Coaching the Connection: Why a Leading Loyalty Rewards & Customer Care Provider Replaced Prompt-Based QA with LLM-Based Quality and Supervisor-Ready Coaching

An Elite Operation, Built on Human Connection

This leading national provider runs a customer care operation most contact centers would envy. It supports loyalty rewards programs on behalf of major national brands — helping cardholders understand their points, redeem rewards, and get more from programs they might otherwise navigate alone. Many of the callers are older, prefer a person to a self-service screen, and remember how the conversation made them feel long after it ends.

That last part is the whole business. The operation has run its own analysis on what happens after a call: when a cardholder has a genuinely positive interaction, they use that card more often. A good conversation moves top-of-wallet behavior — which card someone reaches for first. In a loyalty business, the contact center is not a cost to be contained so much as a lever on the spending it exists to encourage.

So the team measures the thing that is easy to feel and hard to score: did the agent build a real connection? Authenticating an account is table stakes. The bar is whether the customer hangs up thinking they never want to bank anywhere else. The operation posts 94–95% customer satisfaction favorability on its post-call survey — at a 16% take rate, and without ever mentioning the survey during the call. Those are numbers operations of any size rarely see.

Three Pillars: Satisfaction, Margin, and People

Every decision in the operation runs through three pillars.

The first is satisfaction and quality — both client satisfaction (the brands whose programs they support) and customer satisfaction (the cardholders themselves). Quality is treated as the leading indicator of those outcomes, which is why the team holds an 85% quality bar with real accountability behind it: miss it across two of three months and you move to a structured improvement plan. They debate that standard openly and often, and they keep it, because they believe soft-skill quality is what produces the satisfaction numbers.

The second pillar is margin. When the leader took over the contact center, the business was weighing whether to shut it down as a loss-leader. The strategy was direct: make it unassailably a money maker so no one wants to. The operation now runs against a 30%+ net EBITDA margin target — turning the function from a candidate for the chopping block into a profit center the business wants to keep.

The third pillar is people: growing talent, keeping turnover healthy, developing the supervisors and agents who carry the first two pillars. The three reinforce each other — quality drives satisfaction, satisfaction sustains the margin, and people sustain both.

A Stack That Fought Back: Five9, Verint, and the Support Gap

The contact center sits inside a larger company whose technology team is focused on the loyalty side of the business, not on customer care. That constraint shaped every tooling decision: whatever the operation bought had to work without leaning on internal IT.

That requirement is why the team selected Five9 as its CCaaS platform. After a long prior search, Five9 won because it offered an all-in-one solution the operation could run without external technical help. As a phone platform, it has been stable and dependable — exactly what infrastructure is supposed to be.

The friction showed up underneath it. Quality management ran on Verint, which was subcontracted through Five9 rather than contracted directly. That arrangement created a support gap on the capability the operation cared about most: the team couldn't get the Verint support it wanted, and Five9 wasn't positioned to coach them on how to use it. The result was a roughly two-year effort to stand up in-house Verint expertise just to operate a system they had already paid for — and people who truly know that tooling are not easy to find. The phone platform delivered; the quality layer bolted onto it underwhelmed.

When Quality Means Counting the O's in "Hello"

The deeper problem was how that quality layer actually worked. Verint's scoring was prompt-based: analysts hand-built rules for literal phrasing — did the agent say "hello" with the right number of O's, did they hit a specific scripted line. It is a model that rewards compliance with a checklist.

But compliance is not where this operation lives. Roughly 80% of its scorecard rests on whether the agent established a genuine emotional connection — not whether they followed a script word for word. Prompt-based scoring is poorly suited to that. You cannot reliably hand-write a rule for empathy, and the team's analysts were spending their time engineering prompts instead of improving conversations.

Underneath the scoring tool sat a second kind of noise: reporting. At roughly sixty seats, the operation had more visibility into agent behavior than most centers its size — and far too much of it to act on. Cutting through that noise to know what would actually help an agent improve took real analytical effort. Meanwhile, the math on supervision was sobering: roughly $460,000 a year in supervisor cost was producing about ninety minutes of coaching per week. The structure, not the people, was the problem — too much administrative drag between a supervisor and the coaching that moves performance.

From Prompt Engineering to a Model That Learns

What turned interest into a decision was the shift AmplifAI represented on exactly that quality problem: moving from prompt engineering to a model that learns.

Instead of analysts hand-coding whether an agent said a specific phrase, AmplifAI's quality approach is built to learn what a strong interaction looks like and to assess the harder, softer behaviors — the emotional connection that anchors 80% of the scorecard. For an operation that scores connection over script adherence, that is the difference between a tool that fits the work and a tool the work has to bend around.

The commercial case made it easy to act on. The capability upgrade came at net-neutral cost — replacing the prior prompt-engineering spend rather than adding to it. A major enhancement at a flat price point is the kind of proposal that clears internal approval quickly, and it reframed the quality layer from a maintenance burden into a foundation worth building on.

Making the Data Coachable for Supervisors

The selection also spoke to a problem the team had been circling: the supervisor experience.

What first drew the operation to AmplifAI was not the quality scoring at all — it was how the platform makes data actionable for a supervisor. Rather than asking a frontline leader to be an analyst, AmplifAI's coaching workflow is designed to strip the noise out of all that reporting and surface what to do next. That speaks directly to the ninety-minutes-of-coaching problem: when a supervisor doesn't have to assemble the picture by hand, more of their week can go to the conversation that actually develops an agent.

It also fit the operation's stance on automation, which the leader frames as choiceful. The goal is not to automate the contact center away. It is to automate the low-value, nonsensical cost — the noise that adds nothing — and to protect the human element in the places where it differentiates the brand. In a business where a single warm conversation measurably lifts how often a customer reaches for the card, that line is drawn deliberately.

Why It Was the Right Call

The decision came down to a clean alignment of capability, philosophy, and price. The operation got LLM-based quality suited to the soft-skill behaviors it actually scores, a supervisor-ready view of data that had been drowning in its own reporting, and a path to keep adding best-in-class tools on top of a telephony layer that already works — all at net-neutral cost.

For a high-performing loyalty care operation — one already posting 94–95% satisfaction and running to a 30%+ margin — the bar for a new system is not whether it can rescue a struggling team. It is whether it respects what already works and removes what gets in the way. Trading brittle, literal quality scoring for a model that learns, and a pile of reports for coaching a supervisor can act on, did exactly that.