Insights

The Insights page is your "where is the AI failing to deflect?" view. It shows how often the AI handed a question over to your team instead of answering it, what reasons it gave, and which questions keep coming back.

Use it to find gaps in your FAQ - every recurring escalation reason is a hint that you're missing an entry, or that an existing entry needs improvement.

Where to find it

In your SCNX dashboard, open AI FAQ → Insights. It's separate from the broader Support Bot Analytics page because the questions you ask of this data are different.

Time range

A dropdown at the top lets you switch between Last 7 days, Last 30 days, or Last 90 days. Every chart on the page reflects the range you pick.

What the page shows

Top-line numbers

At the top of the page you'll see four numbers at a glance:

• Total answers - how many AI answers the bot produced in the selected window, across both channel auto-answer and the pre-ticket gatekeeper.
• Escalations offered - of those answers, how many included an Open a ticket button. This happens whenever the AI judged the question couldn't be fully answered from your FAQ.
• Escalation rate - the percentage of answers that ended with an escalation offer. Low rates mean the AI is deflecting most questions on its own; high rates mean a meaningful share of incoming questions still need a human, usually because your FAQ doesn't cover them yet.
• Heuristic-fired escalations - the percentage of escalations that came from the safety net instead of the AI's own decision. See Where escalations come from below.

Where escalations come from

Every escalation is tagged with how it happened. There are two sources:

• Clean hand-off (labelled "Tool call" on the page) - the AI clearly decided "I can't answer this from the FAQ" and explicitly asked the bot to offer a ticket, along with a reason. The reason is reliable.
• Safety-net catch (labelled "Safety-net heuristic" on the page) - the AI didn't ask for an escalation outright, but the wording of its answer looked like a non-answer (e.g. it said "I don't know" in prose). The bot adds the escalation button anyway just in case the member needs it.

A small share of safety-net catches is normal - that's what the safety net is for. A very high share (well over half) suggests the AI is hedging in prose instead of clearly handing off, which is worth investigating. Look at the recent escalations list to see if the safety-net entries look like real questions that should have had a clear answer.

Top escalation reasons

A ranked list of the most common reasons the AI gave when it explicitly escalated. Treat this as a to-do list for your FAQ. Each recurring reason is a question the AI doesn't have a good answer for yet - add or improve an FAQ entry covering it and the reason should drop off the list next time you check.

Safety-net escalations don't have specific reasons attached, so they're not shown here. If your safety-net share is high, the recent escalations list is the place to dig.

Recent escalations

A list of the last 20 escalations, with the question the member asked, the reason (if any), and when it happened. Useful for spot-checking individual cases - especially when a top reason looks suspicious or you just want to see how the AI is wording its non-answers right now.

How to use the data

Three concrete things this page is good for. You don't have to do all three - pick whichever matches a question you're trying to answer.

Spot FAQ gaps

The single most useful workflow.

1. Set the window to Last 30 days.
2. Open Top escalation reasons and look for any reason that shows up 3 or more times.
3. For each one, either add a new FAQ entry covering that question, or expand an existing entry so its wording covers what members are actually typing.
4. Check back in two weeks - the reasons you tackled should have dropped or disappeared.

Check that the AI is staying reliable

If you want to make sure the AI isn't regressing or going off-script:

1. Watch the Heuristic-fired escalations number - the percentage of escalations caught by the safety net rather than asked for cleanly by the AI.
2. If it stays above ~50%, sample the Recent escalations table. Are the safety-net entries questions the AI should have been able to answer? If so, the matching FAQ entries probably need rewriting.
3. If a single topic keeps showing up, look at the entry that should cover it - the wording may not match what members type.

Track ticket deflection over time

To see whether the AI is reducing your team's load month over month:

1. Pull up Insights on the 7-day window and note the Escalation rate.
2. Switch to 30 days and 90 days and note the same number.
3. A falling rate means more questions are being resolved without human help; a rising rate means coverage hasn't kept up with the kinds of questions coming in.

Suggested workflow

A loose weekly routine that combines the three workflows above:

1. Open Insights with a 30-day window once a week.
2. Skim the top escalation reasons for any reason that shows up 3+ times.
3. For each recurring reason, either add a new FAQ entry for it or improve the existing one so the AI can answer that wording.
4. Glance at the safety-net catches number. If it's climbing without obvious cause, sample the recent escalations to see whether the AI is regressing on a particular topic.
5. Check back in two weeks - the reasons that were top last time should be smaller or gone.

Insights is for FAQ gaps, not staff performance

For staff metrics, response times, and ticket trends, head to the Support Bot Analytics page. Insights is focused on AI answer quality and FAQ coverage.

Enterprise

Insights tops out at a 90-day window. Longer retention, custom analytics, and exportable reports are available on enterprise plans - see Enterprise & Docs Sync or contact sales.