Skip to main content
Metrics type: Cross-Platform Metrics • Category: Monitoring

​
At a glance

A scatter / line chart that pairs each minute’s 5xx response rate against the same minute’s cart-abandonment rate from the connected commerce sibling. Visually answers “do my server errors translate into customers walking away from carts?” The slope of the relationship is the merchant’s abandonment sensitivity coefficient.

​
Calculation

Calculated automatically from your New Relic data. See the At a glance summary above for what the metric tracks and the worked example below for a typical reading.

​
Worked example

A Shopify Plus brand with NR APM. Baseline cart-abandonment rate (during healthy operations) sits at 68% (industry-typical for non-luxury commerce). 5xx rate baseline is 0.05%. A 5xx spike occurs at 13:14, 13:23 on 02 May 26: Reading the relationship. The scatter for these 11 minutes shows a clear positive correlation: every additional 1% of 5xx rate adds roughly 3% to the abandonment rate. This sensitivity coefficient of 3.0 is the actionable number, it tells the merchant exactly how much customer behaviour responds to server errors. Stores with strong brand loyalty might show 1.5; stores in highly competitive niches show 4, 5. Conversion impact translation. Over the 9-minute spike (13:14, 13:22), 234 carts were created and only 44 checkouts started. Baseline expectation: 234 x 32% completion = 75 checkouts. Lost: 75 - 44 = 31 checkouts, at AOV of £85 = £2,635 of GMV impact in 9 minutes. Apdex calibration interaction. During the worst-spike minute (13:17), Apdex on the storefront was 0.46, well below the 0.5 frustrated boundary. Errors automatically count as frustrated in Apdex math regardless of duration, so the 5xx spike directly drove Apdex into the frustrated zone. The sensitivity coefficient (3.0 abandonment-points per 1% error rate) is roughly consistent with the conversion-loss-per-100ms-latency rule of thumb extrapolated to the error case. Why the 30-day scatter matters. The sensitivity coefficient is calibrated over 30 days, not just one incident, because individual incidents have noise. With ~43,200 minute-tuples over 30 days, the regression line is statistically robust. A merchant whose coefficient moves from 3.0 to 4.5 over a quarter has lost some brand loyalty and customers are now more likely to walk away when something goes wrong, that’s a strategic signal worth a conversation.

​
Sibling cards merchants should reference together

​
Reconciling against the vendor’s own dashboard

Where to look in New Relic: NR doesn’t natively render this cross-platform correlation. The closest equivalent screens for the input signals: For the cart-abandonment input, compare against the merchant’s commerce platform’s funnel report (Shopify Analytics > Behavior > Funnel; BigCommerce Reports > Abandonment). Why our number may legitimately differ from the vendor’s: Cross-connector reconciliation: NR APM 5xx and Datadog APM 5xx should agree within 0.3% on healthy days; during incidents the gap can widen to 1, 2% as one platform sees errors the other doesn’t. The cart-abandonment rate is identical because both platforms read from the same commerce sibling. Cross-connector disagreements on this card therefore trace back to the 5xx series, audit instrumentation parity if it matters. NR APM 5xx vs Browser-observed 5xx (RUM): RUM captures 5xx that happen after the page loads (typically API calls); APM captures 5xx at the server boundary. Some 5xx never reach the browser (the customer’s request was queued and timed out before getting a response); some Browser 5xx never reach APM (third-party API failures called from the browser). The two views are complementary, not redundant.

​
Known limitations / merchant FAQs

NR vs Datadog: should the abandonment-sensitivity coefficient match across both platforms? Yes, within ~0.3 of the slope value. Both pull cart-abandonment from the same commerce sibling, so the y-axis is identical. Differences come from the x-axis (5xx rate); if NR and DD have different APM coverage, the 5xx series differs and the regression slope can shift slightly. A persistent slope-difference of >0.5 indicates real coverage drift worth auditing. Apdex math: why isn’t Apdex on this chart? This card is specifically about errors and abandonment, not latency and abandonment. Apdex bundles latency and errors into a single satisfaction score; using it on this card would muddy the error-specific signal. For the latency-and-conversion view, use Checkout App Health x Sales. NRQL retention: 30-day window is on the boundary of retention, is that a problem? On standard NR plans (8-day full-resolution retention), the 30-day window uses rolled aggregates for the older 22 days. Rate-vs-rate correlation is sample-stable so the slope of the relationship stays valid, but minute-by-minute scatter detail past day 8 reads at hourly resolution. For full 30-day minute-resolution, upgrade to NR Data Plus (13-month retention). NR and Datadog disagree on the slope by 0.4, who’s right? Almost certainly the platform with broader 5xx coverage is reading the higher slope (it sees more errors, so the regression line picks up more points and weights them more heavily). Audit which platform is missing instrumentation; once aligned, slopes typically converge within 0.2. Sampling: can sampling break this card? Rate-vs-rate correlation is sample-stable. NR’s high-cardinality event sampling does not affect the slope or the alert. Cart-abandonment is unsampled. Multi-account: my US and EU stores have different abandonment baselines, can the card handle both? Yes. Connect each pair (NR account + commerce sibling) as a separate integration; each gets its own card with its own sensitivity coefficient. The Nerve Centre stack panel can render both side-by-side; combined slope analysis is not currently supported (would require careful weighting by traffic). Ingest cost vs visibility: can I sample down 5xx events? No, never sample errors. Errors are rare enough that the cost of keeping 100% of them is small (typically <5% of total ingest), and sampling errors breaks alert sensitivity for low-volume failure modes. Sample down successful Transaction events freely; keep errors at 100%. Alert tuning: my alert fires every Black Friday because abandonment naturally rises with traffic, what should I tune? Two fixes: (a) use NR’s adaptive baseline condition type, which compares against the same time-of-day on prior weeks rather than a fixed +5% threshold; (b) require a co-incident 5xx spike to gate the abandonment alert (“abandonment +5% AND 5xx rate >1%”), which prevents firing during traffic-only abandonment surges. Adaptive baselining is generally preferred. My sensitivity coefficient is 5.5, is that bad? It’s high but not unprecedented. A coefficient of 5.5 means every 1% of 5xx rate produces 5.5% additional abandonment. Industry-typical for non-luxury commerce is 2.5, 3.5; coefficients of 5+ usually indicate one of: (a) low brand loyalty (customers don’t retry), (b) high-competition niche (alternatives are one click away), (c) unusually error-sensitive customer segment (B2B buyers, premium customers). The actionable response is to invest in error-recovery UX (helpful error pages, automated retry on the client) which can typically pull the coefficient down by 1, 2 points.

​
Tracked live in Vortex IQ Nerve Centre

Cart Abandonment During 5xx Spikes is one of hundreds of KPI pulses Vortex IQ tracks across New Relic and 70+ other ecommerce connectors. Nerve Centre runs the detection layer; Vortex Mind investigates the cause when something moves; Ask Viq lets you interrogate any number in plain English. Start for free or book a demo to see this metric running on your own data.