1. You are here:
  2. Home
  3. Resource library Search
  4. What is the Third Way to Successful AI Adoption?
Member News Cambridge Management Consulting
18 Sep 2025

What is the Third Way to Successful AI Adoption?

In this article, Tom Burton, Partner at Cambridge Management Consulting for Cyber Security, outlines an approach to leveraging Artificial Intelligence which balances its transformative power with safety, security, and authenticity.

Every day brings fresh AI headlines - euphoria at one end, existential dread at the other. Inside organisations, that same polarity shows up as two unhelpful tribes: ‘adopt at all costs’ and ‘do nothing until it’s proven’.

The first risks chaos; the second risks irrelevance. There is a smarter path between them - a practical ‘Third Way’ that couples ambition with control. The goal here is: deliver measurable value from AI while keeping risk tolerable, repeatable, and visible.


Start With Real Problems, Not Shiny Tools

  1. Before touching models, define the business problems worth solving. Be specific.
  2. Cut cost in targeted operations, not ‘everywhere’.
  3. Improve customer experience for a well-chosen journey, not ‘all CX’.
  4. Accelerate content or decision cycles in defined workflows, not ‘marketing in general’.

Translate big ambitions (your 'Mid-sized Hairy Audacious Goals’ (MHAGs)) into discrete, bounded initiatives that can be prioritised independently. Keep them loosely coupled so a blocked idea doesn’t stall everything else. If a use case can’t be measured, constrained, or owned, it’s not ready.


Define ‘What Good Looks Like’

You can’t design what you haven’t defined. For each initiative, write down:

  • Objective & Approach: Data needed, how it will be used, what the system will do, and where people fit.
  • Quality Thresholds: What a high-quality output is, what a low-quality one is, and the minimum acceptable level.
  • Edge Cases: Don’t test ‘2+2’ if the real task is fourth-order differentials; challenge the solution with the hardest representative scenarios.

This discipline pays off in Proof of Concept (PoC) and Proof of Value (PoV). You’ll know if the concept works to the quality required - and you’ll know it objectively.


Anticipate GenAI-Specific Risks Early

Modern AI isn’t deterministic. Outputs depend on prompts, training data, updates, and context. Map the risk landscape before you scale.

  • Hallucinations: When models lack information, they can produce plausible fiction. Ask: what happens if the system confidently invents an answer in this workflow?
  • Bias (and Model Collapse): Training data reflects human and structural biases; as AI-generated text floods the web, models trained on model output degrade over time. Consider both fairness and long-term quality.
  • Unintended Behaviours: agents can optimise the goal in ways you didn’t expect. If asked to win, do they try to sidestep rules? Design tests that probe for rule-bending rather than assuming it won’t happen.
  • Information security:
  • Contextual Blindness: Without strong scoping, tools may mingle confidential data across clients or teams.
  • Data use in training: Know whether your tenant data can train models beyond your boundary.
  • Scope Creep: Restrict what the model can access to the least necessary.
  • AI-targeted Attacks: Poisoning data, prompt injection, model inversion - plus all the usual cyber risks. Treat these like you would any critical system, with controls aligned to regulation and data sensitivity.

The question isn’t ‘Is there risk?’ but ‘Is the juice worth the squeeze - and how will we contain it?’


Prove Value Before You Scale

With risks mapped and success defined, move deliberately:

  • PoC: Demonstrate the idea can work at small scale and still meet your edge-case thresholds.
  • PoV: Run the system on real work, with guardrails. Compare against the legacy process or add human checks. Confirm benefits outweigh costs.
  • Production: Adopt with training, policy updates, and change management so value actually lands. Keep oversight - models and ecosystems evolve, so trust must be earned and re-earned over time.

Remember the hidden cost: unchanged processes. Dropping new tech onto old ways of working is just a costlier version of the same thing. Redesign the work, not just the toolset.


Conclusion

The Third Way isn’t a compromise; it’s a method. Start with real problems. Define success and edge cases. Confront GenAI’s distinctive risks where they actually live: in data, behaviour, and security. Then prove value and scale with ongoing supervision.

Do that consistently and you’ll avoid the two traps that dominate the headlines: the rush that breaks things and the caution that misses the moment.

To learn more about how Cambridge Management Consulting can refine your AI adoption strategy, contact Tom Burton here: https://www.cambridgemc.com/people/tom-burton

Return to listing