# Impact Sizing: Shield (Asset Nova wedge)

Date: 2026-08-26 · Run non-interactively; all assumptions stated with confidence levels.
Sources: `context-library/research/Orbitron/cpm-agentic-ai-product-discovery.md` (all stats cited there), `outputs/roadmaps/asset-nova-strategy-and-roadmap.md`.
Note: business-info-template.md and metrics/ are empty; no internal baselines exist. This sizing uses industry data + explicit assumptions.

## What we're sizing

An agent that continuously reads contracts + the live project record (correspondence, schedules, reports) on capital megaprojects, detects incoming contractor claims against the owner, evaluates the contractor's position for weaknesses, warns before defense deadlines expire, and drafts the owner's response. Human approves every outbound item. Two value lenses:

1. **Customer value** (what the pilot must prove)
2. **Orbitron revenue** (what the business case must show)

---

## Lens 1: Customer value per project

### Driver tree ($5B representative GCC megaproject, owner-side deployment)

```
$5B project capex
  ↓ × 33% = $1.65B potentially disputed (HKA CRUX avg on affected projects; conservative haircut to 20% → $1.0B)
  ↓ Commercial leakage addressable by better notice discipline + evidence:
      • Claims the owner paid that could have been defended
      • Contractor claims accepted because the owner didn't have evidence to challenge them
      • Claims that arrived as surprises, leaving no time to prepare a defense
  ↓ Shield intervention rate: agent surfaces 70% of trigger events vs ~40% caught manually today (assumption)
  ↓ Value capture on incremental events: 3–8% of at-risk value protected/recovered
  = $30M–$80M protected per project  (expected case: $50M)
```

### Scenarios per project

| Scenario | Key variable | Value protected | Multiple on $2M/yr license |
|---|---|---|---|
| Worst | Agent finds only what humans already catch; value = faster dispute prep only | $5–10M (avoided legal/consulting spend + settlement uplift on 1 dispute) | 2–5x |
| Expected | 1–2 material missed-notice saves + stronger substantiation across ~30 events/yr | ~$50M | ~25x |
| Best | One Kashagan-class save (single defeated claim) | $100M+ | 50x+ |

**The floor case still clears 10-15X ROI, which is Orbitron's published FlowNova claim. The pilot success metric should be dollar-denominated exposures surfaced, validated by the customer's contracts team.**

### Usage funnel (within one deployed project)

| Stage | Volume | Drop-off | Reason |
|---|---|---|---|
| Documents ingested | 100K–500K docs (Aconex archive + weekly increments) | — | Read-only export; no integration barrier |
| Candidate events detected | ~200–400/yr | — | Correspondence + schedule + daily report triggers |
| Events surfaced to humans (above confidence threshold) | ~100–150/yr | 60% | Precision filter; noise kills trust (see AI metrics in PRD) |
| Events reviewed by contracts team | ~90% of surfaced | 10% | Requires weekly review ritual; adoption risk |
| Notices/actions drafted | ~30–50/yr | — | Only where entitlement path exists |
| Approved & sent by humans | 70–85% of drafts | 15–30% | Draft quality + commercial judgment overrides |

Biggest funnel risk: **the review ritual**. If precision is low in week 1-4, the contracts team stops looking. This drives the AI engineering bar (precision > recall early).

---

## Lens 2: Orbitron revenue

### Funnel (GCC + adjacent, 24 months)

| Stage | Count | Assumption |
|---|---|---|
| Target accounts (GCC owner-operators + Tier-1 EPCs + PMCs) | ~40 | Aramco, ADNOC, QatarEnergy, PIF giga-projects, SABIC + 10 EPCs |
| Qualified conversations (vendor-registered, sponsor identified) | 12 | 30%; Orbitron's existing MENA presence accelerates |
| Paid pilots (3–6 months, 1–2 projects each) | 5 | 40%; pilot priced $300–500K |
| Convert to program license | 3 | 60%; pilot proves $ surfaced |
| Avg ACV at license | $1.5–3M/project-year | 0.05–0.1% of project value on $3–5B projects (half the 0.1–0.3% industry anchor, entry pricing) |

### Revenue scenarios (by end of year 2)

| Scenario | Pilots | Conversions | Projects live | ARR |
|---|---|---|---|---|
| Worst | 3 | 1 | 2 | ~$3M |
| Expected | 5 | 3 | 6 | ~$10–12M |
| Best (one Aramco program-wide deal) | 6 | 4 | 15+ | $25M+ |

Key sensitivity: **one program-wide owner deal changes everything** — Aramco's Jafurah-scale programs run 10+ EPC packages. Land one program, not one project.

---

## Assumption risk table

| Assumption | Confidence | Risk if wrong | De-risking action |
|---|---|---|---|
| 33% of capex disputed (HKA) generalizes to GCC megaprojects | Med | Overstates addressable leakage | Use customer's own claims history in pilot scoping; ME dispute values ($91M avg) suggest conservative |
| Agent detects 70% of trigger events at usable precision | Low | Product fails trust test | Retrospective backtest on a closed project's archive BEFORE live pilot — this is the single most important de-risk and doubles as the sales demo |
| 3–8% of at-risk value protected | Low | ROI story collapses to "faster paperwork" | Define pilot metric as exposures surfaced + validated by customer counsel, not recovery (recovery takes years) |
| Owners will share Aconex archives with a vendor | Med | No substrate, no product | In-region VPC/sovereign deployment from day 1 (SACS-002, PDPL); read-only exports lower the ask |
| Contracts teams adopt weekly review ritual | Med | Shelfware | Design agent output as their existing artifact (event register + notice log), not a new dashboard |
| $300–500K pilots land within project-team authority | Med | 18-month procurement cycles | Confirm delegation-of-authority thresholds per account; EPC contractors as faster wedge if owners stall |

## Recommendation

**Proceed.** Expected case: ~$50M/yr protected per project against a ~$2M license (25x ROI), with a worst case that still clears Orbitron's published 10-15X ROI bar. The two low-confidence assumptions (detection rate, value capture) are both de-riskable with one artifact: **a retrospective backtest on a completed project's document archive** — which is also the case-study demo. Strategic fit: dollar-denominated wedge that builds the substrate for Horizons 2-3 (Design Gate, Budget Red Team) per the roadmap.

**For the deck:** lead with "33% of capex ends up disputed; the average Middle East dispute is $91M and takes a year; most losses trace to notices missed and evidence scattered — problems of reading, not judgment."