How to close the gap between AI investment and impact in 2026

2025 has been the year of AI experimentation. According to a recent McKinsey study, almost all companies invest in AI, but many are still finding their feet when it comes to implementing it effectively with just 1 percent believing they are at ‘AI maturity’. While this makes sense given how new AI is, the gap between investment and impact represents a real opportunity.

In our latest webinar, Ophelos co-founder Paul Chong reflected on AI's progress in 2025 and shared his thoughts on where AI is heading in 2026. His main takeaway aligned with what we're seeing across the industry: companies have adopted AI, but many aren't taking full advantage of their investments yet. 

During the webinar, Paul highlighted 5 key steps that companies can take now to bring real business value and close the gap between the vision companies have for AI and where they are currently at.

1. Place bets and accept that some will fail

"You've got to place some bets. You've got to accept that some of these investments are going to go down a rabbit hole and you're not going to get them back."

There won’t always be a safe bet in AI and companies that wait for certainty could find themselves playing catch-up while their competitors build capability through trial and error.

What this looks like in practice:

• Run multiple experiments in parallel
• Set clear timeframes for evaluating success
• Budget for failure as part of your strategy
• Treat unsuccessful pilots as learning opportunities, not wasted investment

The companies that are succeeding at successfully using AI aren't the ones that never failed but are the ones that fail faster, learn quicker, and iterate constantly.

2. Start simple, not complex

"Some of the simpler use cases, the less obvious things like just managing documents better is where you can have a massive impact."

The temptation early on is to chase transformative use cases but simple, high-volume tasks can deliver measurable ROI today and grow over time. For example: Payment notifications. Document workflows. Customer reminders.

At Ophelos, we started using voice AI for simple payment reminders, this wasn't a multi-step problem-solving task. We picked a task that had clear boundaries, high volume, and immediate measurable value. Today that same system handles 80,000+ calls per month.

How to identify the right starting point:

• Look for tasks your team does hundreds of times per week
• Avoid processes with complex system dependencies
• Choose workflows where success is immediately measurable
• Start with limited scope, then expand

3. Have strong observability and monitoring

"You're making sure you're observing and monitoring everything going on. The technology helps you do that, making sure things are working and being able to hit the kill switch if anything goes wrong."

This becomes especially important when your AI touches customers, you can't deploy it and then hope for the best. Guardrails need to be in place so that you always have a safety measure for if anything goes wrong.

Practical steps:

• Start with under 10% of users for any new deployment
• Define clear metrics for what "working" looks like
• Assign a human in the loop to actively watch over the system initially
• Build manual override capabilities and clear escalation paths
• Set regular checkpoints to continue, pause, or stop
• When we deployed Voice AI, we started with small sample sizes and built in constant monitoring. This approach lets you prove you can control it before you scale it. Over time, you can reduce human oversight as confidence builds.

4. Connect use cases carefully to compliance requirements

"You need to go into it very thoughtfully. You need to adhere to compliance in the given market."

In regulated industries, this won’t be optional, but even in less regulated sectors, connecting AI to compliance, legal, and risk frameworks is extremely important.

Questions to ask before deployment:

• What regulations govern this use case?
• What approvals do we need before going live?
• How do we ensure AI decisions are explainable and auditable?
• What happens if the AI makes a mistake?
• Compliance shouldn't be a blocker or slow you down; it should guide your testing. By knowing the boundaries, you can then test aggressively within them.

5. Think about workflows, not just individual tools

"You'll hear a lot more about workflows in 2026. We're shifting from "what task can AI help with?", to "what entire workflow can AI own?"

Paul's analogy is to think of AI agents like team members with specific roles:

• A compliance agent (connected to regulatory documentation)
• A customer service agent (expert at empathetic communication)
• A negotiation agent (specialises in payment plans)

Together, they handle complete workflows that used to require multiple handoffs and human decision points.

The shifts the conversation from: "AI can help me draft emails faster" to: "AI can manage the entire customer onboarding workflow"

The companies treating AI as a productivity tool will see incremental gains but the ones treating it as a workflow transformation will see exponential impact.

What separates leaders from laggards in 2026

"I think it's going to be the year of impact."

The competitive advantage isn't in having AI. It's in building the capability to execute it effectively. That means placing strategic bets, starting with simple high-value use cases, monitoring rigorously, respecting compliance boundaries, and thinking in complete workflows rather than isolated tasks. The companies that succeed at this will lead their markets in 2026.

Learn about our AI-driven debt collection.