The Uncomfortable Truth About AI Investments

An Apollo Global Management economist recently made headlines with a pointed assessment: outside of the technology sector, artificial intelligence has delivered zero financial payoff to corporate America. Despite billions spent on AI tools, subscriptions, and pilots, the needle on business performance has barely moved for most companies.

For Canadian mid-market business leaders who have been watching AI adoption with cautious curiosity — or who have already made early investments — this finding deserves serious attention. Not as a reason to disengage from AI, but as a signal to change the question being asked.

The issue isn't whether AI works. It's whether companies are building the right conditions for it to work.

The Model Obsession Is a Distraction

For the past few years, most of the AI conversation has centered on model capability: which large language model is most powerful, which has the best reasoning, which offers the largest context window. Vendors have been happy to fuel this framing because it keeps the focus on their product.

But according to MIT Technology Review's recent analysis of enterprise AI architecture, the next competitive advantage has less to do with which model you choose than with what you build around it. This is a meaningful shift in thinking — and one that aligns with why so many AI initiatives stall after the pilot phase.

Switching from GPT-4 to Claude, or from one AI platform to another, rarely produces the transformational gains companies expect. The model is, increasingly, a commodity. What isn't commoditized is your data, your workflows, and your ability to connect AI output to real business decisions.

What 'Zero Payoff' Actually Means for Mid-Market Companies

When Apollo's economist describes zero financial payoff, he's describing a pattern that most operations leaders will recognize: AI tools get adopted, employees experiment with them, a few efficiency gains appear — and then the results plateau well below the investment threshold.

This happens for predictable reasons:

  • AI is deployed as a standalone tool, not embedded in the workflows where decisions actually happen.
  • Data quality and accessibility haven't been addressed, so AI operates on incomplete or siloed inputs.
  • There's no measurement framework to connect AI activity to revenue, cost reduction, or speed-to-market.
  • Change management is underestimated — people revert to familiar processes when AI tools create friction.

For a Canadian manufacturer, professional services firm, or regional retailer, these aren't abstract concerns. They're the operational realities that determine whether an AI subscription becomes a line item you justify at renewal or a capability that compounds over time.

The Infrastructure Question Every Leader Should Be Asking

MIT Technology Review describes the solution as an AI serving harness — the technical and organizational infrastructure that surrounds a model and allows it to deliver consistent, reliable, business-relevant output. Think of it as the difference between buying a high-performance engine and actually building the vehicle around it.

For mid-market companies, this translates into three practical priorities:

1. Data pipelines before model selection. Before asking which AI tool to use, ask what data that tool will need, where that data currently lives, and how clean and accessible it is. An AI connected to well-structured, current business data will outperform a more powerful model operating on fragmented inputs.

2. Workflow integration over standalone deployment. AI generates ROI when it changes what people do and how fast they do it — not when it sits adjacent to existing processes. The integration point matters more than the tool itself.

3. Measurement from day one. Define the metric you expect to move — quote turnaround time, customer response rate, cost per lead, inventory accuracy — before deploying. Companies that can't measure AI impact can't manage it, and can't defend the investment internally.

What This Means for Your AI Strategy in 2025

The Apollo finding shouldn't discourage investment in AI. It should sharpen how that investment is structured. The companies that will pull ahead aren't necessarily those who adopted AI earliest — they're the ones who built the operational foundation to extract value from it.

For Canadian mid-market companies, this is actually an opportunity. Most competitors are in the same position: tools deployed, returns unclear. The window to build durable AI infrastructure — and the workflows that make it pay — is open right now.

The question isn't whether to invest in AI. It's whether you're investing in the right layer.