Kroger—the united states’ largest traditional grocer - runs thousands of microservices that power everything from personalized coupons to real-time inventory. Its cloud platform team maintains hundreds of development and staging clusters on Azure and GCP so engineers can test safely before code reaches production.

Background

To avoid cold starts, each team kept at least one node per cluster running 24 × 7. the result: > 20 m node-hours/month—most of them idle.

Challenges

1. idle spend: Dev clusters sat empty nights & weekends, yet nodes kept billing.
2. manual tuning: Teams tweaked HPA (Horizontal Pod Autoscaling) settings for every release, with mixed results.
3. fragmented visibility: Azure & GCP dashboards showed different metrics; leaders lacked a single savings view.
4. time-boxed pov: Kroger needed proof of value in < 30 days, without risking developer velocity.

Solutions

Smart Scaler was first rolled out to Kroger’s development clusters in observation mode, capturing a clean performance baseline without touching workloads. Leveraging reinforcement-learning-powered AI, it then produced detailed right-sizing recommendations—optimal CPU and memory requests for every pod—while confirming that all service-level objectives remained intact. Unlike competing tools that look only at infrastructure signals and treat right-sizing (VPA) and pod-count scaling (HPA) as separate, lag-prone steps, Smart Scaler optimizes both in one continuous loop, using fine-grained application and infrastructure metrics; the result is fewer scaling errors, higher uptime, better performance, and materially lower cloud costs. After Kroger’s engineers reviewed and approved the suggested thresholds, the same clusters were switched to full autoscaling and optimization mode, allowing Smart Scaler to drive HPA and VPA decisions autonomously. That proven template has since been extended across every production and non-production environment company-wide.

Results

Customer Voice

“Smart scaler paid for itself before the pov even ended. we slashed idle spend by seventy percent and handed engineers a platform that scales itself.”  - Senior director, cloud platform engineering, Kroger

Conclusion

Smart Scaler gave Kroger a self-optimizing dev & staging estate that

• Rightsizes pods automatically and predicts when to scale to zero,
• Unifies cost visibility across Azure and GCP,
• Delivers hard savings in weeks—long before production rollout.