The Growing Temptation of AI in Operations
A 2025 study from MIT’s NANDA Initiative suggests that nearly 95% of AI pilots struggle to demonstrate measurable ROI. Despite more than $40 billion in enterprise investment globally, only a small minority of initiatives meaningfully impact P&L performance.
This is not a failure of the technology itself. More often, it reflects gaps in strategic clarity, process discipline, and decision ownership.
Over the last 12–18 months, AI has rapidly moved from a niche discussion to a boardroom priority.
In many organisations we interact with – particularly small and mid-size companies – AI is now being discussed as a solution to:
-
- Planning complexity
- Inventory challenges
- Decision delays
- Data overload
- Talent constraints
- Sales & customer support
The attraction is understandable. Operations are becoming harder to manage. Data volumes are exploding. Decision windows are shrinking. And experienced operational talent is stretched thin.
AI appears to promise clarity, speed, and control — all at once.
Yet, despite the excitement, many AI initiatives in operations quietly fail to deliver meaningful value.
Not because AI doesn’t work – but because of how it is being approached.
Why AI Sounds Like the Perfect Answer
From a leadership perspective, AI appears to offer something deeply appealing:
-
- Intelligence without adding headcount
- Insights without manual effort
- Automation without operational disruption
- Better decisions without deeper structural change
For organisations already under pressure, AI feels like a way to leapfrog operational maturity.
But operations rarely reward shortcuts.
And this is where the gap between expectation and reality begins to widen.
Where AI Initiatives Actually Break Down
When we look at AI initiatives that struggle or stall in operations, the failure is rarely technical.
The patterns are remarkably consistent:
-
- AI is layered over broken or inconsistent processes
- AI models are fed with unreliable, poorly governed, or incomplete data
- Outputs are generated, but no one fully trusts them
- Insights exist, but decision ownership is unclear
- Recommendations are produced, but execution doesn’t change
In such environments, AI doesn’t simplify operations, but it simply amplifies existing confusion!
This leads to a dangerous outcome: AI becomes another reporting layer, not a decision enabler.
It’s important to recognise this clearly:
AI does not fail because it is inaccurate. AI fails because the organisation and processes around it are unprepared.
The Core Mistake: Starting with Data Instead of Decisions
Most AI initiatives begin with questions like:
-
- “What data do we have?”
- “What models can we build?”
- “What insights can we generate?”
But high-performing operations start somewhere else entirely.
They start with decisions – decisions that can uplift operational performance and benefit business.
Questions such as:
-
- Which decisions materially impact cost, service, or working capital?
- How frequently are these decisions made?
- Who owns them?
- What happens if they are wrong?
- What information genuinely improves decision quality?
Without clarity on these questions, AI outputs – no matter how sophisticated the model – struggle to find relevance.
This is why many AI dashboards may look impressive but change very little on the ground.
What “Use-Case Driven AI Adoption” Really Means
A use-case driven approach reverses the usual sequence.
Instead of asking what AI can do, it asks:
Where do we repeatedly make difficult decisions with imperfect information?
A practical way to think about this is:
Decision → Data → Discipline → AI
-
- Decision: Clearly define the operational decision that matters
- Data: Identify the minimum reliable data required to support it
- Discipline: Ensure the process and ownership around that decision are stable
- AI: Apply intelligence only where it enhances speed, accuracy, or consistency
In this model, AI is not the end in itself. It is the force multiplier.
Why Mid-Size Companies Often Misjudge AI Readiness
Mid-size organisations face a unique challenge.
They are:
-
- Too complex for intuition alone
- Too resource-constrained for large experimental programs
- Under constant pressure to “modernise quickly”
As a result, AI is sometimes viewed as a way to compensate for:
-
- Process gaps
- Data discipline issues
- Decision ambiguity
Unfortunately, AI is unforgiving in such environments.
It does not compensate for weak fundamentals; it instead accentuates them through unreliable and misleading outputs.
The Advantage Mid-Size Firms Actually Have
However, there is a powerful upside.
When approached correctly, mid-size firms can often extract value from AI faster than large enterprises, because they have:
-
- Shorter decision chains
- Closer alignment between leadership and execution
- Faster ability to pilot and course-correct
- Less organisational inertia
But this advantage is realised only when AI is introduced after operational clarity, not before it.
Mid-size firms don’t need enterprise-scale AI programs. They need focused, decision-centric intelligence.
A Reality-Based View of AI in Operations
AI is not a maturity accelerator.
It does not replace:
-
- Process discipline
- Data quality and ownership
- Clear accountability
- Operational leadership
What AI does exceptionally well is multiply the quality of what already exists.
Strong processes become faster. Clear decisions become sharper. Reliable data becomes more valuable.
Weak fundamentals, however, simply become more visible by poor outputs.
Conclusion
For operations leaders, the most important AI question is not:
“Which tool should we adopt?”
It is:
“Which decision are we trying to improve, and are we ready to trust the answer?”
Until that question is addressed honestly, AI will remain more promise than performance.
Complimentary AI Readiness Diagnostic (Operations-Focused)
If you are exploring AI for operations and want a grounded, execution-focused assessment, we offer a complimentary 30-minute AI Readiness Diagnostic.
The discussion focuses on:
-
- Which operational decisions are AI-ready
- Where fundamentals need strengthening first
- How to avoid expensive experimentation that doesn’t scale
If this reflects your current AI initiative or roadmap, a focused diagnostic discussion can help clarify the right starting point. Feel free to reach out through our website or connect with us directly.