diff --git a/app/javascript/dashboard/i18n/locale/en/integrations.json b/app/javascript/dashboard/i18n/locale/en/integrations.json
index a39165268..a6f3ef2f2 100644
--- a/app/javascript/dashboard/i18n/locale/en/integrations.json
+++ b/app/javascript/dashboard/i18n/locale/en/integrations.json
@@ -427,8 +427,8 @@
           "HINT": "Share of handled conversations escalated to a human agent."
         },
         "HOURS_SAVED": {
-          "LABEL": "Hours saved",
-          "HINT": "Estimate: Captain replies times the team's average response time. Directional, not measured labor."
+          "LABEL": "Time saved",
+          "HINT": "Estimate: Captain replies times ~2 minutes of assumed agent effort per reply. Directional, not measured labor."
         },
         "REOPEN": {
           "LABEL": "Reopen rate",
diff --git a/app/javascript/dashboard/routes/dashboard/captain/assistants/overview/Index.vue b/app/javascript/dashboard/routes/dashboard/captain/assistants/overview/Index.vue
index 687e97286..b9cc6ae35 100644
--- a/app/javascript/dashboard/routes/dashboard/captain/assistants/overview/Index.vue
+++ b/app/javascript/dashboard/routes/dashboard/captain/assistants/overview/Index.vue
@@ -49,6 +49,11 @@ const resolveTrendGood = (trendValue, direction) => {
 // :point, and a plain number for :absolute counts like conversation depth.
 const TREND_SUFFIX = { percent: '%', point: ' pts', absolute: '' };
 
+// Hours-saved is reported in hours, but large values read better as days. Past
+// 100h we switch the unit so the card stays legible.
+const formatDuration = hours =>
+  hours >= 100 ? `${Math.round(hours / 24)}d` : `${hours}h`;
+
 const metricFor = (statKey, formatValue, direction, trendKind = 'percent') => {
   const data = stats.value?.[statKey];
   if (!data) return { value: '—', trend: '', trendGood: null };
@@ -84,7 +89,7 @@ const metrics = computed(() => [
     key: 'hoursSaved',
     label: t('CAPTAIN.OVERVIEW.METRICS.HOURS_SAVED.LABEL'),
     hint: t('CAPTAIN.OVERVIEW.METRICS.HOURS_SAVED.HINT'),
-    ...metricFor('hours_saved', v => `${v}h`, 'up'),
+    ...metricFor('hours_saved', formatDuration, 'up'),
   },
   {
     key: 'reopen',
@@ -121,7 +126,7 @@ const metrics = computed(() => [
       <CaptainPaywall />
     </template>
     <template #body>
-      <div class="flex flex-col gap-6">
+      <div class="flex flex-col gap-6 pb-8">
         <InboxBanner />
 
         <CoverageBanner :knowledge="stats?.knowledge" />
diff --git a/enterprise/app/builders/captain/assistant_stats_builder.rb b/enterprise/app/builders/captain/assistant_stats_builder.rb
index 7198c39e1..d162406ad 100644
--- a/enterprise/app/builders/captain/assistant_stats_builder.rb
+++ b/enterprise/app/builders/captain/assistant_stats_builder.rb
@@ -3,13 +3,18 @@
 # window, plus a derived trend.
 #
 # Queries are batched to cut round trips: the message-derived counts (handled,
-# public replies, depth) and the average reply time are each computed for both
-# windows in a single scan via conditional FILTER aggregation.
+# public replies, depth) are computed for both windows in a single scan via
+# conditional FILTER aggregation.
 class Captain::AssistantStatsBuilder
   RESOLVED_EVENT_NAMES = %w[conversation_captain_inference_resolved conversation_bot_resolved].freeze
   HANDOFF_EVENT_NAMES = %w[conversation_captain_inference_handoff conversation_bot_handoff].freeze
   BOT_RESOLVED_EVENT_NAME = 'conversation_bot_resolved'.freeze
 
+  # Assumed agent effort displaced by each public assistant reply. Reporting data
+  # only captures reply latency (customer wait time), not handling effort, so hours
+  # saved is a count-times-assumed-effort estimate rather than a measured duration.
+  SECONDS_SAVED_PER_REPLY = 2.minutes.to_i
+
   attr_reader :assistant, :account
 
   delegate :range, :period, to: :window
@@ -26,9 +31,8 @@ class Captain::AssistantStatsBuilder
 
   def metrics
     messages = message_window_metrics
-    reply_times = avg_reply_times
-    current = window_metrics(current_range, messages[:current], reply_times[:current])
-    previous = window_metrics(previous_range, messages[:previous], reply_times[:previous])
+    current = window_metrics(current_range, messages[:current])
+    previous = window_metrics(previous_range, messages[:previous])
 
     build_metrics(current, previous)
   end
@@ -57,8 +61,8 @@ class Captain::AssistantStatsBuilder
     }
   end
 
-  # Combines the per-window message counts and reply time with the reporting-event metrics for one window.
-  def window_metrics(range, message_counts, avg_reply)
+  # Combines the per-window message counts with the reporting-event metrics for one window.
+  def window_metrics(range, message_counts)
     handled = message_counts[:handled]
     public_count = message_counts[:public_count]
     depth_conversations = message_counts[:depth_conversations]
@@ -68,7 +72,7 @@ class Captain::AssistantStatsBuilder
       handled: handled,
       auto_resolution: rate(resolution[:resolved], handled),
       handoff: rate(resolution[:handoff], handled),
-      hours_saved: (public_count * avg_reply / 3600.0).round,
+      hours_saved: (public_count * SECONDS_SAVED_PER_REPLY / 3600.0).round,
       reopen: reopen_rate(range),
       depth: depth_conversations.zero? ? 0 : (public_count.to_f / depth_conversations).round(1)
     }
@@ -96,15 +100,6 @@ class Captain::AssistantStatsBuilder
     }
   end
 
-  # Average reply time (seconds) for both windows in one scan.
-  def avg_reply_times
-    row = account.reporting_events.where(name: 'reply_time', created_at: full_span).reorder(nil).pick(
-      Arel.sql("AVG(value) FILTER (WHERE #{window_clause(current_range)})"),
-      Arel.sql("AVG(value) FILTER (WHERE #{window_clause(previous_range)})")
-    )
-    { current: row[0].to_f, previous: row[1].to_f }
-  end
-
   # Resolved and handed-off conversation counts for one window, in a single scan
   # of the handled set's reporting events.
   def resolution_counts(range)
diff --git a/lib/seeders/reports/assistant_conversation_creator.rb b/lib/seeders/reports/assistant_conversation_creator.rb
index 542da1b42..9462d899e 100644
--- a/lib/seeders/reports/assistant_conversation_creator.rb
+++ b/lib/seeders/reports/assistant_conversation_creator.rb
@@ -90,9 +90,9 @@ class Seeders::Reports::AssistantConversationCreator
     return unless rand < 0.6
 
     travel(rand((1.minute)..(10.minutes)))
-    incoming_message(conversation)
+    follow_up = incoming_message(conversation)
     travel(rand((20.seconds)..(5.minutes)))
-    assistant_reply(conversation, waiting_since: Time.current)
+    assistant_reply(conversation, waiting_since: follow_up.created_at)
   end
 
   def apply_outcome(conversation, created_at, outcome)