Why containment rate alone is a misleading success metric
Containment rate is the number every voice bot vendor leads with, and it is also the easiest number to inflate. A bot can log a call as "contained" simply because the caller did not get transferred to a human, even if the caller hung up frustrated, got a wrong answer, or called right back ten minutes later to try again. None of that shows up in a raw containment number. If you measure success by containment alone, you can end up rewarding a system that is good at ending calls, not good at resolving them. The fix is not to throw containment out. It is to always pair it with a resolution measure, so a high containment rate that is not backed by actual issue resolution gets caught before it shows up as inflated ROI.
The core KPIs, defined in plain language
Five metrics form the backbone of any honest voice bot evaluation. Containment rate is the percentage of calls the automated system handles start to finish without transferring to a human agent. First-Call Resolution (FCR) is the percentage of issues actually resolved in a single contact, with no follow-up call needed, which is the metric that keeps containment honest. Average Handle Time (AHT) is the total time per interaction, including talk time, hold time, and wrap-up work, and it matters even on calls the bot does not fully contain. Cost-per-contact is total operating cost divided by the number of calls handled, and it is the denominator every ROI calculation eventually runs through. Transfer or escalation rate is the percentage of automated calls that get routed to a human, and it is effectively the inverse of containment, useful for spotting where the bot is struggling with specific call types.
What a good FCR benchmark actually looks like
SQM Group, a contact-center benchmarking research firm, tracks First-Call Resolution across industries and puts the aggregate average at roughly 70%, with a typical range of 50-90% depending on industry and call complexity. Under SQM Group's framework, a 70-79% FCR is considered good performance, and 80%+ is considered world-class, a bar only about 5% of contact centers actually clear. That benchmark matters for voice bot ROI because it gives you a reference point independent of the bot itself: if your automated system reports a 90% containment rate but your FCR (measured by tracking repeat contacts) is sitting at 55%, the gap between those two numbers is where your ROI math is quietly falling apart.
How AI containment compares to traditional IVR
Traditional routing-only IVR systems, the kind that just move callers through a menu tree to the right queue, are generally reported across contact-center industry sources to contain somewhere in the 20-40% range, mostly by handling simple lookups or deflecting to self-service. AI-powered voice systems that can actually converse and resolve issues generally average 40-55% containment in typical deployments, and best-in-class implementations, usually ones with strong integration into backend systems and well-scoped use cases, report containment in the 70-80% range. The gap between those numbers is not really about the AI being smarter in the abstract. It is about how well the deployment is scoped, how deep the system integration goes, and whether the bot is being asked to resolve things it is actually capable of resolving.
How to calculate ROI that holds up under scrutiny
The basic formula is straightforward: ROI equals the cost avoided by automation minus the platform and deployment cost. Cost avoided is containment rate multiplied by call volume multiplied by average human-agent cost per call. Platform cost includes subscription or usage fees, integration and setup work, and the ongoing monitoring and QA labor the system needs to keep running well. The nuance that most ROI spreadsheets skip is quality-adjusted containment: if your containment rate is 60% but a chunk of those "contained" calls generate a callback within 24-48 hours, you are not actually avoiding that cost, you are deferring it. Netting containment against your measured FCR gives you a truer picture of what the bot is actually resolving versus what it is just closing out. It is also worth tracking a savings line most teams miss entirely: AHT often drops even on calls the bot does not fully contain, because the bot triages the issue, collects the account details, and pre-qualifies the request before handing off, which cuts the human agent's time on the call substantially. That secondary savings can be a meaningful share of total ROI, even when it never shows up in a containment number.
Tracking these metrics day to day
The honest way to run this is to check containment, FCR, AHT, cost-per-contact, and transfer rate together, on the same dashboard, every week, not just at renewal time. A bot that looks great on containment and bad on FCR needs a different fix than one that looks bad on both. We built Persistence around this same principle: our customers track containment, FCR-adjusted resolution, AHT, and escalation rate directly in the product, rather than reconstructing them from call logs after the fact, because ROI conversations go a lot better when both sides are looking at the same numbers.
Key takeaways
- Containment rate by itself is a vanity metric — a call that "resolves" but gets a callback an hour later isn't actually resolved, so containment has to be read alongside FCR
- SQM Group research puts industry average First-Call Resolution at roughly 70%, with 70-79% considered good and 80%+ considered world-class (only about 5% of contact centers hit that bar)
- Traditional routing-only IVR containment typically runs 20-40%; AI-powered voice containment generally averages 40-55%, with best-in-class deployments reaching 70-80%
- Real ROI = (containment rate x call volume x avoided human-agent cost) minus platform and deployment cost, then adjusted downward for calls that get contained but come back as repeat contacts
- AHT reduction on transferred calls is a savings line most containment-only ROI math misses entirely
