Is the bot earning its keep?

Deflection economics, AI resolution quality, and the knowledge gaps that cost you escalations — one glassy surface for the automation product owner.

Deflect Live Pulse Signal Cadence

Deflection Funnel AI-reconciled

Last 30 days · stages reconcile

312k

Sessions

100%

→63%

198k

Searched

63% of sessions

→54%

106k

Self-resolved

34% deflected

→40%

42k

Started ticket

13% of sessions

→67%

28k

AI-resolved

67% of tickets

→33%

14k

Agent

human handoff

Topic Bubble Map auto-clustered

size = volume · color = AI res. rate

Order status22k81%

Returns14k63%

Billing9k47%

Password8k92%

Shipping6k78%

Account4k66%

Handled well (≥75%) Watch (55–75%) Struggling (<55%)

AI Resolution Rate

67%

last 30 days · ▲ 2.4pp

matches Fin benchmark

Deflection Rate

34%

good band 20–40%

▲ 1.1pp self-serve / no agent

Resolved by AI

41,280

conversations / mo · ▲ 9% MoM

Cost Saved

$148K

/ mo · blended $3.58/contact

▲ $11K vs last month

AI CSAT & Handoff

30-day rolling

91%

AI CSAT

19%

Handoff rate ▼ 1.2pp

handoff analysis 58% low-confidence · 22% out-of-scope · 20% customer asked

Search Health

self-service quality

86%

success

Search success86% ▲0.8pp

"No results" rate1.4%

Avg results / query4.2

Refined searches11.3%

Projected-Lift Simulator AI

Publish suggested articles3

Resolution lift

+3.1%

Est. saved

$6.2K/mo

Drag to model how many AI-drafted articles you approve to the knowledge base this cycle.

Content-Gap Table Knowledge Gap Engine

top no-result & unanswered queries

Query	Misses	Suggested article	Confidence
"cancel after shipping"	412	How to cancel a shipped order	0.88
"partial refund timing"	318	When partial refunds post	0.84
"change delivery address"	287	Edit a delivery address mid-transit	0.71
"international warranty claim"	214	Filing a warranty claim abroad	0.68
"reactivate cancelled plan"	176	Restoring a cancelled subscription	0.82

Article Performance

by deflection $ contribution

Article	Views	Helpful	Deflected
Track your order	18.2k	94%	$21K
Return policy	12.4k	88%	$14K
Reset your password	10.1k	96%	$12K
Update payment method	7.8k	83%	$9K
Shipping timelines	6.5k	90%	$7K

Field Guide Reading Deflect like an automation owner

How to use this

Start at the funnel, not the rates. The two stages that matter are Started ticket → AI-resolved (67%) and AI-resolved → Agent (33% leak). Every point you move off that last drop is salary you don't pay.
Hunt the warm bubbles first. The bubble map is sized by volume and colored by resolution — a big rose bubble (Billing, 9k @ 47%) is the single richest fix on screen. Cyan bubbles are already winning; leave them alone.
Read color, then size. A small struggling topic is a coaching note; a large struggling topic is a roadmap item. Click any bubble to see handoff reasons and the auto-drafted articles that close it.
Let the AI do the writing. The Knowledge Gap Engine clusters unresolved chats, drafts the missing article + bot answer, and projects the lift — your job is to approve, not author. Use the simulator to decide how many to ship this cycle.
Tie every move to money. Cost Saved and the per-article "Deflected $" turn "the bot is doing great" into a number a CFO trusts. Bring those, not resolution percentages, to the budget meeting.
For your own org: if your deflection rate sits below the 20–40% band, the bottleneck is almost always search, not the bot — check Search Health's "no-result" rate before adding model spend.

Watch the walkthrough

Four AI agents walk this dashboard.

In context — automation benchmarks

Sample feed

Industry median AI resolutionretail & e-commerce cohort

62%you +5

Median deflection rateself-serve, no agent

29%you +5

Blended cost / human contactfully-loaded, this region

$5.90−$2.32

Avg knowledge-base coverageintents with a live article

74%you 81%

AI CSAT vs human CSAT gapcohort average

−4ppyou −1pp

Illustrative — wire to your automation analytics feed (Fin / Ada / Forethought) to benchmark live.