Internal · scoping map · not client-facing

Facade.ai — Estimating Workflow & Automation Map

End-to-end teardown of how Imperium Building Solutions produces a facade bid, mapped against where Facade.ai takes over. Reverse-engineered from the 80 West Broadway package (RFP → Exhibit B SOW → leveling form → drawings → RFI samples).

v2 · updated 2026-06-11 · incorporates Arianna's written answers + Exhibit B SOW + RFI_01 sample + Common RFIs reference · v1 source: 2026-06-01 call + 2026-06-02 email

Unknowns resolved since v1

  • PDF-only during bidding — CAD comes post-award only
  • Input IS industry-standard — current PDF drawing set + GC's SOW template
  • Output format IS templatable — Exhibit B categories are always the same
  • Pricing source confirmed — supplier quotes, no internal price book
  • Stable supplier roster — 2–3 accredited suppliers, same every project
  • RFI categories are recurring — 10 standard buckets, automatable as a checklist
  • Takeoff IS mostly codifiable — assembly rules are learnable; judgment is at the margin
  • Design phase matters — SD/DD/CD detection changes scope confidence + flags future COs

The market gap, confirmed: facade firms use Bluebeam for takeoff measurement — but you still manually identify which areas to measure, and nothing facade-specific carries those quantities into estimating (price generation + a knowledge database). The next step Arianna wants: a system that generates an offer with all items to quote, each referencing the drawing/detail/spec it came from, so the supplier can simply put a price next to each line. Facade.ai owns the identification → organization → packaging chain that no tool currently touches.

6–7
robust proposals/year — throughput capped by estimating effort, not demand
~1 week
saved per project just by auto-identifying all items in the GC's scope of work (Arianna's own estimate)
3 suppliers
stable roster per project (CANTORI + glass + specialty panels) — predictable packaging target
High automation Medium Low Stays human MVP v1 scope

The pipeline — what happens vs. what Facade.ai does

◆ MVP Wedge v1 = ingest → extract → take off → reconcile. Output: a verified quantity takeoff aligned to the GC's leveling schema (Exhibit B), packaged per supplier. STAGES 0–3
0
~95%
Intake & triageHighMVP

Manual today

GC emails an RFP + a Dropbox link. Simone opens an all-disciplines design package and pulls out only the facade-relevant sheets & specs by hand.

80 W Bway: 448 PDFs · 84 DWGs · 328 sheets · ~100 spec sections → ~25 relevant

→ Facade.ai

Ingest the entire package, classify every sheet/spec, surface the facade subset. Also: detect the design phase (SD / DD / CD) — this determines scope confidence and flags that a subsequent drawing set will follow.

✓ Always PDF during bidding. CAD comes post-award only. Design phase detection = new value layer.
est. ~95% automatable
1
~92%
Read finish schedule + scopeHighMVP

Manual today

Read the GC's Exterior Finish Schedule (every material code → source, color, location) and the Exhibit B SOW template to establish what's to be furnished. Cross-check against specs sections.

80 W Bway specs referenced: 084400 Curtain Wall · 074213 Metal Panels · 088010 Glass · 089000 Louvers · 087100 Door Hardware · 079200 Sealants

→ Facade.ai

Extract the full TAG# → material → location table and parse Exhibit B line items into a structured scope list. Output categories are always the same: Windows, Curtain Wall, Terracotta, Metal Panels, Storefront, Terrace Doors. Anything not there must be explicitly EXCLUDED.

✓ Exhibit B is the "bible" — confirmed as industry-standard output template. Categories don't vary GC-to-GC at the top level.
est. ~92% automatable — output schema confirmed, more confident than v1
2
~65%
Markup & takeoff — the bottleneckMed–HighMVP

Manual today

Simone prints elevations, hand-colors every surface by material, counts units, measures m²/lm, and applies assembly rules. Bluebeam measures once you highlight — but you still pick what to highlight.

e.g. 1 window = 6 vert + 1 horiz terracotta = 7 pcs → 189 windows = 1,323 profiles + 378 corners. 273 window/door units total. Each TC panel has a substructure connecting to the building.

→ Facade.ai

The codification is clearer now. Two tracks:

Metal surrounds (MT-54): (1) identify perimeter in lm from elevation drawings; (2) extract geometry from detail drawings to determine bent / L-shaped / C-shaped fabrication type; (3) quantity = perimeter × assembly rule. Highly automatable.

Terracotta system: identify TC panels from finish schedule; count panels and map to elevation; identify substructure system (each panel has a bracket substructure — engineer determines anchor count based on weight + dimensions + wind loads). Substructure logic is somewhat standard.

✓ PDF-only confirmed. No CAD/BIM during bidding — resolves swing variable from v1. Estimate stays ~65% — judgment is at the margin for unusual geometry.
est. ~60–70% — PDF-confirmed. Swing is now about geometry complexity, not format uncertainty.
3
~99%
Quantity reconciliationHighest-confidenceMVP

Manual today

The same quantities live in 3 formats — architect (imperial) → Imperium SOW (metric) → GC bid form (imperial/bid categories). Simone converts & cross-checks by hand, every revision.

Reconciles cleanly: surround 1,065 lm ↔ 3,491 LF (=1,064 m ✓) · planters n°42 ↔ 42 EA ✓

→ Facade.ai

Deterministic conversion + Exhibit B category mapping + self-validation. Because the numbers must agree across documents, the system can verify itself with no human check.

est. ~99% — deterministic & self-verifiable
4
~80%
Supplier RFQ packaging & sendHigh — revised upv2+

Manual today

Email the correct material subset to each supplier, then wait 3–4 weeks while each redoes the takeoff on their portion. If the info is pre-organized with RFIs already sent, suppliers say it becomes very easy — just a matter of putting a price next to it.

→ Facade.ai

Supplier roster is stable and known — auto-package the right material set per supplier from Stage 2–3 output. The packaging format is now confirmed as the value unlock: suppliers get organized, reference-linked line items instead of raw drawings.

✓ Stable 3-supplier model confirmed. Estimate revised up from 70% → 80%.
CANTORI (cantorigroup.it)
Primary window manufacturer. Also covers metal panels & aluminum structures. Sends prices for all window components + fabrication (including final packaging). Italy-based.
Glass supplier
Custom IGU (insulated glass unit) to project specs. Separate from CANTORI. Supplies the glazing unit per spec section 088010.
Specialty panels supplier
Terracotta, porcelain, or concrete panels depending on project. Separate specialist from window/metal supplier.
est. ~80% automatable — stable roster removes the main gating unknown from v1
5
~55%
Missing info detection & RFI generationMedium — revised upv1.5 candidate

Manual today

Arianna or a supplier spots a missing callout or drawing conflict → manually creates an RFI and references the relevant drawing. Industry is old-fashioned: supplier emails/calls with a question → manually transformed into formal RFI. Tool auto-generation is not used.

Real example (RFI #1, 80 W Bway): "Confirm door type on West Elevation — missing callout." + "Confirm corrugated vs flat panel on parapet — elevation and detail conflict."

→ Facade.ai

Two automation layers:

(1) Pre-bid checklist: scan every project against the 10 standard RFI categories (performance requirements, glass make-up, system type, structural attachment, trade interfaces, dimensions, finishes, mock-up, procurement, scope gaps). Flag any that can't be answered from the documents. This alone saves multiple back-and-forths.

(2) Drawing conflict detection: cross-reference elevation drawings against section details — catch mismatches like corrugated vs flat panel before suppliers ask.

Performance requirementsWind loads, U-factor, SHGC, air/water infiltration, seismic — extractable from specs
Glass make-upTempered/laminated/Low-E/IGU thickness — in spec sections 088010, drawings
System typeCW vs window wall, stick vs unitized, operable vs fixed — from finish schedule + drawings
Dimension conflictsCross-reference elevations vs plans vs details (like corrugated/flat panel). Needs CV on drawings.
Finish requirementsPVDF vs anodized, AAMA spec, custom colors — usually in finish schedule
Scope gapsInterior trim, sunshades, louvers, fire safing — cross-check Exhibit B against drawings
Structural attachmentSubstrate, embeds, edge distances — needs structural drawings, harder to cross-ref
Mock-up & testingWho pays, required scope — usually a judgment call + negotiation
est. ~55% — checklist + conflict detection automatable; coordination meetings stay human. Revised up from 40%.
6
~10%
Pricing & markupStays humanout of scope

Manual today

Supplier prices arrive; Imperium layers in overhead, PM, customs & duties, contingency, engineering stamps, Italy travel → the final numbers.

→ Facade.ai

Deliberately hands-off — this is Imperium's judgment/margin logic. One future exception: a $/SF rough budget lookup for early developer-stage asks (where GC doesn't yet have a CD set and just needs a ballpark for a bank loan). Arianna notes this depends heavily on supplier relationships and geography — complex to generalize. Could be a per-company historical database feature (upload past proposals/contracts → track original cost vs final + change orders).

est. ~10% — deliberately kept human (the secret sauce)
7
~92%
Bid-form population & outputHighv2 (easy pull-forward)

Manual today

Type every quantity into the GC's Exhibit B template and highlight the relevant scope items in the PDF drawings so the GC can see visually what's being quoted.

Output always includes: (1) highlighted scope PDF — overall elevations + key details; (2) populated Exhibit B with quantities + INCLUDED/EXCLUDED per line item

→ Facade.ai

Populate Exhibit B from Stage 3 output. Auto-generate the highlighted scope PDF (already identified regions in Stage 2 = pre-done markup). Output schema confirmed — high value, low difficulty.

✓ Two outputs confirmed: Exhibit B populated template + highlighted drawing PDF. Both templatable.
est. ~92% automatable — output confirmed, revised up from 90%
8
~75%
Revision diffing & change orders — ongoing 6–12 moMedium — revenue-criticalv2+

Manual today

A new drawing set lands every few weeks for 6–12+ months. DD → 100% DD → CD set → bulletins. Simone manually hunts the deltas and re-prices. Critical business point: changes after the CD set (the contractual basis of design) are additional costs — the ability to flag and document these changes is how Imperium protects revenue on change orders.

→ Facade.ai

Diff drawing versions, flag exactly which quantities changed, re-run Stages 2–3 only on deltas. Added value: automatically flag any change that occurs after the CD set baseline — these are potential change orders. This turns a manual revenue-protection process into an automatic one.

est. ~75% automatable — also doubles as change-order revenue protection. Revised up from 70%.

Target output — Exhibit B (the GC's SOW template, the MVP's north star)

Scope categoryUnitQty (80 W Bway)Furnish + Install
Punched windows (tilt & turn, incl. glass, receptors, straps)SF8,761$1,267,703
Painted aluminum window surround (MT-54)LF3,491$314,627
Storefront systemSF912$237,967
Terracotta panels @ storefrontSF$285,200
Cantilever + mech-screen metal panelsSF3,269 / 3,660 / 655 / 566$1,360,762
Engineering, calcs, submittals, design-assistLS$291,162
Subtotal — Furnish & Install$3,984,619

Exhibit B is the contract's "Exhibit B" — the full scope of work that becomes the binding document. It runs ~230 line items covering office & documentation, safety, installation requirements, glass specs, hardware, louvers, metal panels, testing requirements, and punch list. The top-level categories (windows, curtain wall, terracotta, metal panels, storefront, terrace doors) are consistent across GCs. A structured Alternates block sits below (2-side vs 4-side receptors +$50K/+$128K, terracotta-vs-metal surround −$105K, NY PE stamp +$29K, mockup allowances, bird glass excluded). Note: get the live formula-level file before building against this schema — current values are from cached read.

What stays human (by design)

  • Pricing judgment & markup — overhead, contingency, margin logic. Depends on supplier relationship, geography, project risk.
  • Customs/duties & logistics calls — case-by-case, country-dependent (Italy sourcing adds complexity)
  • Supplier selection & negotiation — relationships, not rules. CANTORI is primary but negotiated project-by-project.
  • Engineering / PE stamping — delegated-design liability sits with a licensed engineer. Anchor count for TC substructure requires engineer sign-off.
  • Final design-optimization decisions — value engineering, flagging coordination issues to GC
  • Supplier technical meetings — engineer-to-engineer coordination on unusual conditions

Open questions — next things to answer

  • TC substructure standard patterns — how often is anchor count truly novel vs. derived from standard panel-weight tables? Could the structural logic be partially codified? follow-up
  • Drawing conflict frequency — how often does an elevation/detail mismatch (like corrugated vs flat panel) occur per project? Calibrates RFI detection value. follow-up
  • Historical $/SF corpus — do they have past proposals/contracts we can look at for the rough-budget feature? Format/size? follow-up
  • Supplier packaging format — what does CANTORI actually want to receive? PDF with line items + drawing refs, or a structured spreadsheet? follow-up
  • Rough budget workflow — how often do SD-phase requests come in? Is a $/SF lookup a real recurring need or occasional? follow-up

Design phase map — what arrives when, and what it means for us

PhaseDrawing maturityWhen RFP arrivesFacade.ai implication
SD Schematic Design Conceptual, few details, low info density Early — developer needs bank loan budget Rough budget only ($/SF lookup). Low automation value — too little data.
DD Design Development ~80% developed — another 100% DD + CD will follow Most common RFP phase Primary target. Full pipeline runs. Must flag "CD set to follow" so revision diffing activates later.
CD Construction Documents Fully developed — contractual basis of design Final bid before award Highest automation confidence. Changes after this = change orders → auto-flag in Stage 8.

Unknowns closed since v1

  • PDF vs. CAD/BIM — resolved: always PDF during bidding. CAD only post-award, post-scope. Stage 2 estimate no longer swings on this. confirmed
  • Input standardization — resolved: industry-standard format confirmed (PDF drawings + Exhibit B SOW). Top-level categories don't vary. confirmed
  • Takeoff codifiability — largely resolved: assembly rules are learnable (metal surround perimeter logic, TC panel count logic). Judgment lives at the geometry margin. confirmed
  • Pricing source — resolved: supplier quotes only. No internal price book. $/SF lookup is a separate, future, and harder feature. confirmed
  • Supplier roster stability — resolved: 2–3 stable accredited suppliers per project (CANTORI + glass + specialty). Enables confident Stage 4 packaging. confirmed
  • Why existing tools fail — resolved: Bluebeam measures once highlighted — but identifying what to highlight is the gap. No estimating software for facades exists. Confirmed exactly as hypothesized. confirmed