What thoughtful AI-augmented support looks like in practice

We’ve been talking about the potential of AI in customer support for years. Now, the shift is becoming a reality and the stakes are higher than ever. The real question is no longer what AI could do, but what happens when it moves from experimentation into the day-to-day operation.

This month's reading shows what thoughtful AI-augmented support looks like in practice, with real outcomes, and how the companies who choose the right outsourcing partner are creating real operational advantage.

What’s worth your attention this month?

1. How Oura built a measurement system that could see beyond traditional CSAT and NPS scores

In Oura's August blog issue, John Moses, their VP of Member Experience, describes how the company developed a new metric called the Total Customer Experience Score (TCX). Their initial problem was one support leaders recognise: traditional metrics like CSAT and NPS reach a small, self-selected group, and manual QA teams can hand-audit only a small percentage of solved tickets. So, as AI handles a growing share of interactions, the blind spot grows, and a single flaw in an AI agent can repeat thousands of times before a survey catches it.

Using large language models, they now evaluate nearly every support interaction across every channel — not just the estimated 5 to 7 percent of tickets a team can manually audit or the fraction of customers who complete a survey. Four AI classifiers score each interaction on issue identification, frustration, resolution, and satisfaction, and an interaction only passes if all four are positive. Since rolling TCX out, their TCX score has climbed nine points, repeated troubleshooting has dropped from 40 to around 20 percent, and median handle time on live chat is down by more than 25 percent.

Read Human-Led at Scale: How Oura Is Redefining Customer Experience in the AI Era

2. How one fast-growing travel technology company reached a 90% AI resolution rate

John Gleeson’s latest article in Success VP profiles Tern, a fast-growing travel technology company now supporting over 12,000 advisors. Their support team is consistently hitting a 90% AI resolution rate month after month, up 20% from last year, while shipping product weekly and scaling the user base significantly. Tern’s high resolution rate is a result of a combination of factors: feeding their Fin model using standing AI projects to continuously process newly shipped product documentation, updating training transcripts, and ensuring they use the findings to shape internal response guidance and public help center content. The result is a knowledge base that understands how customers use the product and how they ask for help. The 10 to 15 percent of tickets that still escalate are handled by a support team that now spends roughly 60% of its time on escalations and the rest on proactive, health-triggered outreach, rather than repetitive Tier 1 tasks.

Read The Brain Behind a 90% Resolution Rate

What do we think about it?

1. The question of what good looks like matters more when AI is doing more of the work

The two articles above share something beyond their results. Both Oura and Tern treat measurement as a continuous activity and distinguish between the interactions AI can handle reliably and the ones that still need a human. And both understand that the knowledge feeding the AI is as important as the AI itself.

What this means in a managed outsourcing context is worth thinking through. The quality of an outsourced support operation depends on the same inputs: how well your partner understands your product, and how thoroughly your knowledge base is maintained. A partner that treats onboarding as a one-time event will find the gap between AI and human performance increasingly difficult to manage as volumes grow, products evolve and customer needs change.

The organisations seeing the strongest results are combining AI technology with experienced support teams who understand both the platforms and the customer journey.

“AI has made our services more valuable, our people more effective, and many of our roles more interesting and fulfilling.”

Aidan O'Shea, Co-Founder and CEO, Otonomee

Read how we think about AI-augmented support

2. Companies evaluating their outsourcing partner are doing more research than you might think

Two recent pieces from Mark Hillary's CX Files Substack offer two perspectives that are worth connecting. The first argues that most BPO companies invest heavily in sales while underinvesting in the trust signals that shape the buyer view before any conversation takes place. By the time a prospective client agrees to a first meeting, they have already reviewed the website, examined LinkedIn commentary, spoken privately to existing clients, and asked an AI platform to compare suppliers and build an initial shortlist. The sales conversation is often closer to the end of the buying journey than the beginning.

The second piece makes the related argument that this is where most BPOs are losing ground quietly. AI-first search rewards evidence, original thinking, and genuine experience instead of volume. A company publishing volume without substance and information that it can't back up with specific client outcomes is just generating noise and decreasing its credibility.

Read Your BPO Doesn't Have a Sales Problem - It Has a Trust Problem

Read Stop Calling AI Slop "Thought Leadership"

Closing thoughts

The thread connecting this month's reading is that quality in AI-augmented customer support is something you can measure, build toward, and show evidence of. The companies doing it well are publishing specific outcomes, keeping their knowledge bases current, and keeping humans in the loop for the inevitable complexity that remains after automation. That specificity is also what makes them visible and credible to the buyers who are now evaluating them, often without ever identifying themselves.

If you are thinking about what the right outsourcing partner looks like for your operation, we would be glad to be part of that conversation.

Get in touch