NPS measures long-term stated loyalty, not what happens in each conversational interaction — and in 2024 it fell from the 2nd to the 8th most-used metric (CMSWire). For channels like WhatsApp and chatbots, six metrics capture the real experience: Customer Effort Score (1.8× more predictive of loyalty than CSAT, per HBR), transactional CSAT (better when inferred), Task Completion Rate (with accuracy verification, to avoid hallucinations), Escalation Rate and handoff quality, First Contact Resolution at the need level, and Value Enhancement Score (the most correlated with retention, per Gartner).
A conversational channel can show a stable NPS while it doubles repeat contacts , drives up escalation costs and silently destroys the customer experience. That's the problem with measuring the wrong thing. For years the Net Promoter Score was the go-to metric in banking: easy to report to the board, easy to put in a deck. But according to the Forrester 2024 US Customer Experience Index , CX quality in banking fell for the third year in a row, to historic lows. In Latin America, digital channels accumulate friction without traditional indicators catching it in time. The problem isn't that NPS is a bad metric — it's that it was designed to measure something else. Why wasn't NPS designed for conversational channels? NPS asks a single question: how likely are you to recommend this company to someone? It's a long-term loyalty metric, not an interaction-experience one. In conversational channels, what matters happens in seconds: did the customer resolve what they needed? Was it hard? Did they have to repeat their information when handed to an agent? Gartner flagged this back in 2021: NPS consistently fails to provide actionable insight for customer service leaders. And market data confirms the direction — according to CMSWire, NPS went from the second most-used metric in 2023 to the eighth in 2024. NPS persists where better alternatives are missing, not where measurement is most effective. Which metrics actually capture the conversational experience? Customer Effort Score (CES): the strongest predictor of loyalty CES measures how much effort the customer had to make to resolve their need. A single post-interaction question: how easy was it to handle your request today? The research by Dixon, Freeman and Toman in Harvard Business Review , based on more than 75,000 interactions, establishes that CES is 1.8 times more predictive of loyalty than CSAT and 2 times more than NPS . And 96% of customers who go through a high-effort interaction become more disloyal. In banking, where reactivation is expensive, that hits the business directly. Transactional CSAT: the most immediate signal CSAT measures point-in-time satisfaction after a specific interaction. The problem in asynchronous channels like WhatsApp is that response rates rarely exceed 15-20%, and those who respond aren't representative: that biases the indicator systematically. A more robust approach is to build an inferred CSAT by combining behavioral signals: how many messages did it take to get an answer? Did the customer repeat the same question? Did they contact again within 48 hours? The inferred version covers 100% of interactions, not 15%. Task Completion Rate (TCR): did they resolve it or not? The most honest metric: did the customer complete the task they came to do? It's measured behaviorally, without a survey. There's a critical risk in channels with generative AI: the bot can confidently close the flow after giving incorrect information. TCR records “success,” but the customer got a wrong answer about their balance, their product or their contract terms. A hallucination about interest rates or the terms of a loan isn't just a CX problem: it's a regulatory liability. TCR without accuracy verification is the indicator that most often creates a false sense of control. Escalation Rate and handoff quality The most documented problem isn't the escalation rate itself, but what happens during the handoff. The CFPB flagged as a frequent complaint that many customers get stuck in loops of repetitive replies with no real path to a human agent. A well-designed escalation transfers, at minimum: the reason for contact, a summary of the conversation and the customer's authentication status. When that doesn't happen, CES collapses no matter how well the bot worked before. The indicator that matters isn't the rate — it's the CES of the interactions that escalated. First Contact Resolution (FCR): at the need level, not the channel level Did the customer resolve their need without contacting the bank again within the next 48 hours? In a channel that promises instant resolution, a repeat contact is a clearer failure signal than anywhere else. The trap: if each channel measures its FCR in isolation, all of them can post good numbers while the customer goes through three contacts to solve a single problem. Useful FCR is measured at the level of the customer's need, not the channel . Value Enhancement Score (VES): the metric Gartner puts on the table VES measures whether, after an interaction, the customer feels they can do more with their product and are more confident in their decision to use it. Gartner identifies it as the indicator most correlated with retention and wallet share. An example: a customer contacts the bank because they were charged $150 they weren't expecting. A TCR-oriented bot refunds the money and closes the conversation. A VES-oriented bot refunds the money, explains why the charge happened and offers to set up an alert so it doesn't happen again. The first puts out the fire; the second puts out the fire and teaches the customer how not to get burned again. That difference is exactly what NPS will never capture. The dashboard that makes sense in 2026 NPS can keep its place as a long-term signal, but using it to measure what's happening in a conversational channel is like measuring a room's temperature with a thermometer that returns results three weeks later. The banks winning at conversational CX aren't the ones with the prettiest channel or the most powerful bot: they're the ones that know exactly what's happening in each interaction and have the processes to fix it without waiting for the next quarterly NPS report. The question is no longer whether your bank has conversational AI, but whether it knows exactly what that AI is doing to its customers' experience. At Delto we instrument and operate conversational channels that measure what matters , not just what's easy to measure. If you want an outside read on how your bank is measuring, book 30 minutes with our team .
Why isn't NPS enough for conversational channels? Because it measures long-term stated loyalty with a single question (“would you recommend us?”), not the experience of each interaction. On WhatsApp or chatbots, what matters is whether the customer resolved their issue, how much effort it took, and whether they had to repeat their information — and NPS captures none of that.
What is the Customer Effort Score (CES)? It measures how much effort the customer had to make to resolve their need. According to Harvard Business Review, it is 1.8 times more predictive of loyalty than CSAT and 2 times more than NPS.
Which metrics should a bank measure in its conversational channels? Customer Effort Score, transactional CSAT (ideally inferred), Task Completion Rate with accuracy verification, Escalation Rate and handoff quality, First Contact Resolution at the need level, and Value Enhancement Score.
What is the risk of Task Completion Rate with generative AI? The bot can close the flow as a “success” after giving incorrect information (a hallucination about rates or terms). Without accuracy verification, TCR creates a false sense of control and can become a regulatory liability.