Peak season: what we changed after last year

The pattern repeats every year. A programme sends comfortably for ten months, then triples its volume in three weeks, and somewhere in the middle of the biggest weekend the inbox placement quietly moves. Nobody sees it on the day, because open rates are noisy and revenue is high anyway. It shows up in January, when the same list performs worse than it did in October.

We have now run enough peaks to have opinions about which precautions pay for themselves and which are theatre. Last year produced a short list of changes we have carried into this season, and none of them are about creative.

The preparation checklist we published before the last peak still stands, and if you have not been through it, start there rather than here. This piece is about the five things we do differently now.

We stopped adding volume in the week that matters most

The instinct is to save the biggest push for the biggest weekend. The problem is that mailbox providers judge a sudden change in pattern, and the worst week to be judged is the week when every other retailer is also behaving strangely.

So the ramp now starts early and rises gently. We increase volume from early autumn in steps, each step small enough that no single day is a large multiple of the day before it. Peak-level daily volume is reached a clear fortnight before the trading weekend.

Two practical consequences. The content of those early sends has to be worth sending, so the calendar needs real editorial reasons to mail more often, not filler. And any new sending domain or IP has to be in use months ahead, because a warm-up compressed into November is a reputation problem waiting to happen.

By the time the weekend arrives, the volume is not new. Only the offer is.

We ramp the engaged part of the list, not the list

The other instinct we have abandoned is widening the audience as peak approaches. Reactivating a dormant segment on the biggest sending week of the year puts your least engaged addresses in front of providers at exactly the moment your sender reputation is under most pressure.

The sequence we use instead:

• Suppress the deepest inactive tier before the ramp begins, using your normal inactivity window rather than a special peak exception.
• Run any reactivation attempt in early autumn, well clear of the ramp, and accept the result.
• Increase frequency first for the most engaged tier, then for the middle tier, and leave the tail alone entirely.
• Hold the same frequency capping structure you use all year, with the caps raised per tier rather than removed.

Removing the cap is the change we see most often and regret most often. A cap raised from four to six messages a week for engaged customers is a trading decision. A cap switched off is an unbounded one, and over a fortnight it produces complaint rates that take months to recover from.

We write the freeze and the kill switch down

Two operational rules now sit in the peak plan as named items with owners.

The freeze is a date after which nobody changes template code, data feeds, audience logic or the ESP configuration. Ours starts about ten days before the first major send. Copy and offers can change after the freeze; the machinery cannot.

The kill switch answers a question every team should settle in under a minute. Who can stop a send that is going out wrong, what is their phone number, and what is the sequence. Write it down with actual names, test that the named people have the platform permissions the procedure assumes, and put it somewhere findable at eight in the evening on a Saturday.

Alongside those, we pre-build the contingency assets before the freeze: a correction template, a plain apology send, and a holding version of the homepage-linked email that works if a landing page fails. Building these under pressure produces the mistakes you were trying to avoid.

We keep the holdout running through peak

Every year somebody proposes suspending the holdout group for the trading period, because withholding messages from any customer during peak feels like leaving money on the table.

We now argue against it every time, for two reasons.

The first is that peak is when the holdout is most informative. Customer intent is high enough that a reasonable share of held-out customers will buy anyway, and knowing what that share is tells you what the programme actually added during your most expensive weeks. Without it, every peak review reports gross revenue and calls it performance.

The second is continuity. A holdout that is paused for six weeks cannot support a clean year-on-year comparison afterwards, because the treated and held-out groups are no longer comparable populations.

Keep the group small, keep the selection random, and keep it stable across the season. Read it after the returns window closes rather than on the Monday, since peak revenue is inflated by purchases that come back in January.

We monitor placement, not just delivery

Delivery rates stay high right up until the moment they do not, and they say nothing about which folder a message reached. During peak we watch a narrower set of signals daily, and a person is assigned to look at them rather than an alert being assumed to work.

What is on the daily list:

• Complaint rate by mailbox provider, not blended, with a trigger level agreed in advance that pauses the ramp.
• Placement from a seed list, read as a directional signal about a provider rather than as a precise figure.
• Hard bounces by domain group, watched for the sudden appearance of a domain that was fine yesterday.
• Time from send start to completion, since queueing is often the first visible sign of throttling.
• Unsubscribe rate per send against the same send's rate last year, rather than against the programme average.

The rule that makes the monitoring useful is agreeing the response before the season. If complaints from one provider cross the trigger level, the next day's volume to that provider is cut and the audience narrows to the engaged tier. Deciding that in advance turns a judgement call at nine in the evening into a procedure.

What to do in the next fortnight

• Set the ramp schedule in the calendar now, with the target daily volume and the date it must be reached.
• Name the freeze date and the two people who hold the kill switch, and check their platform permissions today.
• Confirm the holdout will run untouched through the season, and book the post-returns review date while people still have diary space.
• Agree the complaint-rate trigger levels per provider, and write the response next to each one, because deliverability decisions made under pressure are usually the wrong ones. The groundwork for all of this sits in the peak season preflight.