Post-peak: what your data is telling you in December

The January meeting usually opens with "did it work?", which is a question about a number that is not settled and would not be very informative if it were. Total peak revenue tells you what happened; it rarely tells you what to do.

Three questions are worth the analyst’s time, and all three are answerable.

Where did people fail rather than decline? A visitor who abandoned because the price was too high has told you something about pricing. A visitor who abandoned because a variant selector broke on their phone, or a discount code was rejected, or a delivery date was not shown, has told you about a defect. The second group is fixable and the first largely is not, and peak volume is the best chance all year to see the second group clearly.

Who bought, and what did they buy? Product-level and customer-level facts survive the distortions that ruin rate-level metrics. A product that outsold its forecast by four times is a merchandising and forecasting signal regardless of what the traffic mix was doing.

What broke operationally? Oversells, integration backlogs, support spikes, fulfilment delays. This is the most actionable data you collect at peak and the least likely to be in the report, because it lives in inboxes and spreadsheets rather than in analytics.

Before any analysis, know what is in the number. Shopify documents that its marketing reports include cancelled, pending and unpaid orders, and exclude test and deleted orders.

That is a sensible reporting choice and a poor basis for a January board slide. Peak produces an unusually high share of orders that do not convert to kept revenue: cancellations, fraud holds, failed payments on pending orders, and a returns wave that arrives weeks later. Reporting gross peak revenue in the first week of January means reporting a figure you will quietly revise.

The discipline is simple and unpopular: agree before peak which number the business is measuring — orders placed, revenue captured, or revenue net of refunds and cancellations — and report the same one every year. The third is the honest one and the one you cannot produce until February, which is precisely why most teams report the first.

Channel comparison is the analysis most requested after peak and the one most likely to mislead, because several mechanisms that are harmless in a normal month all misbehave at once.

Sessions are time-boxed in a way that peak violates. A Shopify session ends after 30 minutes of inactivity or at midnight UTC. A sale that launches at midnight, or a shopper who browses in the evening and buys in the morning, becomes two sessions — which inflates session counts and deflates the conversion rate computed from them. During peak, that pattern is not an edge case, it is the behaviour.

The model you are reading is a choice, not a fact. Shopify provides five attribution models — last non-direct click, last click, first click, any click and linear — with last non-direct click as the default for marketing activity. They will disagree, and the disagreement is largest in periods with long consideration windows and heavy multi-touch exposure, which is exactly what peak is. The "any click" model gives credit to every channel in the journey and can therefore allocate more than 100% of your orders, which makes it useful for understanding a single channel’s involvement and unusable for splitting a budget.

First interactions reset. Shopify documents that if a visitor does not purchase within 30 days, the first-interaction referrer resets, and that after an order the next referrer is treated as a new first interaction. A shopper who first arrived in October via search and bought in late November may not be attributed the way anyone in the room assumes.

The marketing report is not the sales report. "Sales attributed to marketing" includes only trackable effort — campaigns created through Shopify’s marketing section, or UTM-tagged traffic. It will not reconcile with total sales, and trying to make it reconcile is a week nobody gets back.

None of that means attribution is worthless. It means the right output of a peak attribution review is a ranked list of hypotheses to test in a quieter month, not a budget reallocation signed off in January.

Peak’s most valuable output is not the revenue. It is several thousand people who have now bought from you once. Whether that was a good trade depends entirely on what they do next, and nothing in your December data can tell you.

Shopify’s cohort analysis groups customers by the date they placed their first order and tracks their behaviour forward in time, which is the correct instrument for this. The thing to understand about it is that the picture is not stable: customer reports are based on a customer’s entire order history rather than only the orders in the selected period, so a customer who was new in November becomes a repeat customer in your November data once they buy again in December. Your peak cohort therefore keeps improving after peak, and any judgement made in the first week of January is made on the worst version of the figures you will ever see.

Two practical consequences. First, put a date in the calendar — three months and six months out — to look at the peak cohort properly, because nobody does it spontaneously once the quarter has moved on. Second, be sceptical of any "we acquired X customers" claim made in January, and of the acquisition cost derived from it: whether it was expensive depends on a second purchase that has not happened yet.

One smaller caveat on timing: Shopify notes that customer reports may not include activity from the past 12 hours, with the new-versus-returning report being the exception and current within seconds. Rarely material, occasionally the explanation for two dashboards disagreeing on a Monday morning.

Comparing peak conversion rate to your annual average and drawing a conclusion. The traffic is different, the intent is different, the discounting is different and the device split is different. A lower rate at ten times the volume is not a problem to solve; it may be the best week you had.

Reallocating next year’s budget from peak attribution. See above. Use it to generate hypotheses, test them in a normal month where the result is readable, and decide then.

Launching a test in January to fix what peak revealed. Tempting, and usually premature. January traffic is its own anomaly — gift-card redemptions, returns, discount-trained shoppers — so a test run in the first three weeks is measuring another unrepresentative period. Fix defects immediately; test preferences in February.

What is worth doing this week is the least analytical item on the list: write down everything that failed, while people still remember it. The support inbox, the oversells, the codes that did not work, the page that had to be edited by someone on leave. That list is more useful than any report you will produce from the peak data, and it has a shelf life of about a fortnight before everyone has rationalised it into "it went fine".

• Shopify conversion rate optimisation
• Black Friday: the changes to make and the ones to postpone
• Loyalty and retention: what actually brings people back