Alvarez & Marsal · July 2026

A perspective on AI at Gold Fields

Pragmatic. Focused on early value, sustainable foundations, and clarity on ROI. This is not a crisis — but the opportunity cost is real and already accruing.

6 priority questions 5 proven deep dives A 3-decision path this quarter
Our view

AI in mining is an opportunity-cost consideration — pursue it with early value in mind, alongside careful foundational thinking.

Unlike other industries, mining isn't facing an existential AI crisis. So it's fine to be careful. But the opportunity cost of not pursuing AI is substantial and should prompt immediate, staged momentum — the biggest risk today is inaction while "shadow AI" already spreads across the workforce.

01 · Identify

Pinpoint the "no-regrets" moves and pursue them with execution expertise, not a leap of faith.

02 · Establish

Stand up guardrails and a staged rollout of AI tooling with a careful, governed foundation.

03 · Embed

In parallel, build the foundational DNA of an AI-enabled organisation — culture over any single tool.

The agenda

Six priority questions — and one that's critical

You'll have many questions about how to address the AI opportunity. These are the ones that matter most.

Q1 · Strategy

Where is the heaviest concentration of potential AI value in our business?

Q2 · Investment

What is the cost of doing nothing?

★ Critical · Exposure

What is our exposure today from unsanctioned AI use — "shadow AI" — by our own people?

Q3 · Operating model

Where should AI be positioned in our business — under the CTO/CIO, or a new CAIO?

Q4 · Execution

What can we get going with tomorrow, to prove value from AI in the short term?

Q · Strategy & Value

How can we transform our extensive historical data into a legitimate competitive asset?

The critical exposure

Shadow AI is already inside Gold Fields

The question isn't whether our people use unsanctioned AI — it's how much proprietary data has already left the building. It's happening now, it's not hypothetical, and for a miner it's dangerous.

~1 in 2

employees use AI tools their employer never approved — adoption often led by senior staff, not juniors.[1]

69%

of organisations suspect or have evidence staff are using prohibited public GenAI tools (Gartner).[2]

77%

of data pasted into GenAI flows through personal, unmanaged accounts that bypass every corporate control.[3]

20%

of all data breaches now involve shadow AI, at a higher average cost per breach (IBM, 2025).[4]

Crown-jewel geological IP

Drill results, ore-body and reserve models pasted into consumer LLMs move outside our control — and could inform a rival's bid on adjacent ground.

Commercial & M&A leverage

Offtake terms, JV and acquisition positions leaked mid-negotiation destroy deal value.

Market-sensitive disclosure

Guidance, reserve or results text drafted in a public tool risks selective-disclosure and JSE/SEC breaches before official release.

+ Others: safety, ESG leaks, as well as POPIA / GDPR breaches.

  1. 1~1 in 2: BlackFog (2025) — 49% of workers use AI in ways not sanctioned by their employer. Senior-led: TrustedTech Shadow AI white paper (2025) — ~63% of senior decision-makers admit using unapproved AI vs 31% of junior staff.
  2. 2Gartner — 69% of organisations suspect or have evidence their employees are using prohibited public generative-AI tools.
  3. 3LayerX Security, Enterprise AI & SaaS Data Security Report 2025 — 77% of GenAI users paste data into the tools, and ~82% of that pasting occurs via personal, unmanaged accounts that bypass corporate controls.
  4. 4IBM, Cost of a Data Breach Report 2025 — 20% of organisations suffered a breach involving shadow AI, adding ~US$670K to the average breach cost.
Where value concentrates

The mature, high-value, easy-to-deploy options are a small subset

Use-case options are extensive — but the ones worth chasing first are concentrated across the value chain. Follow the flow, stage by stage; ★ marks a flagship deep dive.

Central & support functions

Cross-cutting use cases, grouped by function. Select a function to expand its full list — ★ are flagship deep dives; the $ tier is an indicative cost signal.

Proven, not hypothetical

Five flagship deep dives

The flagship opportunities are concentrated and proven — not a leap of faith. Select any to explore how it works and the value at stake.

Set the foundations first

Just get going — then commit to a regimented change plan

The worst decision is inaction, especially given the likely spread of shadow AI. Ideally, appoint a Chief AI Officer, negotiate terms with AI providers, track token and cost usage on accessible dashboards, and set a roadmap to sharing knowledge and upskilling from the basics.

MalkaA word from Malka · A&M Chief AI & Knowledge Officer
How it comes together

A three-layer operating model, delivered in partnership

A&M and NTT DATA deliver pragmatic early value on sustainable foundations. The partnership stacks as a pyramid — A&M builds AI capability and delivers value at the top, NTT DATA provides the intelligence and infrastructure beneath. Select any tier of the pyramid to reveal its full detail below; click it again to collapse. The A&M capability & value layer is shown first.

Prioritisation

Filter for mature, high-value and easy to implement

Every use case from the assessment is plotted by technology maturity (vertical) against ease of implementation (horizontal, the inverse of change-management difficulty). Dot size reflects value — scaled to the investment tier ($, $$, $$$) as a proxy for the scale of the prize. The sweet spot, top-right, is where mature, high-value and easy-to-land opportunities cluster. Hover or tap any point to read the use case, its value, maturity and ease. Gold rings mark the flagship deep dives. Positions are derived from the assessment's 1–5 maturity and change-management scores; indicative, not a precise ranking.

High maturityLow
Hard to implementEasy
↑ Maturity
Ease →
Value-chain use cases
Central-function use cases
Flagship deep dive
Larger dot = higher value ($ → $$$)

Hover a point to see the use case, its type and where it sits.

The path forward

A value-gated roadmap

Momentum now, discipline throughout — every phase passes a gate before the next unlocks.

Phase 1 · 0–3 months

Prove & design

"Just get going"

Appoint an AI lead, commit seed funding, set the mandate and risk appetite. Run the strategic assessment; design phases 2–3 and target the high-priority rollout areas. Deploy the governed LLM with guardrails from day one.

Phase 2 · 3–9 months

Build the engine

Assessment validates the path

Approve stage-gated funding; endorse the operating model. Stand up the delivery COE — prioritise, pilot, scale — and embed governance, host the first amplified pilots and provide domain owners with sanctioned capability.

Phase 3 · 9+ months

Scale & compound

Pilots prove ROI

Federate ownership to sites; hold the portfolio to ROI. Run a continuous-improvement engine that matures and scales the portfolio, hardening model governance and the value-tracking mechanism.

The ask

Every month of inaction is realised opportunity cost. Convert it to value with three decisions this quarter.

1

Agree the no-regrets path — the de-risked, proven moves that are ready to pursue now.

2

Mandate the team and governance to run it, with clear ownership and accountability.

3

Set the guardrails that make broad AI use safe — turning shadow-AI risk into sanctioned capability.

One last thing

The opportunity is here now