diff --git a/static/app/views/insights/queues/charts/latencyChart.tsx b/static/app/views/insights/queues/charts/latencyChart.tsx
index 361102f7d0f537..18cd0442756837 100644
--- a/static/app/views/insights/queues/charts/latencyChart.tsx
+++ b/static/app/views/insights/queues/charts/latencyChart.tsx
@@ -1,4 +1,7 @@
+import cloneDeep from 'lodash/cloneDeep';
+
 import {t} from 'sentry/locale';
+import {defined} from 'sentry/utils';
 import {useProcessQueuesTimeSeriesQuery} from 'sentry/views/insights/queues/queries/useProcessQueuesTimeSeriesQuery';
 import type {Referrer} from 'sentry/views/insights/queues/referrers';
 
@@ -21,13 +24,36 @@ export function LatencyChart({error, destination, referrer}: Props) {
     referrer,
   });
 
+  const messageReceiveLatencySeries = cloneDeep(
+    data['avg(messaging.message.receive.latency)']
+  );
+
+  if (
+    !isPending &&
+    !error &&
+    defined(messageReceiveLatencySeries.data) &&
+    defined(messageReceiveLatencySeries.meta) &&
+    !defined(
+      messageReceiveLatencySeries.meta?.fields['avg(messaging.message.receive.latency)']
+    )
+  ) {
+    // This is a tricky data issue. If Snuba doesn't find any data for a field,
+    // it doesn't return a unit. If Discover can't guess the type based on the
+    // unit, there's no entry in the meta for the field. If there's no field,
+    // `TimeSeriesWidgetVisualization` dumps that data onto its own "number"
+    // axis, which looks weird. This is a rare case, and I'm hoping that in the
+    // future, backend will be able to determine types most of the time. For
+    // now, backfill the type, since we know it.
+    messageReceiveLatencySeries.meta.fields['avg(messaging.message.receive.latency)'] =
+      'duration';
+    messageReceiveLatencySeries.meta.units['avg(messaging.message.receive.latency)'] =
+      'millisecond';
+  }
+
   return (
     <InsightsAreaChartWidget
       title={t('Average Duration')}
-      series={[
-        data['avg(messaging.message.receive.latency)'],
-        data['avg(span.duration)'],
-      ]}
+      series={[messageReceiveLatencySeries, data['avg(span.duration)']]}
       aliases={FIELD_ALIASES}
       error={error ?? latencyError}
       isLoading={isPending}