1 of 28

AACEi Southern California Chapter��Economic Market Intelligence | �Applying Machine Learning to Predict Project Cost and Schedule��April 23, 2026

AtkinsRéalis - Sensitive / Sensible (FR)

2 of 28

Introductions
US Market Intelligence
Southern California Market Intelligence
Applying Market Intelligence and ML to Projects
Recap Key Messages
Q&A

AtkinsRéalis - Sensitive / Sensible (FR)

Agenda

3 of 28

INTRODUCTIONS

AtkinsRéalis - Sensitive / Sensible (FR)

Who we are
Where we work
What we do
What I do

4 of 28

Who We Are

4

AtkinsRéalis

We’re AtkinsRéalis, a world-class engineering services and nuclear company.

Together, with our industry partners and clients, and our global team of consultants, designers, engineers and project managers, we can change the world and engineer a better future for our planet and its people.

UK & Ireland

14,000

Canada

8,500

United States

5,500

Latin America

2,000

Middle East & Africa

4,000

Asia Pacific

6,000

Our global team of over 40,000 employees speaks over 70 languages, and represents 130 nationalities across six continents.

5 of 28

Where we work

From designing entire cities to delivering nuclear power stations and transforming manufacturing systems, we focus our business in the areas that have the most impact on the way we all live and the resources we demand from the planet.

Building �& places�

Cities & communities

Social

Commercial

Residential property

Transportation

Rail & transit

Roads

Aviation

Port

Defense�

Aerospace

Defense

Security

Industrial�

Life sciences /

Pharma

Advanced / Hi-tech manufacturing

General manufacturing

Water�

Water and

wastewater utilities �– Treatment and resources

Environment protection,

regulation and resilience

Nuclear�

New build

Reactor support

and life extension

Environmental remediation

Power & �renewables�

Power grids

Hydropower

and dams

Alternative energies

and technologies

Minerals �& metals�

Minerals�Metals

5

6 of 28

Our commitment to a whole-life view of major development programmes enables us to lead projects at every stage and ensure that, wherever we are involved, our people have a wider view of the challenge to better guide our clients and partners.

Project �delivery

Decommissioning

Consulting, strategy �& advisory

Engineering �& design

Project & program management

Operations & maintenance (O&M)

Capital

What we do

6

7 of 28

What I do

7

Cost Estimating

Detailed estimates

Labor, materials and equipment build-up capability

Forced detail at early stages of design – to inform target value design approach

Construction Market Intelligence

Analyze data on labor, materials, market activity, supply chain considerations and bid costs to advise on market risks

Benchmark Database Development

Enable clients to develop top-down benchmark databases

Greatly enhance their budgeting and ability to challenge contractor estimates and schedules

Cost & Schedule Modelling

Tools to dynamically develop a budget and design parameters that match the project requirements

Reducing the waste from designing a concept a project cannot afford

Preconstruction Reports

Bringing these services together in support of our PM/CM delivery services, as interactive deliverables

8 of 28

US MARKET INTELLIGENCE

AtkinsRéalis - Sensitive / Sensible (FR)

Long-term construction material and bid prices trends
Global supply chain considerations
Trends in construction activity

9 of 28

National Bid Price and Escalation Trends

AtkinsRéalis - Sensitive / Sensible (FR)

Pre-COVID (pre-2020)

Steady 2%-3% bid price escalation / year
Limited variation in of input prices

Pandemic Period (2020-2022)

Massive ~40% increase in 2-year period
Material prices rose first, bids caught up 6-months later

Post-Pandemic (2023-2024)

Two years of historically low escalation in bids and materials

New Uncertainty (2025-now)

Tariffs and global conflict have spiked material prices and bids are expected to increase

10 of 28

Global Supply Chain – Tariff Rates

10

AtkinsRéalis

Tariff	Authority Issued	Approximate Tariff Rate	Effective Date	Risk Level for Construction
Steel	Section 232	50% raw metal, 25% for derivatives, 15% for equipment	6/4/2025 (restructured 4/6/26)	High - Evidence of high tariff pass-through
Aluminum	Section 232	50% raw metal, 25% for derivatives, 15% for equipment	6/4/2025 (restructured 4/6/26)	High - Supply squeeze and tariffs combine for high price pressure (substantiated in PPI data)
Copper	Section 232	50% raw metal, 25% for derivatives, 15% for equipment	8/1/2025 (restructured 4/6/26)	High - Copper a significant element of electrical components
Brand-Name or Patented Pharmaceuticals	Section 232	25% (with potential escalation)	4/2/2026	Driving greater investment in Life Sciences in US
Automobile Parts	Section 232	25% (reduced effective rates for UK, EU and Japan under bilateral frameworks.)	4/3/2025 for vehicles	Low - Added fleet maintenance and procurement costs
Automobile Parts	Section 232		5/3/2025 for parts	Low - Added fleet maintenance and procurement costs
Medium and Heavy Trucks	Section 232	25%	11/1/2025	Medium - Added procurement costs
Softwood Timber and Lumber	Section 232	10%	10/14/2025	Medium – big impacts for residential, some impact for all projects (E.g., formwork)
Kitchen Cabinets and Bathroom Vanities	Section 232	50% (reduced effective rates for UK, EU and Japan under bilateral frameworks.)	10/1/2025	Significant – Interiors projects seen notable escalation
Canada (Fentanyl-related)	IEEPA^a	35%; 10% on potash and Canadian energy. USMCA exemption.	8/1/2025 (35% rate)	Medium – Indirect, but lots of lumber and steel imported from Canada.
Mexico (Fentanyl & Migration)	IEEPA^a	25%; 10% on potash. USMCA exemption.	3/4/2025 (25% rate)	Low - Indirect effect due to BABA requirements
China (Fentanyl-related)	IEEPA^a	20%	3/3/2025	Low - Indirect effect due to BABA requirements
India (Russian Oil)	IEEPA^a	25% (with exceptions)	8/27/2025	Low - Indirect effect due to BABA requirements
Reciprocal/Global Tariffs (Trade Deficit)	IEEPA^a	10-41% by country of origin (with exceptions)	8/7/2025 (10/5/2025 for goods in transit)	Mixed

11 of 28

Global Supply Chain – Oil Price Volatility

“Largest oil supply shock in history!”

In terms of by peak daily supply loss

The Iran war and effective closure of the Strait of Hormuz caused peak daily supply losses exceeding 12 million barrels per day (mb/d)
By peak daily volume disrupted, this exceeds:

1973–74 Arab oil embargo (~4.5 mb/d)
1978–79 Iranian Revolution (~5.6 mb/d)
1990 Gulf War (~4.3 mb/d)
2022 Russia–Ukraine oil shock (net ~2–3 mb/d after rerouting)

11

AtkinsRéalis

Then why aren’t oil prices higher?

400 million barrels of emergency supply released (unprecedented!)
This crisis has not gone on long
The market has largely bet it won’t last long
Demand-destruction -> less economic and construction activity

12 of 28

Global Supply Chain – Oil Price Volatility

Diesel prices have risen fastest

More globalized supply chain
Middle east crude suitable for diesel production
Existing supply constraint on Russian crude (also Diesel feedstock)

Immediate impacts

Increased freight costs
Increased on-site fuel cost

Medium-term impacts

Additional price pressure for oil dependent or energy-intensive materials

Asphalt
Steel
Concrete

12

AtkinsRéalis

13 of 28

National Construction Activity Trend

AtkinsRéalis - Sensitive / Sensible (FR)

14 of 28

Data Center - Hyperscale-at-Hyperspeed!�

Massive growth since 2023
Only ~3.5% of non-res construction
Gargantuan Investments Planned

Amazon ~$200bn
Microsoft ~$140bn
Google ~$180bn
Meta ~$125bn
Oracle ~$50bn
~40%/60% split of construction/server spend

~$280bn in construction cost on accelerated schedules

AtkinsRéalis - Sensitive / Sensible (FR)

15 of 28

SOUTHERN CALIFORNIA MARKET INTELLIGENCE

AtkinsRéalis - Sensitive / Sensible (FR)

Different trend in construction activity
Labor availability overview
California construction process

16 of 28

CA Not invited to AI Data Center Party…

AtkinsRéalis - Sensitive / Sensible (FR)

Some significant investments in CA

Biggest investments are elsewhere

Virgina
Texas
Ohio
Southeast

17 of 28

Planned Activity in California

Massive Rail & Transit Programs

HS Rail ($200bn+)
BART (~$113bn)
LA Metro ($50bn+)

Significant Multi-Family Residential
Power Infrastructure

Major Solar Projects
Battery Storage

Airports

Significant activity at LAX
SAN T1 Replacement ($3.8bn)

Leisure & Hospitality

SD and LA convention centers
Limited hotel activity before World Cup and Olympics

AtkinsRéalis - Sensitive / Sensible (FR)

18 of 28

Labor Market – Declining Construction Employment

AtkinsRéalis - Sensitive / Sensible (FR)

19 of 28

Labor Market – Better Than Most States

AtkinsRéalis - Sensitive / Sensible (FR)

*Data from recent labor availability study covering 100-mile radius of Thousand Oaks, CA.

20 of 28

Challenges: California Construction Process

Typical California Construction Headlines:

“California’s largest data center may be built in their backyard. Not if they can stop it.”

(inewsource, April 2026)

“The CEQA graveyard: Projects delayed by California’s powerful environmental law”

(Las Vegas Sun / Bay Area News Group)

“California’s broken permitting system faces scrutiny in Assembly report”

(Davis Vanguard, March 2025)

“Navigating the permitting thicket: Major energy project on hold”

(POLITICO, January 2026)

“Local permitting blamed for cost overruns and slow delivery of state megaprojects”

(Streetsblog California)

Big, unpredictable impact on schedule and indirect costs of projects

AtkinsRéalis - Sensitive / Sensible (FR)

21 of 28

APPLYING MARKET INTELLIGENCE AND ML TO PROJECTS

AtkinsRéalis - Sensitive / Sensible (FR)

ML escalation forecasts
Enhancing uncertainty analysis with ML models
Applying ML to cost and schedule modelling tools

22 of 28

ML Escalation Forecasts

Escalation forecasts = blend of 3 models.

Two traditional models

ARIMA (auto-regressive integrated moving average)
Long-term average (excluding COVID period)

One machine learning model

NNETAR (neural-network auto-regression)

Strengths

Performs better when costs jump faster than history suggests (e.g. pandemic)
can learn threshold effects:

escalation accelerating after sustained inflation
flattening after demand destruction

Limitations

Black box – not easy to explain why it is
NNETAR extrapolates patterns—it does not “reason”.

Guardrails

Needs experienced judgement to reason and apply market understanding
Useful to shape ranges - not the answer

AtkinsRéalis - Sensitive / Sensible (FR)

23 of 28

Enhancing Uncertainty Analysis with ML

AtkinsRéalis - Sensitive / Sensible (FR)

Market Analysis

Local market history

Price trends

Construction activity levels

Predictions for tariffs and oil price impacts on US material prices

Labor Analysis

Quantify:

Labor availability,

Labor cost

Labor productivity

Enables labor impact on bid prices to be stripped away

Survey Data

Survey internal staff

Survey contractors

Survey vendors

Understand their concerns regarding cost and schedule risk

Cost Model

Utilize benchmark data

Break out labor, materials, equipment and mark-ups for capital program.

Break out proportion of capital cost where uncertainty (e.g. tariffs/oil price) applies

ML Analysis of Bids

Elasticity Analysis

Calculate how much a change in material price results in a change in bid price

Monte Carlo Analysis

We don’t know how tariffs and oil prices will change in the future

Scenario analysis for direct tariff/oil price impacts

Risk register of indirect impacts

Approach to Assess Tariff and Oil Price Impact:

24 of 28

Enhancing Uncertainty Analysis with ML

AtkinsRéalis - Sensitive / Sensible (FR)

Which ML models suit different market shocks?

Any answer requires project specific justification

Slower acting issue (E.g. Tariffs)

OLS Regression
Allows isolation of effects of time and economies of scale
Explainable, defensible elasticity estimates
Results were used as one input, not the sole forecast driver

Faster acting issues (e.g. Oil price shocks)

More sophisticated models required also required
Need to account for non-linear impacts and different responses of the market

25 of 28

Applying ML to Cost and Schedule Modelling

Case Study

Challenge: make a cost and schedule model to give instant answers (class 5 accuracy)
Program of 18 small projects – refurbishing gas regulator stations
Success Requirement – needs to be much easier to accurately replan projects when budgets and priorities shift

AtkinsRéalis - Sensitive / Sensible (FR)

Program Information	Helpful for ML Modelling
6 completed, 12 in progress	🗶🗶
$3m - $20m cost range for projects	🗶
Standard P6 schedule (not resource or cost loaded)	✔
Standard cost reports for each project	✔
Alignment between cost and schedule WBS	✔✔
Equipment list for each project	✔✔✔
No clear scaling factor (e.g. cost/lf, cost/cfm, etc.)	🗶🗶🗶

26 of 28

Applying ML to Cost and Schedule Modelling

Approach

Simple parametric cost model – based on equipment specification
ML model to find ‘most-similar’ comparison projects – based on equipment specification and project estimate characteristics
Develop schedule modelling tool – ML function averages durations and start dates of most similar projects and adapts to estimate characteristics
Now we have estimated cost and duration for each WBS item
Spend curve become Sigmoid function – fitting shape of s-curve between 0 and 1
ML Model utilizes ‘most-similar’ historic data to generate ‘most-likely’ sigmoid function for spend curve – then applies the actual cost over the dates

AtkinsRéalis - Sensitive / Sensible (FR)

27 of 28

KEY MESSAGES

AtkinsRéalis - Sensitive / Sensible (FR)

US Market Outlook is Most Uncertain it has been in Post-Pandemic Period

Activity relatively flat, growth driven by data centers

Future material price pressures from tariffs and oil price shock

Expect prices to rise faster than recent past, but not as fast as pandemic era

California Market is Significantly Different to rest of US

Spend driven by massive rail and public infrastructure works, not invited to hyperscale DC party ☹

Craft labor availability better than most states, with some specific shortages (ironworkers)

Challenges around unpredictability of permitting durations and legal challenges on large-scale projects

Applying ML to Cost and Schedule Data

Every application that we have developed so far still needs human input for decision making

Market intelligence is key to sense-check the outputs from ‘black-boxes’

We’re just scraping the surface of what is possible

28 of 28

Q&A

AtkinsRéalis - Sensitive / Sensible (FR)