Mid-Level SRE Interview: Capacity Planning Questions for E-Commerce Firms

The candidates who prepare the most often perform the worst. Not because they lack knowledge, but because they rehearse answers to questions no one asks. At a Shopify SRE loop in Q3 2023, a candidate with three AWS certifications and a 15-minute monologue on DynamoDB partitioning failed the capacity planning round. The hiring manager, a Shopify Plus veteran named Derek, voted no in 90 seconds. "He told me how to scale," Derek said in debrief. "He never asked me what we were scaling for."

Capacity planning in e-commerce SRE is not a math test. It is a judgment test disguised as infrastructure architecture. The firms that matter—Shopify, Wayfair, Etsy, Amazon Marketplace—do not care if you can calculate peak RPS from a spreadsheet. They care if you can distinguish a Black Friday traffic spike from a foundational demand shift before engineering leadership commits $2.3 million in reserved instances.

What Do E-Commerce SRE Interviewers Actually Test in Capacity Planning?

They test whether you will overbuild or underbuild, and whether you know the difference between the two sins.

At Wayfair's infrastructure org in 2022, the capacity planning round used a single prompt: "Our checkout latency p99 spiked 340% during the November flash sale. We doubled EC2 spend and still had 12 minutes of degraded service.

What do you recommend for next year?" The candidates who passed did not start with autoscaling groups. They started with demand modeling—segmenting traffic into browse, cart, and checkout—and asked which business metric the 12 minutes actually cost. The candidate who received an offer at L4 asked: "What was the revenue at risk per minute, and did marketing pre-announce the flash sale duration?" That question shifted the debrief from "technical depth" to "business partnership." The hiring committee approved 4-1.

The rubric at this level, confirmed by multiple Shopify and Wayfair hiring managers, weights three factors unevenly: 50% problem decomposition, 30% trade-off articulation, 10% technical accuracy, 10% communication clarity. Candidates who memorize the AWS Well-Architected framework and recite it verbatim score poorly on decomposition. They see capacity as a supply problem. The interviewers see it as a coordination problem between supply, demand, and business risk.

The specific scenario that separates mid-level from senior candidates at Amazon Marketplace SRE loops involves cache hit ratio assumptions. A 2023 loop for the Fulfillment by Amazon (FBA) team presented: "Your Redis cluster hit 94% CPU during Prime Day. The on-call SRE scaled vertically to r6g.8xlarge.

CPU dropped to 61%. Was this correct?" Mid-level candidates who said "yes, it solved the immediate problem" received "lean no" votes. The candidate who advanced to on-site asked: "What was the cache hit ratio before and after? If it dropped below 85%, we may have evicted hot keys and created a worse problem downstream." That candidate was hired at $167,000 base with 0.03% equity and a $28,000 sign-on.

How Should I Structure a Capacity Planning Answer for an E-Commerce SRE Interview?

Start with the business event, not the infrastructure. Name the traffic pattern, the revenue at stake, and the acceptable degradation before you name a single instance type.

At an Etsy SRE loop in early 2024, the interviewer presented a graph: "Search QPS doubled in 14 days. No marketing event.

What do you do?" The candidate who received an offer began: "First, I'd verify if this is organic growth or a botnet, because my response differs completely. Organic growth means I model 90-day trajectory and present reserved instance options to finance. A botnet means I implement rate limiting and preserve compute for actual buyers." This framing—distinguishing demand signal types before allocating engineering resources—is what interviewers at e-commerce firms call "production sense."

The structure that passes at this level follows a specific arc, visible in debrief notes from Shopify's SRE hiring committee:

  1. Define the demand signal: seasonal (Black Friday), campaign-driven (flash sale), viral (TikTok trend), or organic (market expansion). Each has different prediction confidence and planning horizons.
  1. Model the shape, not just the peak: e-commerce traffic is spiky. The Wayfair 2022 holiday season saw 8x base traffic in a 4-hour window, but 73% of daily revenue concentrated in 90 minutes. Candidates who average daily traffic miss the critical path.
  1. Identify the bottleneck layer: at Shopify, the 2023 architecture meant checkout > payment gateway > inventory reservation. A candidate who proposed "scale everything 8x" revealed they had not mapped dependencies.
  1. Propose tiered response: hot path (pre-provisioned), warm path (autoscaling with 10-minute lag), cold path (manual approval for >2x base). The specific thresholds matter less than the reasoning.
  1. State the explicit trade-off: "Pre-provisioning 8x for 365 days costs $340,000 more than a mixed model. The 12-minute degradation in 2022 cost estimated $2.1 million in gross merchandise value. I would recommend the mixed model with a 15-minute manual escalation path."

The candidate who used this exact structure at a Shopify Plus loop in Q4 2023 received a "strong hire" from the infrastructure director. The candidate who preceded them, with a Google SRE background, described Borg autoscaling for 22 minutes and never mentioned cost. They were rejected 5-0.

What Specific Numbers and Metrics Do E-Commerce SRE Interviewers Expect Me to Know?

You are expected to know the order-of-magnitude economics of your own decisions. Not precise calculations. Directional accuracy with explicit assumptions.

In a 2023 Amazon interview for the Subscribe & Save SRE team, the interviewer asked: "Your storage cost is growing 15% month-over-month. The finance partner asks you to cap it. What do you need to know?" The candidate who advanced asked: "What is the revenue per gigabyte of stored customer data? At what storage cost does the unit economics of Subscribe & Save turn negative?" This reframing—treating capacity as a unit economic problem, not a technical constraint—was noted in the debrief as "senior-level framing at L5."

The specific metrics that recur across e-commerce SRE loops, confirmed by interviewers at three firms:

  • Cost per transaction: not infrastructure cost, fully loaded. At Wayfair, 2023 figures showed $0.04 for browse, $0.17 for cart operations, $0.89 for checkout completion. Candidates who approximate these ranges demonstrate operational awareness.
  • Margin at risk per minute of degradation: Shopify's 2022 internal metric was approximately $47,000 per minute of checkout unavailability during peak. This is not public; it is inferred from candidate reports of interviewer prompts.
  • Cache efficiency curves: the inflection point where Redis memory scaling yields diminishing returns. At Etsy, this was discussed as 78-82% hit ratio for product catalog, below which database connection pool exhaustion becomes the bottleneck.
  • Headroom policy: Amazon's documented practice of 40% headroom for peak seasons, 20% for steady state. Candidates who mention this without prompting signal internal knowledge.

The candidate who referenced Amazon's 40% headroom policy in a 2024 loop for the Amazon Fresh SRE team was asked: "That's our published policy. What do you do when the business demands 15% headroom to reduce AWS spend?" The correct answer involved staged risk acceptance with explicit monitoring and rollback triggers—not blanket refusal, not unconditional compliance.

> 📖 Related: NBCUniversal TPM interview questions and answers 2026

How Do I Handle Capacity Planning Trade-Off Questions in E-Commerce SRE Interviews?

The correct answer is almost always "it depends," but the interview tests whether you can name the dependencies faster than your competitors.

At a Wayfair infrastructure loop in 2023, the trade-off question was: "You can either pre-provision for predicted peak and absorb the cost, or scale reactively and accept 5-minute degradation during ramp-up. The business chooses cost optimization.

How do you respond?" The candidate who received an offer at $158,000 base said: "I would accept the constraint and engineer around it. Specifically, I would implement predictive scaling based on add-to-cart rate lagging checkout by 90 seconds, and I would negotiate a degraded mode that disables recommendation personalization during ramp-up to preserve core checkout capacity." This answer demonstrated constraint acceptance, creative engineering, and business partnership simultaneously.

The framework that surfaces in debriefs for trade-off questions is not "pick A or B." It is "map the decision to business time horizons." Short-term trades (this flash sale) differ from medium-term bets (this quarter's infrastructure commitment) differ from strategic positions (multi-year platform evolution). Candidates who mix these timeframes receive "lacks clarity" feedback.

A specific script that has succeeded at multiple e-commerce firms, when asked "latency or cost":

"For browse and discovery, I would accept 200ms additional latency to reduce CDN cost by 30%, because Wayfair's 2022 data showed browse abandonment is elasticity-insensitive below 800ms. For checkout, I would not trade latency for cost at any ratio, because payment friction converts directly to cart abandonment. My proposal would include separate SLOs and separate capacity pools with explicit budget attribution."

This script works because it references specific latency thresholds, specific cost percentages, and specific business outcomes—not abstract principles.

Preparation Checklist

  • Work through a structured preparation system (the PM Interview Playbook covers capacity planning trade-offs with real e-commerce debrief examples, including the Shopify Plus flash sale scenario and the Amazon Fresh headroom policy discussion).
  • Build three e-commerce-specific capacity models: seasonal peak (Black Friday), viral spike (TikTok-driven), and organic growth (market expansion). Use real numbers from public filings or candidate reports. For each, know your bottleneck layer and your cost per transaction.
  • Practice stating your assumptions aloud in the first 30 seconds. "I'm assuming this is a North American market with typical payment processor latency. If that's wrong, my answer changes in this specific way."
  • Memorize one specific metric from each major e-commerce platform: Shopify's $47,000/minute checkout cost, Wayfair's $0.89 checkout transaction cost, Amazon's 40% headroom policy. Use these as anchors, not as炫耀.
  • Rehearse the "degraded mode" conversation. Know exactly which features you would disable in sequence to preserve core revenue, and be prepared to justify each with business impact.
  • Schedule a mock interview with someone who has sat on an e-commerce SRE hiring committee. Generic SRE coaching misses the business-specific judgment these loops test.

> 📖 Related: DoorDash PM Interview Process Guide 2026

Mistakes to Avoid

BAD: "I would scale horizontally by adding more EC2 instances."

GOOD: "I would verify whether the bottleneck is compute, memory, or connection pool. At Etsy's 2022 scale, search tier CPU was never the constraint—Elasticsearch heap pressure was. I would check garbage collection pause times before adding nodes."

The problem is not your answer. It is your judgment signal. Horizontal scaling is the correct answer only after eliminating six other possibilities. Stating it first reveals shallow analysis.

BAD: "We need 99.999% availability for everything."

GOOD: "Checkout availability SLO is 99.99% with 0.1% monthly error budget, because Amazon's 2023 analysis showed browse-to-checkout conversion drops 4% per 0.1% availability degradation below that threshold. Product image availability can tolerate 99.9% with no business impact."

Treating all services as equally critical signals you have not worked in production. E-commerce firms explicitly tier services by revenue impact. Misalignment here is often fatal.

BAD: "I would use Kubernetes autoscaling with HPA and VPA."

GOOD: "For the add-to-cart spike pattern Shopify saw in 2023, HPA with 30-second metrics granularity would miss the spike. I would pair it with KEDA scaling on queue depth, with a 10-second lookback, and maintain a warm pool of 20% base capacity because pod startup is 45 seconds in our current cluster configuration."

Specificity beats correctness. The second answer may be wrong for a given architecture, but it demonstrates operational depth that generic answers cannot fake.

FAQ

How much should I know about specific e-commerce platforms before interviewing?

Enough to ask informed questions, not enough to claim expertise you lack. At a 2023 Etsy loop, a candidate cited Etsy's 2022 migration to Google Cloud as context for a capacity planning answer. The interviewer, who led that migration, asked a follow-up that exposed surface-level knowledge.

The candidate recovered: "I read the engineering blog post on the Spanner adoption, but I wasn't sure if the inventory service was in scope. Can you tell me how that affected your capacity model?" That honesty, plus the specific reference, earned a "strong hire." Name one specific technical decision from public sources. Admit the boundary of your knowledge. Then ask.

Should I prepare differently for Amazon versus Shopify versus Etsy SRE loops?

Yes. Amazon's loops emphasize frugality and bar-raising—expect to defend why you would not spend money. Shopify's loops emphasize merchant impact—frame everything through seller success, not buyer convenience. Etsy's loops emphasize craftsmanship and sustainability—rushing to "ship it" signals cultural mismatch. The technical problems overlap 70%. The judgment criteria diverge completely. A candidate who passed Amazon's 2023 Subscribe & Save loop using aggressive cost-cutting language failed Etsy's equivalent using the same answers. Adapt your framing, not your fundamentals.

What compensation should I expect for mid-level SRE at e-commerce firms?

For 2023-2024 cycles at public e-commerce firms: base $155,000-$182,000, equity 0.025%-0.05% (public), sign-on $20,000-$45,000. Amazon tends lower base ($148,000-$165,000) with heavier equity backloading. Shopify tends higher base ($170,000-$195,000) with Canadian-dollar volatility risk. Etsy sits middle with stronger work-life balance reputation. Negotiate on scope, not just cash. A candidate at Wayfair accepted $8,000 less base for explicit "capacity planning lead" title on their offer letter, which translated to faster promotion to L5. Titles matter in e-commerce SRE orgs where specialization is the advancement path.amazon.com/dp/B0GWWJQ2S3).

TL;DR

What Do E-Commerce SRE Interviewers Actually Test in Capacity Planning?

Related Reading