A worked example on a fictional but realistic system. The Analyst was given one question. What follows is how it investigated, what it found, and the evidence it used — end to end.
A European tour operator running 150+ services across multiple clusters and three data centers. 55 million API calls per day at peak booking season. The estate has grown organically over a decade through three platform migrations, two acquisitions, and constant feature pressure.
The team believes they are operationally healthy — uptime is fine, latencies are within SLO, the dashboards are green. But every quarter the cost-to-serve climbs and nobody can say exactly why.
They asked ServiceHop a single question: where is our complexity coming from?
"Where is our complexity coming from?"
The team asked the Analyst a single question. What follows is the investigation it ran in response.
The Analyst planned four steps, ran them in sequence, and narrowed from 150 services to a single endpoint in under five minutes. Every step used a real tool against the ServiceHop model.
Step 1 · Traffic concentration
Tool: portfolio.facts
Computed Gini across 150 services over 14 days.
Result: Top 3 services absorb 82% of traffic.
→ Narrowed to: catalog-service (22%), booking-service
(34%), search-service (26%).
Step 2 · Endpoint profile
Tool: profile.endpoints(catalog-service)
Classified fields and fingerprinted response shapes.
Result: /catalog/items absorbs 94% of catalog-service
traffic. Average payload 18 KB. Average latency 2.1s.
→ Flagged for deeper analysis.
Step 3 · Cacheability simulation
Tool: cacheability.simulate(/catalog/items, 14d)
Replayed 186M calls through cache strategies.
Result: 78% of responses identical within 5-minute
windows. No cache in place. Simulated hit rate at
5-min TTL: 61%. Staleness risk: <0.3%.
→ Candidate finding.
Step 4 · Drift check
Tool: comparison.snapshot(now, -30d)
Compared against 30-day-old snapshot.
Result: /catalog/items traffic share grew from 18%
to 22% over 30 days. No behavioral change. Growth
is volume, not structure.
→ Finding confirmed and priority raised.
Four tools, four questions, one finding. Every number came from the model — none from the Analyst's narrative layer.
Across the 14-day sample, /catalog/items accounted for 186M of ServiceHop's 770M call observations — 24% of all catalog-service traffic and 22% of the estate overall. The endpoint was the single largest source of latency-weighted traffic in the portfolio: 2.1s average response time, called at an average rate of 154 requests per second during peak booking hours.
ServiceHop hashed every /catalog/items response using simhash clustering. Responses grouping into clusters larger than 1000 accounted for 78% of all calls within rolling 5-minute windows. The Analyst verified that the clusters corresponded to identical payloads, not just structurally similar ones — the product catalog changes slowly, and the endpoint is called on every page load by every upstream client.
The Analyst ran a 14-day replay of the endpoint against three cache strategies — a 1-minute TTL, a 5-minute TTL, and a content-hash-keyed strategy. The 5-minute TTL produced a 61% simulated hit rate with a measured staleness risk of 0.27%. The content-hash strategy reached 74% but at the cost of implementation complexity the team did not want to absorb. The Analyst recommended the 5-minute TTL as the practical optimum.
Cache /catalog/items with a 5-minute TTL. The simulation predicts a 61% hit rate, eliminates ~13M upstream calls per day, and carries a measured staleness risk below 0.3% — lower than the estate's existing cache tolerance on comparable endpoints. At this scale, on an estate of this shape, the typical infrastructure reduction for the catalog tier is in the mid-single-digit percent range annually. The exact number is available once ServiceHop is pointed at your traffic.