Right-sizing

Sencai has two separate screens that both aim at the same goal - telling you when an instance is bigger (and more expensive) than it needs to be - using two different, and quite differently reliable, methods. Neither is currently reachable from the sidebar navigation; both are direct URLs under Intelligence.

Right-Sizing Recommendations

/gravity/analytics/rightsizing lists downsize suggestions generated by comparing each cloud instance’s type against a known hierarchy of instance sizes for its provider - for example, suggesting a smaller instance type one or two steps down from what you’re running.

Caution

Read this carefully before acting on anything here: these recommendations are rule-based, not usage-based. They compare instance type names against each other - they do not look at your actual CPU, memory, or I/O utilization at all. The screen itself labels every recommendation’s confidence as low for exactly this reason. Always check real utilization in your cloud provider’s own console before resizing anything based on a recommendation here.

A summary row at the top shows how many recommendations are pending, the total potential monthly saving across them, and how many you’ve already applied. Filter chips let you narrow the table to all, pending, applied, or dismissed.

Each row shows the instance, its current and recommended type, estimated monthly savings, confidence, provider, and status (pending, applied, or dismissed). For a pending recommendation:

• Apply marks it as applied in Sencai and shows you the exact resize you’d need to make (current type → recommended type, provider, region) - Sencai does not perform the resize for you; you do it in your cloud console
• Dismiss removes it from the pending list

Note

This data source is not enabled in every environment. If the table is empty, that’s the expected state until your account team turns on the underlying analysis for your organization - it isn’t a sign that nothing on your account could be right-sized.

AI Workload Profiling

/gravity/analytics/workload-profiles is a separate, more capable mechanism: it analyzes each instance’s actual behavior - average CPU and memory usage, and which hours of the day are peak versus idle - and classifies it as CPU-bound, memory-bound, I/O-bound, idle, or mixed. Where it finds a genuine mismatch, it suggests a migration target instance type with an estimated monthly saving and a confidence percentage.

Each profile’s detail view shows a 24-hour activity chart marking peak hours (CPU ≥ 80%) and idle hours (CPU < 20%), plus a written recommendation summary. Filter the list by profile type or by status (pending review, accepted, implemented, dismissed). For a pending profile:

• Accept the recommendation
• Dismiss it

Note

Like the rule-based screen above, this one depends on a recurring analysis job that has to be turned on for your organization. If you don’t see any profiles yet, ask your account team whether workload profiling is enabled for your account - profiles are meant to be generated once a day for running instances when it is.

Which one should you trust more?

If both have data for the same instance, prefer AI Workload Profiling - it’s looking at what the instance actually does, not just comparing type names. Treat Right-Sizing Recommendations as a rough, low-confidence starting point at best, never as something to act on without verifying utilization yourself first.

The risk of acting on thin data

Both screens estimate a monthly saving, and it’s tempting to treat that number as reliable enough to act on immediately. Resist that for the rule-based screen in particular - a “low confidence” label isn’t decorative, it means the suggestion has no idea what your instance is actually doing. Even for AI Workload Profiling, a confidence percentage below roughly 70% should be read as “worth a look,” not “safe to apply.” Downsizing a genuinely busy instance based on a bad recommendation can turn into a real incident, not just a wasted saving - verify utilization yourself first, every time, regardless of which screen the suggestion came from.

What’s next

• Cost allocation & savings - the broader savings-recommendation list these two screens complement
• Instances - where you’d actually perform a resize
• Cloud cost management - how cost data is ingested in the first place
• Recommendations - proactive, organization-wide suggestions from Lumen