cite mandate. Redesign the pursuit meeting into a 15-minute decision review with two outputs. That's the playbook for using AI in architecture firms to make pursuit selection a competitive advantage instead of a tax.
AI doesn't decide. It clears the runway so the principal can. If you're running a firm that wants a partner for the rollout, that's for founders running mid-market firms what Dan Cumberland Labs does.
⚠️ EVERYTHING BELOW IS PIPELINE METADATA — NOT PUBLISHED
References
- 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
- QorusDocs (via Stargazy), "4 Numbers Every AEC Proposal Leader Should Know in 2026" (2026) — https://stargazy.io/resources/4-numbers-every-aec-proposal-leader-should-know-in-2026
- Royal Institute of British Architects, "RIBA AI Report 2025" (2025) — https://www.riba.org/work/insights-and-resources/ai-report/riba-ai-report-2025/
- CRM Software Blog, "AEC Firms Can Make Smarter Decisions About Pursuing Work by Automating the Go/No-Go Process" (2024) — https://www.crmsoftwareblog.com/2024/01/aec-firms-can-make-smarter-decisions-about-pursuing-work-by-automating-the-go-no-go-process/
- Coleman & Erickson, LLC, "The Go/NoGo Process: Risk Analysis of Client, Project and Team for Architecture and Engineering Firms" — https://www.jwcolaw.com/highlights-from-previous-roundtables/2018/12/05/risk-analysis-of-client-project-and-team-for-architecture-and-engineering-firms/
- Treble Hook, "Mastering the Go/No-Go Decision in AEC Projects" (2024) — https://treblehook.com/blog/mastering-the-go-no-go-decision-in-aec-projects/
- CRM Software Blog, "AEC Firms Can Make Smarter Decisions About Pursuing Work by Automating the Go/No-Go Process" (2024) — https://www.crmsoftwareblog.com/2024/01/aec-firms-can-make-smarter-decisions-about-pursuing-work-by-automating-the-go-no-go-process/
- AutoRFP.ai, "AI Go/No-Go Prompt for RFP Tender Analysis" (2025) — https://autorfp.ai/downloads/ai-go-no-go-prompt-rfp-analysis-download
- ContraVault AI, "Best Go/No-Go Analyzers for Construction RFPs in 2026" (2026) — https://www.contravault.com/blog/best-go-no-go-analyzers-for-construction-rfps-in-2026
- ContraVault AI, "Best Go/No-Go Analyzers for Construction RFPs in 2026" (2026) — https://www.contravault.com/blog/best-go-no-go-analyzers-for-construction-rfps-in-2026
- Dan Cumberland Labs, "How Engineering Firms Are Using AI to Write Better Proposals" (2026) — https://dancumberlandlabs.com/blog/ai-engineering-proposals/
- QorusDocs (via Stargazy), "4 Numbers Every AEC Proposal Leader Should Know in 2026" (2026) — https://stargazy.io/resources/4-numbers-every-aec-proposal-leader-should-know-in-2026
- Dan Cumberland Labs, "How Engineering Firms Are Using AI to Write Better Proposals" (2026) — https://dancumberlandlabs.com/blog/ai-engineering-proposals/
- ContraVault AI, "Best Go/No-Go Analyzers for Construction RFPs in 2026" (2026) — https://www.contravault.com/blog/best-go-no-go-analyzers-for-construction-rfps-in-2026
- HSO, "Maximizing Proposal Win Rates for AEC Firms" (2024) — https://www.hso.com/blog/maximizing-proposal-win-rates-for-aec-firms
