Product Discovery · Deep Dive · Plain-English edition
Oil & gas, petrochemicals, infrastructure & giga-projects. Every acronym and technical term is explained in plain English inline. Includes real, sourced quotes from industry players validating the pain and the willingness to pay. — 26 August 2026
Giant construction projects almost always finish late and over budget — and have done so at roughly the same rate for 70 years.
The reason isn't a shortage of software; it's that the information needed to run a project well is scattered across dozens of separate systems, dozens of separate companies, and millions of documents nobody can read fast enough. That's exactly the problem AI agents (software that can read across systems, reason over documents, and take actions with a human's approval) can now attack. This report maps the pain, the money, the competitors, where agents fit, and how to sell into the Gulf oil companies — backed by real quotes from the people who live it.
Every mega-project moves through the same funnel of stages, each ending in a gate (a formal go/no-go decision you must pass to get funding for the next stage). The industry calls this stage-gating or FEL (Front-End Loading — doing the heavy planning up front, before committing the big money).
Two frameworks govern it: AACE estimate classes (AACE = Association for the Advancement of Cost Engineering; its "classes" rate how accurate a cost estimate is, 1–5) and CII/IPA stage-gate models (CII = Construction Industry Institute; IPA = Independent Project Analysis — research bodies that define best practice).
FEED = Front-End Engineering Design (the detailed design stage that produces the blueprint and the budget the board votes on).
EPC = Engineering, Procurement, Construction (one firm takes responsibility for designing, buying and building it).
The budget is decided at the very start (FEED/FID) but the overrun only shows up years later during construction. Most software only helps during construction — by then the damage is already baked in.
Each party keeps its own system with its own coding structure (private numbering — a WBS, the hierarchical to-do list, and a CBS, the matching cost buckets). Owner SAP doesn't talk to contractor cost systems; the schedule is disconnected from what's ordered and designed; correspondence sits as unstructured PDFs. KPMG (a Big Four consultancy): a third of owners haven't connected their project system to accounting. NIST (US standards institute): ~$15.8bn/yr lost just to systems not exchanging data. Reporting is monthly batch — weeks old before anyone sees a problem.
2026 is a consolidation year — incumbents are buying AI startups. Almost all of it is contractor-side and construction-phase. The owner-side, cross-system, front-end and claims layers stay open — that's the gap.
McKinsey: ~79% average overrun (532 projects); 98% of megaprojects overrun >30%. Flyvbjerg (the Oxford professor whose 16,000-project database is definitive): only 0.5% finish on budget, on time and on benefits — the "Iron Law." EY: in 365 oil-and-gas megaprojects, 64% over budget, 73% late, 59% average cost growth, $500bn total — and the Middle East is the worst region on Earth (89% over cost).
The through-line: the information that could save a project already exists — it's just scattered and unreadable fast enough. That is exactly what AI agents are built to do.
Scored from the owner's seat on three axes — value to the owner, delivery confidence today, and how little effort/data the owner must give up. Lead with A (claims-defence) and G (design-feasibility gate): both are high-value, buildable now, and run largely on data the owner already holds (contracts & correspondence for A; design files for G), so they cost the owner least to adopt. B (FID red-team) is the highest-value prize but a harder first sale — it needs the owner's own historical data and lands an unwelcome message at the highest-stakes gate, so it's the ideal second move. C (cost), D (schedule-truth), E (handover) and F (interface) rank lower: deeper system integration or data the owner can't hand over cheaply. Net: lead with A and G, follow with B.
Actual, sourced statements from inside this world — confirming both that the pain is real and that they will spend money to fix it.
On most complex capital programmes, a significant amount of money that one party is entitled to recover is never recovered… What most projects lose to poor claims management is not a dispute they should have won, but entitlement they never assembled the evidence to claim.
— Kairos (capital-projects claims advisory), 2026 · wearekairos.com
The difference between the amount submitted for a claim versus the amount actually paid at settlement… is as much as half. … The owner too must ensure profitability of his project by defending unsupportable claims by a "low ball" contractor.
— IPA (Independent Project Analysis) via National Academies · nationalacademies.org · and Long International · long-intl.com
Company leaders were faced with a startling finding… their world-spanning capital-projects portfolio had yielded only minimal net present value, with more than half of its potential lost to underperformance… all shared a root cause — bias.
— McKinsey, Don't cancel or coddle at-risk capital projects, 2025 · mckinsey.com
For an average size project, even a 10 percent overrun can lead to a $5 million hit to project profitability… By the time you see the project status, it's already outdated, and can be too late to catch a delay or cost overrun.
— Deltek, The True Cost of Project Delay · ceacisp.org (PDF)
[The customer's] needs are simply a facility with no defects, delivered on time and within the given budget… To address the current challenges in program execution — including severe resource constraints and longer procurement lead times — Saudi Aramco has developed and implemented several innovative procurement and contracting strategies.
— Saudi Aramco, via PMI case study on its Quality Management Information System · pmi.org
ADNOC is on a mission to become the world's most AI-enabled energy company, maximizing the potential of AI to drive efficiency and value creation across our operations.
— Musabbeh Al Kaabi, ADNOC Upstream CEO, on a $340M three-year agentic-AI contract (after a proof-of-concept), described as "reducing time for essential business processes from months to days." AIQ/Presight, 10 Mar 2025 · prnewswire.com
With Document Crunch, we can safely and confidently bid more jobs than before. When you understand your risks upfront, you can be more competitive — you're not padding fees just to cover the unknowns.
— Jeff Brannen, SVP & Chief Legal Officer, Balfour Beatty · trimble.com. He also reports risk-review meetings "gone down from four hours… to two hours" · documentcrunch.com
Every contractor has a story about a contract clause that cost them — and at $200/month, Document Crunch pays for itself with a single caught risk.
— Velocity AI Insights review, 2026 · velocityaipartners.co
nPlan's AI is already in use on more than $500 billion worth of live projects… To date, nPlan has saved its customers well over $1.2B. The Transpennine Route Upgrade has gone further, contractualising the use of nPlan.
— nPlan Series B announcement, 17 Oct 2025 · nplan.io
The pain (bias at FID, lost entitlement, late detection) is confirmed by McKinsey, Aramco and specialist claims advisors; the willingness to pay is proven by a $340M ADNOC contract, a Tier-1 contractor's CLO crediting the tool for competitive bidding, and a customer writing AI into its contracts. The two owner-side opportunities we favour (A and B) are the least served by today's vendors.
The pain and value here are unusually well-quantified, and the strongest signal is that a government is building this itself.
A fully integrated digital system capable of issuing building permits automatically and without human intervention by reading submitted drawings and documents, verifying compliance with the Dubai Building Code.
— Dubai Municipality, on its AI permit system (26-week rollout), 2026 · zawya.com
We reduced our SBC compliance review cycle from 6 weeks to 3 days. The AI caught two fire environmental gaps that would have delayed our building permit by months.
— Customer testimonial, WhiteHelmet (Saudi-based construction intelligence), 2026 · whitehelmet.sa
What used to take us weeks is now done in days… Every coordination issue that Buildcheck caught saved us from expensive field modifications and potential delays.
— Customer testimonial, Buildcheck (AI design review) · buildcheck.ai
Rework ≈ 5% of project cost on average (CII IR-153; 90th-pct 12.4%), design rework +1–3 pts. ~70–80% of cost deviations trace to design, not execution. An error caught in design costs ~10× less than in construction, ~100× less than after handover (CII 1-10-100); pre-construction review catches 60%+ of rework-causing errors. Independent ROI: 4:1 to 30:1 cost-avoidance; an RFI is $200–$500 at drawing stage vs $2,000–$8,000 in the field. Documented outcomes: design errors reaching the field cut 50–75%, schedule compressed 8–18%. On a $5bn giga-project, ~5% rework = ~$250M, of which ~28% (~$70M) is design-caused; catching half early avoids ~$35M on a single project — before schedule/first-oil value. On a $20bn project the design-caused figure is ~$280M. (Sources: reworkcost.com/CII IR-153, mycomply.net, helonic.com, miragemetrics.com, tommasomariaricci.com)
Strong practitioner and vendor validation (WhiteHelmet, Buildcheck) and a decisive regulator signal (Dubai Municipality), but no on-record quote from an Aramco/ADNOC owner executive naming this specific tool — treat owner-persona demand as inferred from the regulator move and the mandatory SBC 2024 reset, and confirm in primary interviews.