CX Frontline Performance Dashboard

Handle Time is phone + chat only (Assembled reports no handle time for email). For Previous Week and MTD the chat component is inflated by a known Assembled mapping defect introduced with the 2026-08-03 Sunshine Conversations migration — see the note under By Channel below. Service level for Previous Week is provisional: Assembled restates recent CCR email service_level downward for roughly 8 days after the fact, so this week’s 94.4% will likely settle lower. Recomputing the prior build’s MTD window on today’s data moved email SLA 93.7% → 79.9% and overall 91.9% → 86.1%. Do not read Previous Week vs. the prior build as an improvement.

Channels here follow Assembled's ticket_channel mapping (phone / chat / email), which is independent of the Phone/Chat/Async bucketing on the Customer Sentiment tab. Assembled routes Sunshine Conversations into email. This dashboard follows Assembled’s mapping on every tab (ratified 2026-08-18) — see the Customer Sentiment tab’s methodology note for the full rule and the one place it diverges from Zendesk. Separate open defect, from 2026-08-03: the Sunshine migration moved chat solves into email but left the corresponding engagement time in chat — through the week of Aug 9 chat occupied time per day stayed roughly flat while chat solves per day halved again (CCR 63.5 → 32.0, CRA 32.3 → 15.6 vs. July), with the corresponding solves landing in email. Chat handle time for Previous Week and MTD (CCR 5,077 s, CRA 10,421 s) is therefore overstated by roughly 2x and is not an efficiency regression; it also inflates the combined “Handle Time (phone + chat)” figure above. Volume and service level are unaffected. This is a defect in the Assembled↔Zendesk channel mapping, not a reporting choice — the same conversation has its solves counted in email and its engagement time counted in chat. It needs an upstream fix; until then treat chat AHT for any period overlapping 2026-08-03 onward as unusable.

Total inbound volume divided by active subscribers in the period — volume from BigQuery (futurhealth.dbt_base.base_zendesk__tickets / base_zendesk__talk_calls), active subscribers from the FH+ subscription retention model (dbt_agg.agg_fh_plus_subscription_retention, weekly cohort grain). Frontline = CCR + CRA + Abandoned Phone volume. Total = all queues. Denominator is the average of weekly active-subscriber counts in the period. Because the retention model needs elapsed time to fully populate a given week, weeks within ~12 days of the cut-off are held at the last reliable week's value (currently 2026-08-03) rather than using their own undercounted figure — this affects the trailing week of the Last 6/12 Weeks trend. The cutoff is re-derived each build from a log-linear trend fit plus a snapshot-stability test; fill-in completes about 7–9 days after a week ends. Methodology correction in this build: the prior build's guard was over-conservative and held 2026-07-27 and 2026-08-03 flat at a stale 1,577. Both weeks are now settled and use their own values, so those two trend points read higher than previously published (07-27 1.21 → 1.40, 08-03 1.07 → 1.18) and the apparent decline across the tail is no longer there. That is a correction, not new demand. The final trend point (2026-08-10) understates on two counts — 6 of 7 days of volume and a held-flat denominator, both pushing down; grossed up it is roughly 1.46 frontline / 4.24 total, not the 1.25 / 3.63 printed. Multi-week periods print cumulative period volume over average subscribers and are not weekly-normalised, so longer windows print larger by design: QTD covers 46 days against Previous Quarter's 91. This methodology also requires data from on/after 2026-06-01 (new payment system cutover); it should not be applied to periods starting earlier than that.