Building the Business Case for AI: The NAFTA Framework in Action

In a previous article, I explored why AI doesn’t replace people — it reimagines their role. But that raises a tough question:

If you’re not reducing headcount, how do you justify the investment?

AI strategies are easy to write. Much harder to fund, implement, and scale. The gap between hype and impact usually comes down to one thing: the business case doesn’t hold up.

That’s where NAFTA comes in — a five-step framework for identifying, testing, and scaling AI opportunities that actually deliver value. In this article, I’ll break down how it works — and why this structured approach matters more than ever in 2025.

📌 Needs: Solve a Real Business Problem

The best AI initiatives don’t start with a model. They start with a problem worth solving. AI is a prediction engine — it creates value by estimating what comes next: the next intent, next action, next risk (Agrawal, Gans & Goldfarb, 2018). Its strength lies in augmenting strategic decisions, not making them for you. So ask:

• What’s broken today?
• Where are decisions too slow, manual or inconsistent?
• What signal, if predicted better, would drive real value?

This is where AI excels: ticket triage, sentiment detection, churn prediction, fraud detection — narrow, high-volume use cases with measurable returns. A 40% reduction in backlog? That gets executive buy-in faster than a moonshot (Davenport & Ronanki, 2018).

🎯 Alignment: Tie AI to Strategy, Not Just Innovation

AI should accelerate your strategic goals, not create side projects. Many businesses fail here — launching AI pilots without tying them to clear KPIs or functional owners. The most successful companies do the opposite. They integrate AI into existing business drivers:

• Reducing operating cost
• Improving agent or customer experience
• Increasing conversion or retention
• Enhancing decision speed and accuracy

And they focus on domains where decisions are frequent and structured — like customer service, IT ops, logistics, finance, or sales (Chui et al., 2018).

🧱 Finance: Build a Scalable, Commercial Model

AI must make financial sense. It’s not just about how much it costs to build — it’s about:

• Whether it scales without linear costs
• Whether models are reusable across use cases or markets
• Whether it reduces cost per contact or boosts revenue per head

Top performers treat AI like a product: with P&L accountability, reuse strategy, and ROI guardrails (Ransbotham et al., 2018). Cost models must account for total cost of ownership (tech + people): whether you buy a solution like Zendesk AI, fine-tune open-source models, or build in-house, your platform needs automation-ready workflows, clean training data, and sustainable infrastructure (Bughin et al., 2017).

🧪 Test: Prove Value, Fast

You wouldn’t launch a product without a beta. Treat AI the same way. A strong pilot answers five questions:

• Is it technically feasible?
• Does it fit into existing workflows?
• Are early outcomes positive?
• Is the data stable and clean?
• Do users find it usable?

Start small, measure everything, and iterate fast. High-impact pilots become foundations for scale — not one-off experiments (Flanding, Grabman & Cox, 2019). In fact, AI leaders are 3x more likely to run frequent tests and iterate based on KPIs ( McKinsey & Company , 2023).

🔍 Analyse: Manage Risk, Build Trust

From biased hiring tools to flawed credit models, AI has already caused reputational damage in real-world systems (O’Neil, 2016). Responsible teams build in explainability, auditability, and human oversight from day one — not as an afterthought (Rahwan et al., 2019). Even great AI fails without responsible governance. That means:

• Data integrity: Is the training data current, complete, and representative?
• Fairness: Could the model entrench bias or produce discriminatory outcomes?
• Accountability: Who owns the output when AI fails?

As I mentioned in my article The Hidden Costs of Building AI for CX, the EU AI Act and global regulation ramping up, governance isn’t optional — it’s operational and a potential €55+ million fine if you get it wrong!

📈 From Test to Scale: Monitor, Retrain, Improve

AI doesn’t stay accurate forever.

• Models drift
• Data shifts
• Users evolve
• Interfaces change

That’s why post-deployment governance is essential. The best AI teams:

• Track precision, recall, and business KPIs
• Capture human feedback loops
• Retrain regularly to keep up with signal changes
• Embed AI into workflows — not bolt it on top (Chui et al., 2021)

According to McKinsey & Company (2023), mature adopters are 4x more likely to have defined AI ownership structures and retraining cycles in place.

🛠️ In Practice: Scaling AI in Customer Service

Many Zendesk customers follow this structure to scale AI in customer experience. They start with a defined issue — like a spike, a backlog or inconsistency — and align AI to outcomes like resolution time or CSAT. They test with real tickets, validate ROI, and expand gradually — with control, not chaos.

Zendesk’s latest capabilities reflect this shift: moving from passive support to agentic AI — where autonomous AI Agents can handle multi-step workflows, trigger updates, reference knowledge, and collaborate with humans. Combined with tools like Copilot, and intelligent triage, this empowers teams to:

• Resolve complex queries without human input
• Support asynchronous journeys
• Deliver faster, more consistent outcomes — at scale

All in a platform that’s flexible, secure, and auditable by design.

✅ The Bottom Line

The companies winning with AI aren’t chasing headlines. They’re applying structure. Testing rigorously. Aligning to strategy. Measuring value. And building trust — not just models. NAFTA helps you do exactly that — turning AI from hype into habit, one use case at a time.

References

• Agrawal, A., Gans, J.S. & Goldfarb, A. (2018). Prediction Machines: The Simple Economics of Artificial Intelligence. Harvard Business Review Press.
• Bughin, J. et al. (2017). Artificial Intelligence: The Next Digital Frontier? McKinsey Global Institute.
• Chui, M. et al. (2018). Notes from the AI Frontier: Applications and Value of Deep Learning. McKinsey & Company.
• Chui, M. et al. (2021). The State of AI in 2021. McKinsey & Company.
• Davenport, T.H. & Ronanki, R. (2018). Artificial Intelligence for the Real World. Harvard Business Review.
• Eccles, R.G., Newquist, S.C. & Schatz, R. (2007). Reputation and its Risks. Harvard Business Review.
• Flanding, J.P., Grabman, G.M. & Cox, S.Q. (2019). The Technology Takers: Leading Change in the Digital Era. Emerald Publishing.
• Floridi, L., Cowls, J. & Taddeo, M. (2023). AI and the Law: Understanding the EU AI Act. Oxford Internet Institute.
• McKinsey & Company. (2023). The State of AI in 2023.
• O’Neil, C. (2016). Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy. Crown Publishing.
• Rahwan, I. et al. (2019). Machine Behaviour. Nature, 568(7753), pp.477–486.
• Ransbotham, S. et al. (2018). Artificial Intelligence in Business Gets Real. MIT Sloan Management Review.