See your AI customer service savings in 30 seconds.

Most calculators and vendor pitches treat deflection rate as a single number. In practice it has three layers, and only one of them is the one you actually want to maximize.

There’s intent recognition (the AI understood what the customer wanted), containment (the customer didn’t escalate to a human within that conversation), and resolution (the customer’s actual problem got solved and they didn’t come back with the same issue 24 hours later through a different channel).

Vendors mostly report containment. It’s the easiest to measure and the most flattering. A customer who clicks “talk to a human” right after the AI answered is counted as a successful deflection in most platforms, even though they’re frustrated and your CSAT just took a hit.

The deflection that matters is resolution. A 30 percent resolution rate beats a 50 percent containment rate every time. The calculator above models the labor savings from human-handled tickets that don’t happen, which maps closest to resolution. If you’ve seen vendor numbers that look impossibly high, like 60 to 70 percent on day one, they’re almost certainly containment with a generous definition of success.

Each input changes the result differently, and a few of them carry assumptions worth surfacing.

Monthly ticket volume is straightforward, but pull it from the last 90 days, not the last month. CX has seasonality (holidays, launches, supply issues), and a single-month snapshot can mislead by 20 to 30 percent in either direction.

Loaded agent cost should include benefits, overhead, and tooling. The bare hourly rate misses 40 to 60 percent of the actual cost. For US in-house, $25 to $35 per hour loaded is typical. US BPO runs $15 to $22. Offshore is $7 to $12. If your team uses a mix, weight the average.

Average handle time is the most-fudged number in CX. The reported AHT in your help desk usually captures the human-touch portion only. It skips queue time, internal lookups, and follow-up tasks after the conversation. The true AHT is typically 1.3 to 1.5x what your tool reports. Use the higher number if you want a defensible result.

Current deflection rate is what your existing chatbot, FAQ, or canned macros already resolve. If you’ve never measured, leave it at zero. If you’ve measured but it’s low (1 to 5 percent), that’s probably accurate. Macros and FAQ rarely deflect more than 10 percent on their own.

Target deflection rate is the planning assumption. The default of 30 percent reflects a realistic 90-day program floor. The slider goes to 80 percent because fully autonomous AI agents (Siena, Decagon, Sierra) can reach that on a clean foundation. Most programs don’t have a clean foundation on day one.

AI cost per resolution is usually the most negotiated line item. Published prices are starting points; volume contracts often land 20 to 40 percent below them. Use the Custom option once you have a real quote.

After advising or running a few dozen of these, four failure modes account for most of the gap between projected and actual deflection.

Knowledge base content quality. The single biggest predictor of deflection success is whether your KB articles are written for AI consumption (clean, declarative, scoped to one issue per article) versus human consumption (narrative, full of context). AI agents trained on the latter underperform by 15 to 30 percentage points. Most teams underinvest in KB rewriting and overinvest in vendor selection.

Intent overlap. When two intents look similar to a model (think “where is my order” versus “I need to change my shipping address”), the AI confidently answers the wrong one and tells the customer their order is on its way when they’re trying to redirect it. Cleaning intent boundaries before launch is dull, hard work, and the difference between a 30 percent and a 45 percent program.

False-positive resolution. The AI answers, the customer doesn’t escalate (containment), but didn’t actually solve their problem. They contact you again 24 hours later via a different channel, post a 1-star review, or churn quietly. False-positives don’t show up in the AI vendor’s dashboard but they show up in CSAT, retention, and your inbox.

Over-aggressive deflection on emotional intents. Refunds, complaints, account closures, and damaged-item reports work best with empathetic humans. Routing them to AI to save labor cost burns more goodwill than it saves money. The realistic scenario in this calculator assumes you’ll exclude these intents from automation.

Vendor marketing on AI customer service has gotten louder through 2025 and 2026, and the claimed numbers have inflated accordingly. Here’s what the realistic data looks like in DTC implementations.

Expected resolution rates in a well-implemented program, by ticket type:

• Order status, tracking, shipping inquiries: 70 to 85 percent
• Subscription management (pause, skip, swap, cancel): 60 to 80 percent
• Returns and exchanges (initiation, not resolution): 50 to 70 percent
• Product Q&A on factual attributes: 40 to 60 percent
• Account management: 40 to 60 percent
• Technical troubleshooting: 15 to 30 percent
• Complaints, damaged items, refund requests: 5 to 15 percent (and that’s intentional)

Blended across a typical DTC ticket mix, these average to 30 to 45 percent deflection on full volume, which is what the Realistic scenario reflects. Programs that hit 50 percent or more on blended volume are running on a clean foundation, with autonomous AI agents, and have invested 200-plus hours in KB cleanup before the AI ever talks to a customer.

The 70-percent-plus numbers in vendor case studies are usually one of four things: containment rather than resolution, a narrow ticket subset (often order-status only), a specific category like FAQ deflection, or marketing. If you’re hearing claims above 60 percent on blended volume, ask the vendor for the methodology and which intent types are excluded. If they can’t answer cleanly, the number isn’t real.

Time to first results: typically week 6 to 8 for measurable lift, day 90 for steady-state numbers. Programs that promise day-one results either start with very low expectations or count what wasn’t really deflected.

The honest read: AI customer service is the highest-leverage CX investment for DTC brands in 2026, and also the easiest to over-promise on. The calculator’s three scenarios are designed to keep you anchored on what is realistic, not what is pitched.