A hyperscale electrical RFP arrives Tuesday at 4:18 p.m.: 400 pages of MEP coordination spec, 90 single-line drawings, an 80-page Division 26 narrative, PLA language buried at page 312, bid due in 14 days. The platform's estimating bench is two senior estimators and a junior. The competition includes three regional subs running the same spec, two with bid-prep AI capability. The platform's national-prime relationship wants the bid back by day 10 to roll into the prime submission. The platform either has the workflow — read the spec, surface the buy-list, generate the labor-by-trade build, draft the union/non-union decision, answer the seven-question gate — or it does not bid. This lesson is the operational workflow. It turns a 400-page MEP spec into a 12-page trade execution brief, runs the union-vs-non-union decision matrix for $200/hr+ work, sets the prime-vs-sub positioning math for the specific bid, and runs the seven-question gate the platform CEO answers before committing the first hyperscale bid. The strategic positioning lesson (Ch2 L3) selected the path; this lesson executes the bid.

The Bid-Prep AI Workflow Architecture

The workflow has four stages and one gating decision. Stage one ingests the RFP package — MEP coordination spec, Division 26 electrical narrative, Division 23 HVAC, Division 21 fire-suppression, Division 25 controls, single-line drawings, equipment schedules, addenda, and reference standards (Uptime Institute Tier classifications, ASHRAE 90.4, NFPA 70 NEC, NFPA 75/76, manufacturer-specific commissioning). Every spec section parses against the platform's reference taxonomy so outputs are comparable across bids. Stage two surfaces the buy-list — structured equipment-and-materials list with manufacturer call-outs, quantities, lead times, and price-volatility flags (R-454B refrigerant, copper line-set surcharges, transformer 60-90 week lead, generator 30-50 week lead, switchgear 40-70 week lead). Stage three generates the labor-by-trade build — electrical (rough-in, conduit, cable tray, equipment install, terminations, testing), HVAC (precision air handling, chilled water, CRAC/CRAH, liquid cooling), mechanical (piping, refrigeration), fire-suppression (Inergen, FM-200, water mist, sprinkler), controls (BMS, mission-critical SCADA, EPMS). Stage four drafts the bid narrative, schedule of values, and risk register against execution history.

The gating decision sits at the end of stage four — the seven-question gate. Before the bid goes out, the platform CEO answers seven questions on paper, in front of the bid-review team, against the AI-generated outputs. The seven tactical questions are the complement to the five strategic-positioning questions from the previous lesson. Strategic questions decided whether to pursue hyperscale and on what path; tactical questions decide whether to bid this specific opportunity given the platform's current state and the bid's characteristics. Skipping the gate is the failure mode that turns a winnable bid into a margin-destroying overcommitment.

The AI capability is a stack, not a single tool. Platforms use GPT-class large language models for spec ingestion and narrative generation, custom-tuned models for trade-specific takeoff, Procore Hyperscale or Trimble Estimation MEP for project-management plumbing, and the platform's internal reference taxonomy and execution-history database for risk-register inputs. Mature platforms run the stack as a versioned workflow with documented audit trail; immature platforms run point solutions producing inconsistent outputs that require manual reconciliation, defeating the AI speed advantage.

The Hyperscale Spec-Translation Prompt — 400 Pages to 12-Page Brief

The single most leveraged prompt in the workflow translates the MEP coordination spec into a 12-page trade execution brief field operations can read and use. The prompt structure is consistent across bids; inputs vary by RFP. (1) Project executive summary (customer, site, capacity, schedule, prime/sub structure, key constraints). (2) Scope of work by trade with explicit inclusions and exclusions per the spec. (3) Equipment schedule with manufacturer call-outs, quantities, lead times, price-volatility flags. (4) Labor-by-trade build with journeyman-hours estimate, supervisor-hours, and bench availability against the demand curve. (5) Schedule by phase (rough-in, equipment install, commissioning) with critical-path milestones. (6) Coordination interfaces (electrical-to-HVAC, HVAC-to-controls, fire-suppression-to-all) with named handoff points. (7) Commissioning requirements (factory acceptance, site acceptance, level 1-5 per Uptime Institute discipline). (8) Risk register (schedule, materials, labor, customer-coordination). (9) Assumptions and clarifications. (10) Pricing narrative (cost structure, margin band, sensitivity). (11) Union/non-union recommendation. (12) Prime/sub recommendation.

Senior estimator reviews the AI-generated brief, marks discrepancies, and either accepts, edits, or rejects-and-reruns. Mature platforms achieve 70-85% acceptance after 6-12 months of workflow tuning; the remaining 15-30% is high-judgment content (assumptions, clarifications, prime/sub positioning) that benefits from senior override. Compression is real — 60-80 hours of estimator time compresses to 8-14 hours of review. The platform capturing this compression bids 3-5x more opportunities per quarter at the same bench.

The brief is the artifact field operations reads. Project executive, PMs, superintendents, and senior journeymen read the 12-page brief to understand execution structure before commitment. The brief becomes the basis for the bid-review meeting, prime negotiation if sub, and customer-direct discussion if prime. The brief's quality is the bid's quality — undisciplined briefs produce bids that win on price, lose on execution, and damage relationship and bench in the same project.

The Union vs. Non-Union Shop Decision for $200/hr+ Work

Hyperscale wages run $145-$220/hour fully loaded in 2026. IBEW prevailing-wage rates run $165-$220/hour by geography. Open-shop rates run $125-$180/hour. Prevailing-wage requirements on PLA-governed hyperscale projects impose IBEW-equivalent wage and benefit structures on open-shop platforms — eliminating the wage differential and adding compliance overhead. The decision is rarely "union vs. open-shop forever"; it is "for this specific bid in this specific geography under this specific PLA structure, what is the operationally and financially defensible labor structure."

Six inputs, one output. (1) Geography labor structure — Northern Virginia largely open-shop with selective IBEW; Chicago, NorCal, PNW, Northeast largely union; Phoenix, Columbus, Austin mixed; Atlanta, Dallas, Salt Lake City, Reno, Iowa vary. (2) PLA structure on the specific project — most Microsoft, Meta, Google projects operate under PLAs mandating prevailing-wage; AWS varies by region; CoreWeave, QTS, Equinix often allow open-shop in non-PLA geographies. (3) Platform operating structure — union, open-shop, or dual-shop legal entity. (4) IBEW local hiring-hall capacity in the geography — can the local fill the platform's manpower request? (5) Platform prevailing-wage compliance capacity — can payroll, benefits, and certified payroll reporting handle the burden? (6) Project duration and labor intensity — short-duration high-labor favors traveling crews; long-duration favors IBEW hiring-hall sourcing.

Four output decisions. (A) Union shop under IBEW local — when platform is already union-structured or geography is union-dominated with sufficient hiring-hall capacity. (B) Open-shop with prevailing-wage compliance — when geography is mixed or open-shop-dominant, PLA permits open-shop, and platform has compliance capacity. (C) Dual-shop legal entity with project-specific structure — when platform's residential is open-shop but hyperscale project requires PLA compliance. (D) Sub under a national prime that handles labor structure — when platform's labor-structure capacity is undersized and the prime provides labor and compliance infrastructure. Decision D is the default for path-one strategic-sub; A/B/C apply when bidding prime.

The wrong decision compresses margin or kills compliance. A platform bidding open-shop on a PLA-governed Microsoft Azure $25M scope and discovering prevailing-wage compliance mid-project absorbs $1.5M-$3M in retroactive wage adjustments, back-pay claims, and remediation. A platform bidding union on a non-PLA CoreWeave project where IBEW hall cannot fill manpower adds $400K-$1.2M in traveling-crew per-diem and schedule slippage. The matrix run with discipline at bid prep prevents both failure modes. Many 2026 hyperscale platforms execute under dual-shop legal entities specifically because entity-level discipline lets them apply the matrix per project without disrupting residential.

Prime vs. Sub Positioning for the Specific Bid

The strategic-positioning lesson decided the platform's general posture — path one strategic sub, path two dedicated division, or path three specialist. The bid-specific positioning decides, for this RFP, whether the platform leads as prime or subs under a named national prime. The decision is bid-specific because path-two and path-three platforms may still sub on bids where the prime relationship is strategic or the risk profile favors sub.

Five variables. (1) Bid size relative to surety capacity — $25M bid against $40M surety is bondable as prime; $60M against $40M is not. (2) Schedule pressure relative to bench — 14-month schedule at 280,000-380,000 journeyman-hours against 200,000-quarter bench is executable as prime; 9-month at the same hour count is not. (3) Customer relationship — direct relationship with the hyperscaler (typical for path-two/three with 4+ projects) makes prime viable; without, sub is the access path. (4) Trade-coordination complexity — multi-trade scope captures coordination margin as prime; pure-electrical platforms sub-coordinate with other-trade subs under prime. (5) Risk-adjusted margin — prime at 22% gross with $4M-$8M working capital and $25M bonding vs. sub at 15% gross with $1M-$3M working capital and no bonding.

Output: lead the bid, sub under named prime, or no-bid. Named-prime decision has three sub-decisions: which prime (Rosendin, Cupertino Electric, MMR Group, Faith Technologies, Mortenson Electrical, Hooper Corporation, Mona Electric — selection depends on relationship history, prime's bid history with this customer, prime's labor-structure capacity for the geography); what scope (full electrical, partial like switchgear-and-distribution only, or trade-specific); margin floor before walking (typically 11-13% floor with 15-17% target). Mature platforms maintain 2-3 active prime relationships and rotate the lead by project; immature platforms depend on one prime and lose flexibility.

The Seven-Question Gate Before the Platform Commits

Before submission — prime or sub — the platform CEO answers seven questions on paper, in front of the bid-review team. Each has a binary answer; a single no is sufficient grounds to no-bid or restructure.

Q1 — does the AI-generated 12-page brief reconcile against the source RFP at 85%+ acceptance after senior estimator review. Below 85%, the workflow has not captured the bid's discipline; a 60-70% reconciled brief wins on price and loses on execution. Remediation is rerunning stages one and two with corrected reference taxonomy; no-bid is the discipline if remediation is not feasible inside the bid window.

Q2 — does the labor-by-trade build show 110%+ bench coverage against the demand curve. Bench undersizing is the most common failure pattern at platforms graduating from path one. If the demand requires 280,000 journeyman-hours in Q3 and bench has 240,000, the deficit must be sourced (per-diem traveling crews, hire ramp, IBEW hall) with sourcing cost in the bid. A bid submitted against 90% coverage and a 50% mobilization plan absorbs $1.5M-$4M in unplanned labor cost mid-project.

Q3 — does the equipment schedule reconcile against current 2026 lead times and price-volatility flags. Transformer 60-90 week lead, generator 30-50 week, switchgear 40-70 week, R-454B refrigerant volatility, copper surcharges. If a transformer's expected delivery is week 38 against an 18-week-from-contract requirement, the bid includes alternate equipment (customer/prime validated), schedule extension cost, or no-bid.

Q4 — does the union/non-union decision matrix output match platform structure and project PLA requirements. Mismatch (open-shop platform on PLA-governed project without dual-shop) is the compliance failure that absorbs $1.5M-$3M in retroactive adjustments.

Q5 — does prime/sub positioning match surety, bench, customer relationship, and risk-adjusted margin. A platform with $25M surety bidding a $40M scope as prime either misrepresents surety or finds itself unbondable mid-bid. Positioning must be defensible against each of the five variables.

Q6 — does the risk register cover the four categories at platform tolerance. Schedule (slippage probability and exposure dollars), materials (price volatility and lead-time exposure), labor (mobilization and prevailing-wage compliance), customer-coordination (specification ambiguity and change-order). Each gets probability, exposure, mitigation. Above-tolerance categories trigger remediation or no-bid.

Q7 — does pricing reflect defensible margin band against documented execution history. A 22% gross bid where documented execution averages 17% loses money on execution. Pricing must reflect actual capability, not aspirational margin. Path-two/three platforms with 6-12 documented hyperscale projects have richer execution data; path-one with 1-3 sub projects relies on prime's data calibrated against documented variance.

The gate run with discipline produces binary go/no-go. Platforms holding discipline bid-to-win at higher rates because each national prime's commissioning team and project executives recognize disciplined bid posture and route more bid opportunities accordingly.

How the Workflow Runs at a Platform Versus at a Single Shop

At a single-location electrical shop, the workflow runs the same stages at smaller scale with less automation. The senior estimator runs stages one and two manually with AI-assisted tools; stage three uses local bench data; stage four drafts faster but with less reference taxonomy. The seven-question gate runs identically. Single-shop platforms can compete on path-one strategic-sub at $2M-$15M scope if workflow discipline is in place; bid-prep capability is the lever that lets a single shop bid against regional subs without 5-10 senior estimators on bench.

At a multi-location platform with mission-critical division, the workflow runs as versioned production capability. The Director of AI Operations (per L5 Ch4) governs stage-one taxonomy, stage-two equipment database, stage-three labor benchmarks, stage-four narrative templates. The bench operates with documented quality reviews, the bid-review team runs the seven-question gate, bids roll into pipeline tracking. The quarterly board review includes bid-to-win ratio, average bid response time, and prime-relationship roster. Mature platforms run 30-60 active bids per quarter; immature 5-15 with longer cycle times.

At a path-three focused specialist, the workflow runs at highest discipline. Estimating bench is large (20-50 senior estimators across regional offices); AI capability is custom-tuned with specialized models for transformer scheduling, switchgear coordination, mission-critical SCADA integration, and Uptime Institute commissioning. The seven-question gate runs with documented audit trail; bid-to-win benchmarks against the named primes. Path-three platforms compete directly with Rosendin, Cupertino, MMR, Faith Technologies, and Mortenson; workflow discipline is what enables that competition at scale.

The Workflow Discipline That Separates Bidders From Builders

The bid-prep workflow converts strategic positioning into operational reality. Platforms selecting path one without workflow investment capture 5-15% of available sub revenue. Platforms selecting path two operating workflow informally absorb the $3M-$8M write-off pattern from the previous lesson. Platforms selecting path three without production-scale workflow cannot compete with Rosendin, Cupertino, MMR, Faith Technologies, and Mortenson on prime positioning. The workflow is the operational competency that gates the strategic ambition.

The discipline compounds. A platform running 30-60 active bids per quarter with documented bid-to-win ratio, prime-relationship roster, and execution-history database builds structural advantage across 24-36 months. AI capability tunes to actual bench; prime relationships strengthen because bid quality is documented; customer-direct relationships emerge because execution history is defensible. By year three of the 36-month build-up, the workflow produces bid-to-win ratios of 22-35% on path-one sub work and 12-20% on path-two prime work — ratios not achievable without workflow discipline.

The workflow also produces the operator bench. Senior estimators trained on AI-assisted workflow become next-generation project executives and division leaders. The Director of AI Operations role expands across workflows (bid prep, dispatch optimization, ride-along coaching, lead nurture) as the broader AI capability matures. The bid-prep workflow distinguishes a hyperscale-positioned platform from a residential-platform-with-electrical-capacity that talks about hyperscale at the board meeting but does not have the discipline to execute. The strategic positioning lesson selected the path; this lesson is the workflow that delivers it.

Key Takeaways

• The four-stage bid-prep workflow: (1) ingest the RFP — MEP coordination spec, Division 26 electrical, 23 HVAC, 21 fire-suppression, 25 controls, single-line drawings, equipment schedules, addenda; (2) surface the buy-list with manufacturer call-outs, quantities, lead times, price-volatility flags; (3) generate the labor-by-trade build (electrical, HVAC, mechanical, fire-suppression, controls); (4) draft the bid narrative, schedule of values, and risk register.
• The hyperscale spec-translation prompt turns a 400-page MEP coordination spec into a 12-page trade execution brief in 12 structured sections. Senior estimator review achieves 70-85% AI-output acceptance after 6-12 months of workflow tuning. Compression: 60-80 hours of estimator time compresses to 8-14 hours of review.
• The union vs. non-union decision matrix has six inputs (geography labor structure, project PLA, platform operating structure, IBEW hiring-hall capacity, platform compliance capacity, project duration and labor intensity) and four output decisions (union shop under IBEW, open-shop with prevailing-wage compliance, dual-shop legal entity, sub under national prime that handles labor structure).
• The wrong labor-structure decision costs $1.5M-$3M in retroactive compliance adjustments (open-shop on a PLA-governed project) or $400K-$1.2M in traveling-crew per-diem and schedule slippage (union bid where IBEW hall cannot fill manpower request). Dual-shop legal entities are the common solution for platforms with mixed operations.
• The prime/sub positioning math for the specific bid considers five variables: bid size vs. surety capacity, schedule pressure vs. bench availability, customer relationship structure, trade-coordination complexity, risk-adjusted margin. Output is lead-the-bid, sub-under-named-prime, or no-bid. Mature platforms maintain 2-3 active prime relationships (Rosendin, Cupertino, MMR, Faith Technologies, Mortenson) and rotate the lead prime by project.
• The seven-question gate before submission: (1) 12-page brief reconciles at 85%+ acceptance; (2) labor-by-trade build shows 110%+ bench coverage against demand curve; (3) equipment schedule reconciles against 2026 lead times (transformers 60-90 weeks, generators 30-50 weeks, switchgear 40-70 weeks) and price-volatility; (4) union/non-union decision matches platform structure and project PLA; (5) prime/sub positioning matches surety, bench, customer relationship, risk-adjusted margin; (6) risk register covers schedule/materials/labor/customer-coordination at platform-defined tolerance; (7) pricing reflects defensible margin band against documented execution history.
• Workflow scale by platform tier: single shop runs 5-15 bids per quarter on path-one sub work at $2M-$15M scope; multi-location platform with mission-critical division runs 30-60 active bids per quarter with versioned production workflow; path-three focused specialist runs at the highest discipline level with 20-50 senior estimators and custom-tuned AI capability competing directly with Rosendin, Cupertino, MMR, Faith Technologies, Mortenson.
• Bid-to-win ratio benchmarks: path-one sub work at 22-35% bid-to-win after 24-36 months of workflow discipline; path-two prime work at 12-20% bid-to-win. Ratios not achievable without the workflow discipline at production scale.
• The workflow is the operational competency that gates the strategic ambition. Path-one platforms without workflow capture 5-15% of available strategic-sub revenue instead of the 15-30% the path is designed to deliver. Path-two platforms operating workflow informally absorb the $3M-$8M write-off pattern. Path-three platforms without workflow at production scale cannot compete on prime positioning.
• The workflow produces the operator bench. Senior estimators trained on AI-assisted workflow become the next generation's project executives and division leaders. The Director of AI Operations role expands across multiple workflows (bid prep, dispatch optimization, ride-along coaching, lead nurture) as platform's broader AI capability matures.