The Deflection Equation: Volume x Cost x Rate

Somewhere right now a VP of support is sitting in a budget review, defending a number nobody in the room understands. The number is big. The CFO wants it smaller. The VP says something about headcount and seasonality and ticket spikes, and everyone nods, and nothing changes. This happens every quarter. It happens because support has spent twenty years describing itself in feelings instead of dollars.

The fix fits on a napkin.

Three terms, one number

Support cost isn't a mystery. It's a product. Ticket volume times cost-per-ticket times the share you can deflect, plus the efficiency you claw back on whatever's left. Four inputs. One bottom line. Any operator with a pen can run it before the coffee gets cold.

Start with the middle term, because it's the one people fight about. The global average cost to resolve a support ticket runs $6 to $7 across all channels, and it stretches from around $2 on the simple end to $60 and up where the industry is complex or heavily regulated (LiveChatAI). That spread matters. A fintech support contact and a "where's my password" reset are not the same animal, and pretending they cost the same is how budgets get built on fiction. Pick the number that's actually true for your queue.

Now the lever. Conversational AI paired with decent self-service deflects 25 to 45 percent of tickets and returns 2x to 5x ROI inside the first year (LiveChatAI). Deflection isn't a fancy word for "ignored." It means the question got answered without a human touching it. The article that finally surfaces at the right moment. The bot that handles the refund status instead of routing it to a person who would have read a script anyway.

And the quiet fourth term, the one most decks forget. On the tickets that still reach a human, the same tooling produces roughly a 30 percent efficiency gain (LiveChatAI). The agent opens the ticket with the context already pulled, the history already summarized, the suggested reply already drafted. They're not starting cold. That 30 percent doesn't show up in the deflection headline, but it's real money, and it compounds on the hardest tickets, the ones that survived precisely because they were worth a person's time.

Walk the napkin

Here's the worked case, and it's a clean one. A mid-sized SaaS firm running 200,000 tickets a year. Annual support spend: $900,000. That pencils out to $4.50 a ticket, which tells you they're already reasonably efficient, no low-hanging fruit, no obvious mess.

They deploy conversational AI and self-service. Deflection lands at 25 percent, the conservative end of the range. Volume drops to 150,000 tickets. The 50,000 that vanished were the repetitive ones, the questions a good help center should have caught years ago. Annual spend falls to $572,500 (LiveChatAI).

That's a 36 percent cut. Read that twice, because the arithmetic has a kicker buried in it. Deflection was 25 percent. The savings were 36. The gap is the fourth term doing its work: the surviving tickets got cheaper to handle, so the firm saved more on dollars than it shed in volume. Deflection knocks out the easy ones. Efficiency discounts the rest. Together they bend the line harder than either does alone.

Run it on your own queue. Take last year's support spend, divide by ticket count, and there's your real cost-per-ticket, no benchmark required. Multiply your volume by a 25 percent deflection rate, the floor of the published range, not the ceiling. Apply your true per-ticket cost to what's deflected. Then shave 30 percent off the handling cost of what remains. The number that comes out the other side is the conversation to have in the budget review. Not a vibe. A figure with four traceable inputs anyone can audit.

Where the napkin lies

Honesty check, because every equation flatters its author. Deflection rate is the term people inflate, and a bad deployment deflects nothing, it just annoys customers into a second contact, which is worse than no bot at all. The cost-per-ticket figure assumes a steady mix, and complex tickets cost multiples of simple ones, so blending them hides where the money actually goes. And the efficiency gain only lands if agents trust the tooling enough to use it instead of routing around it.

None of that breaks the equation. It just means the inputs have to be honest. The napkin doesn't make the case for you. It makes the case checkable, which is the part the VP in the budget review never had.

That's the whole trick. Support stopped being a feeling the moment someone wrote it as multiplication.

Sources: LiveChatAI (global average support cost of $6-7 per ticket, ranging $2 to $60+ by industry complexity and regulation; 25-45% deflection and 2x-5x first-year ROI from conversational AI plus self-service; ~30% efficiency gain on tickets that still reach humans; the worked SaaS example cutting annual spend from $900,000 to $572,500, a 36% cut, as volume fell from 200,000 to 150,000 tickets at 25% deflection).