Back-and-forth depth — high = unclear answers or complex issues.
At a glance
Avg Replies per Conversation is a Conversation Intelligence metric tracked from Intercom data. It measures how many message parts — agent and customer replies — a typical conversation takes before it closes, averaged over the last 30 days. For the support lead at Blitz it is a proxy for effort and clarity: low and steady means issues resolve in a couple of exchanges; rising means agents are going back and forth more, which points to unclear first answers, weak macros, or genuinely more complex problems. It is the efficiency twin of resolution time.
| What it counts | The mean number of conversation parts (replies/notes that represent message exchanges) per conversation, across conversations active in the trailing 30 days. Derived from the conversation_parts on each conversation in the Intercom conversations model. |
| Sample type | Backend API data from Intercom (conversations with their conversation_parts), refreshed on the standard data refresh. |
| Why it matters | Every extra reply is agent time and customer wait. A rising average raises handling cost and erodes CSAT even when first-response time looks fine — customers judge the whole journey, not just the first reply. A low average usually means clear answers and good self-serve; a climbing one is an early signal that something is making conversations harder. |
| Reading the value | Read the level against your own baseline and watch direction. Two to four replies is common for retail support. A drift upward without a matching rise in issue complexity points to first-reply quality — answers that do not resolve, so customers come back. Read it next to Median Resolution Time: more replies usually means slower resolution. |
| Currency | decimal |
| Time window | 30D |
| Alert trigger | — |
| Sentiment key | null |
| Roles | owner, operations |
Calculation
Calculated automatically from your Intercom data. For each conversation active in the trailing 30 days, Vortex IQ counts its message parts — the back-and-forth replies recorded inconversation_parts — and averages that count across all in-scope conversations. Internal notes and purely system events are excluded where they can be distinguished, so the figure reflects genuine customer/agent exchanges. The result is a single decimal: average replies per conversation. See the worked example below for a typical reading.
Worked example
A representative reading of Avg Replies per Conversation for Blitz on Intercom. The baseline reads 2.8 — a question, a clear answer, a thank-you. After a returns-policy change, the average climbs to 4.6 over two weeks. Reading it next to Median Resolution Time (also up) and Top Topics (Tags) (a swellingreturns tag), the support lead sees the pattern: the new policy is being explained inconsistently, so customers keep replying with follow-up questions. The fix is not more staff — it is a single, clear saved reply and an updated help article so the first answer resolves the question. Within a week the average falls back toward 3.0 and resolution time recovers. The founder treats the sustained drop as proof the content fix worked. For deeper investigation, use Vortex Mind to see which topics carry the longest threads; for natural-language exploration, ask Ask Viq “which tags have the most replies per conversation?”.
Sibling cards merchants should reference together
| Card | Why merchants reach for it |
|---|---|
ic_median_resolution | More replies usually means slower resolution — read the two together. |
ic_top_tags | Tells you which topics carry the longest back-and-forths. |
ic_volume_by_channel | Heavier channels (email) tend to run longer threads than chat. |
ic_csat | Confirms whether more replies are hurting the customer’s experience. |
ic_reopen_rate | A reply average that hides reopens points at first-answer quality. |
Reconciling against the vendor’s own dashboard
Where to look in Intercom’s own dashboard: Intercom does not expose “average replies per conversation” as a headline metric, but Reports → Conversations offers related effort metrics (replies sent, conversations with replies) that you can divide to approximate it over a matching 30-day range. The closest concept is Intercom’s replies-per-conversation in the conversation-effort reporting; expect the shape to track this card even if the exact mean differs. Why the Vortex IQ value may legitimately differ:| Reason | Direction | What to do |
|---|---|---|
| What counts as a “reply”. Vortex IQ excludes internal notes and system parts; Intercom’s reply counts may include or exclude these differently. | Variable | Confirm whether notes are in scope on both sides. |
| Conversation scope. Vortex IQ averages over conversations active in the window; a Reports figure may use only conversations created or closed in the range. | Variable | Match the scoping (active vs created vs closed). |
| Bot replies. Automated bot/Series replies may inflate the count if not filtered. | VortexIQ lower if filtered | Match the bot-inclusion setting to your profile. |