What counts as a resolved ticket?
A ticket is only resolved when it reaches a defined support outcome and stays closed without follow-up contacts during your observation window. Spell out that threshold per ticket category: an order-tracking reply is far simpler to verify than a processed refund or a corrected billing record.
Keep your internal quality bar separate from vendor billing events. Intercom’s Fin AI Agent documentation, for example, distinguishes confirmed resolutions from assumed resolutions where a user stops replying. That distinction governs how the vendor charges per outcome, not whether the customer’s issue was genuinely fixed.
Which metrics show whether a chatbot creates value?
Look at three numbers together: verified resolution rate, agent minutes spent picking up chats, and repeat-contact rate across email, chat, and phone within a fixed follow-up window.
Slice performance by identical ticket categories, otherwise an unexpected spike in simple password resets can make a mediocre bot look dramatically better month over month.
| Measure | What it reveals | Decision use |
|---|---|---|
| Verified resolutions | Tickets meeting your written closure criteria | Serve as the denominator for unit cost |
| Human handling minutes | Time spent on QA, handoffs, and manual fixes | Subtract directly from gross time saved |
| Recontacts | Customers returning through any channel inside your window | Correct inflated chatbot resolution counts |
| Quality and complaints | Policy hallucinations or frustrated customer transcripts | Audit and tighten guardrails before expanding |
How do you calculate cost per resolved ticket?
Divide all support expenses tied to the chatbot lane over a period by the count of verified resolutions in that same period. Capture software and outcome fees, knowledge-base maintenance hours, and agent handling on escalated chats—explicitly noting whether setup costs are amortized or tracked separately.
Comparing a vendor’s quoted price per resolved chat against your support team’s fully loaded cost per ticket mixes apples and oranges unless bot maintenance and escalations are already baked in.
What does a chatbot ROI calculation look like?
Suppose 1,000 eligible monthly tickets take 10 minutes each manually. In a bounded pilot scenario, 400 verified chatbot resolutions remove 4,000 minutes of agent work, while chat QA, bot updates, and extra repeat contacts absorb 1,000 minutes.
Net time saved comes to 50 hours a month. At a $30 loaded hourly rate and 60% capacity utilization, monthly value hits 50 × $30 × 0.60 = $900. Subtracting $500 in monthly platform and upkeep costs leaves $400 net monthly value, paying back a $2,400 rollout in six months on a capacity basis.
With the remaining tickets staying manual, modeled first-year ROI works out to (12 × $400 − $2,400) ÷ (12 × $500 + $2,400) = 28.6%.
When does chatbot ROI overstate the benefit?
Spreadsheet projections break down fast when abandoned chat windows get booked as wins, complex failures vanish from test averages, or every freed minute gets treated as direct payroll reduction. Never double-count the same saved hour as both a cut in support wages and extra upsell capacity.
When headcount stays constant, frame the gain strictly as absorbed ticket growth or shorter queues. Put dollar values on CSAT shifts only when cohort data proves lower churn—and keep pilots long enough to catch delayed complaints and seasonal spikes.
What should you check before increasing chatbot coverage?
Hold off on opening new ticket queues until verified resolution accuracy holds steady and support managers have a concrete plan for redeploying freed shift hours.
- Blind-audit a sample of both assumed and confirmed resolutions against identical closure rules.
- Track escalations and repeat contacts across phone and email queues outside the chat widget.
- Stress-test the business case at lower resolution and utilization assumptions.
- Name clear owners for help-center freshness, complaint triage, and human handoff paths.
Frequently asked questions
Is chatbot deflection the same as resolution?
No. Deflection merely measures conversations that did not reach a live agent; resolution confirms the customer’s issue was actually solved. Track both metrics separately when comparing vendors.
Should I include the support team’s salaries?
Yes—always price freed time using loaded hourly labor rates, while separating usable team capacity from actual cash payroll changes when headcount stays constant.
Can a chatbot be worthwhile without reducing headcount?
Yes—freed agent capacity helps absorb volume spikes, clear backlogs, or handle VIP accounts without extra hiring or overtime. State that operational target upfront and track it directly.