Finding 1
Most threads were short, but a meaningful tail required more work
The median conversation contained four messages. The 75th percentile contained 10 and the 90th percentile contained 21.
A median alone therefore understates the operational burden. A smaller group of high-touch families or complex scheduling situations can account for a disproportionate share of communication work.
Finding 2
Classifiable replies were usually fast
Across 8,890 adjacent message transitions where Swum could classify one side as the provider and the other as the family, the median interval was approximately 18 minutes. 84.9% occurred within 24 hours.
This is not a customer-support service-level metric. Multiple quick messages within one active conversation can contribute several transitions, and long periods without a reply are not represented unless another message eventually arrived.
Finding 3
Two-way communication appeared in 41.9% of conversations
Swum could identify messages from both a provider and family in 1,271 of 3,035 conversations. Other threads may contain one-way notices, welcome messages or older sender fields that could not be classified.
The result supports separating announcements from conversations in operational reporting. A sent message count alone says little about whether a family needed or received a response.
Finding 4
Messaging volume moved with the operating season
Monthly message volume rose from 360 records in January 2026 to 4,999 in June before easing to 4,284 in July. The growth coincided with more active businesses and summer lesson volume.
The series is therefore useful for staffing and workflow planning, but not for inferring that a feature caused engagement growth.
Conversation depth
| Measure | Messages |
|---|---|
| 25th percentile | 2 |
| Median | 4 |
| 75th percentile | 10 |
| 90th percentile | 21 |
2026 message volume
| Month | Message records |
|---|---|
| January | 360 |
| February | 611 |
| March | 1,650 |
| April | 3,343 |
| May | 3,893 |
| June | 4,999 |
| July | 4,284 |
Methodology
How the benchmark was produced
Swum selected message metadata only. Message content, sender names and personal information were not read or emitted.
- Messages were grouped by their parent conversation path.
- Sender classification matched the stored sender identifier to the stored coach or parent identifier when possible.
- Reply intervals used adjacent classifiable messages from opposite parties and excluded intervals longer than 30 days.
- 12,078 older or differently structured messages remained unclassified and were excluded from reply-direction measures.
Operator response
What to do with this finding
- Measure announcements, one-way notices and two-way support conversations separately.
- Review the high-message tail to find workflows that repeatedly create questions.
- Track the first provider reply per family request before defining a formal response-time goal.
- Preserve structured sender roles so future communication reporting has complete coverage.
Limitations
What this data cannot establish
- Message content was intentionally excluded, so topic and sentiment were not analyzed.
- The reply interval is between messages, not necessarily between a customer request and a complete resolution.
- Older sender fields reduced classification coverage.