Demand load
Work arriving per time unit.
Business Domains · Module 16
Operations Analytics.
Operations analytics is about flow: work enters a system, consumes capacity, waits in queues, and either meets or misses a service promise.
Core idea
Operational averages can look healthy while customers suffer in the tail. Measure flow, queue, capacity and service distribution.
01
Context
Average delivery time is stable. SLA misses doubled.
“The average still says 31 minutes. Why are complaints exploding?”
The distribution widened. The median may be stable while the p90 or p95 deteriorates badly.
Operations analysis should isolate queueing, processing, handoff and capacity constraints rather than rely on one average.
02
Operations system
Measure the flow, not just the final duration.
Capacity
Resources available to process the load.
Backlog
Work waiting because arrival exceeds throughput.
Service level
Percentiles and share meeting the SLA.
Rule
If the SLA is about the tail, the average is not the primary metric.
03
Challenge
Find the bottleneck behind SLA misses.
Exercise · 20 minutes
Trace the process.
- Which process stages contribute to total time?
- What is p50/p90/p95 by city and hour?
- How does arrival rate compare with capacity?
- Where does backlog accumulate?
- Are failures concentrated in a handoff?
- What capacity change would affect the bottleneck rather than a non-bottleneck?
04
AI assist
Ask AI to build a bottleneck analysis.
Operations analysis prompt
Investigate why 45-minute SLA misses doubled while average delivery time stayed near 31 minutes.
Break total time into:
order confirmation, preparation, courier assignment, pickup, delivery.
Report:
p50, p90, p95, SLA pass rate, backlog and capacity/load ratio
by city and hour.
Identify bottleneck evidence.
Do not recommend staffing changes unless the constrained stage is identified.05
Validation
The operations checklist.
01
Are process timestamps complete and ordered correctly?
02
Are percentiles used where tail performance matters?
03
Is backlog defined at a consistent point in the process?
04
Are demand load and capacity aligned in time/geography?
05
Could mix shifts explain slower cases?
06
Does the recommendation target the actual bottleneck?
06
Worked takeaway
Fix the constraint that controls the system.
Use percentiles and SLA pass rate.
Break end-to-end time into stages.
Compare load to available resources.
Target the stage where queues actually form.
Analyst habit
When service quality worsens, ask where time is being spent and where work is waiting.