Where AI Actually Belongs in EPC Procurement (and Why 9 in 10 Projects Still Overrun)

What EPC Procurement Controls — and Why It Matters

In an EPC contract, the contractor assumes full responsibility for engineering, procurement, and construction— delivering a complete, operational facility at a guaranteed price by a fixed date³. Procurement decisions are widely estimated to govern 40–60% of total project cost⁴. That's not a minor operational function. That's where most of the money gets committed before construction crews arrive.

Procurement decisions lock in three things simultaneously:

• Cost ceiling — supplier commitments fix what materials and equipment will cost before construction begins
• Schedule viability — lead times for specialized equipment, which can stretch from months to years⁴, determine whether the construction schedule is even physically achievable
• Scope constraint — what engineering can actually build is bounded by what has been sourced and at what specification

Under lump sum turnkey (LSTK) contracts, the contractor absorbs all financial risk. This makes procurement decisions doubly consequential: cost overruns come out of contractor margin, not the owner's budget. Which raises an obvious question— if procurement controls this much of the outcome, why isn't AI being aimed at the decisions that actually drive failure?

In EPC, procurement doesn't just affect cost. It sets the ceiling for what engineering can accomplish and the floor for how long construction will take.

What Procurement AI Is Actually Doing in EPC

AI procurement tools deployed in EPC are primarily optimizing three things: bid evaluation speed, purchase order and invoice matching, and unit cost reduction. These are real efficiency gains. They are not, however, the main reasons EPC projects fail.

Current AI applications in EPC procurement:

• Bid evaluation automation — faster screening and scoring of supplier submissions
• PO and invoice matching — reducing manual processing time for purchase orders
• Cost benchmarking — comparing unit prices against historical data and market rates
• Supplier onboarding automation — streamlining vendor qualification workflows

Some platforms include supplier risk monitoring as a secondary capability. At least that nods in the right direction. But it's not the primary optimization— and the distinction matters.

More than 72% of procurement leaders are piloting or deploying AI¹. Only 28% have scaled adoption beyond pilots. That gap— wide adoption intention, narrow transformation signal— is the tell. High pilot rates with low transformation rates usually mean the tool is being aimed at the wrong problem.

Current EPC procurement AI is sophisticated at the transaction layer. It's mostly silent about the sequencing layer— which is where project outcomes are actually determined. Speeding up procurement on an incomplete engineering scope doesn't fix the scope. It just gives you a faster path to the wrong answer.

The Problem — 9 in 10 Large Projects Still Go Over Budget

Nine out of ten large infrastructure projects exceed their original budgets— with average overruns of 28%². That's not an outlier problem. That's a structural one.

In energy-specific EPC projects, more than three out of five projects experience cost overruns, across a 2025 peer-reviewed study examining 662 projects spanning 83 countries and $1.358 trillion in investment.⁵

Independent studies. Consistent pattern. Technology investment is not the gap.

Consider what that implies: EPC contractors have been investing in construction technology for decades — project management systems, BIM, cost engineering platforms. The adoption curve is steep.

Construction industry productivity grew at just 0.4% annually between 2000 and 2022— compared to 3.0% in manufacturing over the same period⁶. The problem isn't access to tools. It's what those tools are being aimed at.

The root causes of EPC cost overruns aren't slow procurement cycles. They're engineering estimation errors, procurement-engineering sequencing failures, and supplier relationship structures that optimize for price rather than project outcomes. AI applied to procurement cycle time doesn't touch any of those.

There's a pattern worth naming here. It's the automation of failure risk. AI makes you faster at whatever you're already doing— which means if your procurement process is missequenced, you just get missequenced faster. The question worth asking isn't how to speed it up. It's what you're actually pointing it at. For a look at how this pattern appears across industries, see the hidden costs of AI implementation.

Three Things EPC Procurement AI Should Be Optimizing Instead

The root causes of EPC cost overruns point to three optimization targets that current procurement AI largely ignores: engineering completeness before procurement begins, total cost of ownership over unit price, and supply chain resilience through strategic supplier relationships.

The goal of procurement AI should be better project outcomes, not faster procurement. Those are different objectives. They require different optimization targets.

Engineering Completeness Before Procurement

The most under-addressed problem in EPC procurement is procurement that begins before engineering is finished. The numbers here are worth sitting with. According to Alex Modon, CEO of Unlimited Industries, speaking to Latitude Media⁷, EPC projects typically reach only 30% engineering definition before financial commitment is made— leaving 70% of scope unresolved when supplier commitments begin.

That's the number worth sitting with. Seventy percent of scope undefined. At financial commitment.

PLC Construction research indicates that targeting 95%+ engineering completion before major procurement can reduce total project costs by 15–30%⁸. Early contractor involvement (ECI) is a formal contracting approach that brings construction teams in before engineering is finalized. Done well, it avoids the field installation rework costs that typically represent 3–8% of construction budgets⁸.

What AI could do here: quantify engineering readiness before procurement commitments are made; flag when scope definition is insufficient to support reliable sourcing; model the downstream cost risk of procuring on incomplete specifications. Some fast-track projects deliberately begin procurement early to compress schedules— that's a legitimate tradeoff. But AI should help quantify that tradeoff's risk, not automate it blindly.

Total Cost of Ownership Over Unit Price

EPC procurement optimized for lowest unit price routinely produces the highest total project cost. Equipment that's cheaper per unit but arrives late, fails quality inspection, or requires non-standard installation generates change orders and idle time that dwarf the upfront savings. Change orders— often stemming from exactly this kind of procurement decision— typically generate 5–10% cost increases⁸.

Total cost of ownership (TCO) in EPC accounts for delivery speed, quality failure rates, logistics complexity, installation requirements, and supplier resilience over the project lifecycle— not just unit price. Strategic sourcing that optimizes for TCO rather than unit cost can achieve 8–15% cost savings on materials and equipment versus project-by-project purchasing⁸.

Applied correctly, AI would model TCO at bid evaluation stage rather than defaulting to unit cost as the primary selection criterion; score suppliers on delivery reliability and quality history alongside price. A procurement process that moves faster on an incomplete scope isn't efficient. It's just a faster path to a change order.

Supply Chain Resilience and Strategic Supplier Relationships

Treating suppliers as interchangeable vendors— selected fresh for each project at the lowest bid— is one of the most expensive practices in EPC. Long-term strategic supplier relationships reduce lead times, improve quality consistency, and create the trust needed to absorb project complexity when (not if) things change.

Transactional supplier relationships miss opportunities to reduce lead times and improve product quality⁹. And with specialized equipment lead times stretching from months to years⁴, supplier reliability is a schedule variable— not just a cost variable.

EPC firms that treat each project as "n of one" lose the compounding benefit of relationship-based supplier management. At the portfolio level, AI could score supplier relationships across project history; identify performance patterns that project-level procurement misses entirely.

Where current AI optimization targets miss the mark:

The gap between each column isn't a technology problem. It's a question of what you're pointing the technology at. If you're mapping our AI strategy services to actual operational goals, this reframing is the conversation worth having first.

The Structural Problem Software Can't Solve

Before you redirect your AI budget, there's a structural factor most procurement conversations avoid.

Some of the mis-optimization in EPC procurement isn't a technology problem at all. Traditional cost-plus contracts create a structural perverse incentive: EPC contractors make more money when projects take longer and cost more.

"We only make money when the project costs more and takes longer." — Alex Modon, CEO, Unlimited Industries⁷

And the implication for software:

"There's no real incentive structure for an EPC to want better software." — Alex Modon, CEO, Unlimited Industries⁷

Both LSTK and cost-plus contracts create perverse incentives, though in different directions. LSTK discourages transparency— because admitting problems increases contractor risk exposure. Cost-plus rewards inefficiency. Neither structure naturally aligns contractor incentives with project owner success.

Incentive misalignment is the problem behind the problem. AI deployed into a perverse incentive structure doesn't fix the incentive— it serves it. Building AI governance for complex operations on top of misaligned incentive structures doesn't change the underlying equation.

Integrated project delivery (IPD) and target-cost contracts are beginning to realign these incentives. They're worth tracking. The critique here is narrower: most AI procurement investment today assumes the current contract structure, not a reformed one.

Where to Redirect Your AI Investment in EPC

Redirecting AI investment in EPC procurement starts with a different question. Not "how can AI speed this up?" but "what upstream problem, if solved earlier, would make this downstream process unnecessary?"

Three specific investments worth considering:

1. Engineering completeness scoring — quantify what percentage of scope is defined before supplier commitments are made; create a trigger threshold below which procurement is flagged as high-risk
2. Supplier relationship scoring across project history — evaluate delivery reliability, quality performance, and resilience, not just unit price at bid time
3. TCO modeling at bid evaluation — build total cost of ownership into the selection decision before commitments are locked in

The strategic framing is straightforward. AI is a force multiplier. It rewards what you're already optimizing for. A solid AI decision framework for your organization begins with auditing what metrics you're currently rewarding— before selecting a tool designed to optimize them faster.

The firms that will outperform in EPC aren't necessarily the ones with the most AI. They're the ones with the clearest answer to "what are we optimizing for?" If navigating that question requires external perspective, that's exactly the work Dan Cumberland Labs does with AEC firms mapping AI investment to actual operational goals.

Frequently Asked Questions

What does EPC stand for in construction?

Engineering, Procurement, and Construction— a project delivery model where a single contractor designs, sources all materials and equipment, builds, and commissions a complete facility under a fixed-price contract. The contractor assumes full responsibility for delivering the facility on time, on budget, and to specification³.

What percentage of EPC projects go over budget?

Approximately 90% of large infrastructure projects exceed their original budgets, with average overruns of 28%². In energy-specific EPC projects, more than three out of five projects experience cost overruns, across a study examining $1.358 trillion in investment across 83 countries⁵.

How is AI currently being used in EPC procurement?

AI in EPC procurement is primarily deployed for bid evaluation speed, purchase order and invoice matching, cost benchmarking, and supplier onboarding automation— optimizing for transaction speed and unit cost reduction¹. Secondary applications include supplier risk monitoring, though this remains a smaller share of total AI deployment in the category.

What is total cost of ownership in EPC?

Total cost of ownership (TCO) accounts for the full cost of a procurement decision across the project lifecycle— including delivery reliability, quality failure rates, logistics complexity, installation requirements, and supplier resilience— rather than just unit price. TCO is widely considered procurement best practice but remains underweighted relative to unit cost in most current EPC procurement decisions⁸⁹.

What You're Actually Optimizing For

The case for AI in EPC procurement is real— but it depends entirely on what you're using it to do. AI applied to transaction speed and unit cost reduction produces incremental gains on the metrics that don't drive project outcomes. Applied upstream— to engineering completeness before procurement begins, supplier relationship quality across the project portfolio, and total cost of ownership at bid evaluation— it can move the needle on the failures that actually matter.

Most EPC firms that are struggling with AI adoption aren't struggling with the technology. They're struggling with the question that comes before the technology: what are we actually trying to improve?

The firms that will outperform in EPC aren't the ones with the most AI. They're the ones with the clearest answer to that question. If you're working through how to assess whether your AI investment is actually delivering results, that's the right place to start.

References

1. Global CPO GenAI Survey 2024, via Ivalua, "The Role of AI in Sourcing and Procurement" (2025) — https://www.ivalua.com/blog/ai-in-sourcing-and-procurement/
2. MDPI, "An Analysis of Factors Contributing to Cost Overruns in the Global Construction Industry" (2025) — https://www.mdpi.com/2075-5309/15/1/18
3. PMI, "Realizing Engineering, Procurement and Construction Projects" (2020) — https://www.pmi.org/learning/library/realizing-engineering-procurement-construction-projects-7173
4. GEP, "EPC Procurement & Sourcing: Process, Risks, Challenges and Strategies" (2024) — https://www.gep.com/blog/strategy/epc-procurement-risk-challenges-strategies
5. ScienceDirect/Elsevier, "Beyond Economies of Scale: Learning from Construction Cost Overrun Risks and Time Delays in Global Energy Infrastructure Projects" (2025) — https://www.sciencedirect.com/science/article/abs/pii/S2214629625001380
6. McKinsey & Company, "How the construction industry can boost productivity through technology" (2023) — https://www.mckinsey.com/uk/our-insights/the-mckinsey-uk-blog/how-the-construction-industry-can-boost-productivity-through-technology
7. Latitude Media, "Can AI Revolutionize EPC?" (2024) — https://www.latitudemedia.com/news/catalyst-can-ai-revolutionize-epc/
8. PLC Construction, "EPC Procurement Strategies to Reduce Project Costs in 2026" (2025) — https://www.plcconstruction.com/epc-procurement-strategies-to-reduce-project-costs-in-2026/
9. PurchaserAI, "Procurement in EPC: Where Execution Gets Stuck" (2024) — https://purchaser.ai/blog/procurement-in-epc-where-execution-gets-stuck