Paid and lifecycle channels: auctions, inboxes and partners

1 · What it is. Someone types a query, advertisers compete in an auction for the paid slots beside the free results, and the brand pays when its ad is clicked. Then the platform counts the conversions that follow under the account's attribution model.

2 · What marketers watch. Cost per click (CPC), click-through rate (CTR), conversions, and return on ad spend (ROAS): revenue the platform attributes to the ads, divided by what the ads cost.

3 · The vanity trap. Two numbers. First, platform-reported conversions include modeled ones: Google Ads Help says conversion modeling uses observed conversions to predict unobserved ones, for cases where cookies and identifiers were not available, and adds them to reported totals "only when there's a high confidence" (read at source, October 2026). That is a reasonable engineering choice and still not a row in your order table. Second, brand-term ROAS. Ads on the brand's own name look spectacular because the people typing "kettle and co" were already looking for Kettle & Co. Many of those clicks would have arrived through the free result directly below the ad, so the ad is partly buying traffic the brand already had.

4 · Where data adds value. A brand-term pause test: switch the brand-term ads off in some weeks or some regions and watch total brand traffic and orders, paid plus free, in the company's own data. If total orders barely move, the brand-term ROAS was mostly free clicks wearing a price tag. Join paid clicks to the company's own orders rather than to the platform's conversion count. The same ask, "Paid brand search buying free clicks?", sits on the triage desk in Partnership session 2 as ask 5, tagged as one the data team owns, because the evidence lives entirely in internal data.

1 · What it is. Ads compete in an auction for space in a person's feed, the platform's delivery system decides who sees them, and conversions are credited inside an attribution window the advertiser picks from the platform's menu. Where data is partial, the platform models it, then reports a ROAS.

2 · What marketers watch. ROAS, cost per acquisition (CPA), cost per thousand impressions (CPM) and reach.

3 · The vanity trap. Platform ROAS is the platform grading its own homework, under a window and a model it chose. Meta's Business Help Center, as listed in October 2026, offers click-through windows of 1 or 7 days, a 1-day view-through window and a 1-day engage-through window, and says the attribution setting is used to inform ad delivery as well as to credit conversions. It also says Meta may use statistical modelling where conversion data is missing or partial. A 7-day click, 1-day view setting is commonly reported as the default (reported, by the practitioner Jon Loomer; Meta's page does not state it). Meta's own page warns that comparing across attribution models "will lead to inaccurate conclusions". So a ROAS from Meta and a ROAS from Google are not two readings of one quantity. Constructed: Kettle & Co's Meta campaign reports ROAS 4.1 on spend of 24,000 pounds, and the marketing lead asks whether to double the budget.

4 · Where data adds value. A holdout. Meta's Conversion Lift, as described on its help page in October 2026, randomises people into a test group and a control group of people who do not see the ads, reads the result with an intent-to-treat approach, and reports the lift as an estimate. That is a different kind of evidence from the attribution screen, because the comparison group was chosen at random rather than inferred. Angles session 6 reads Kettle & Co's constructed lift test in full; the short version is that the incremental ROAS comes out far below 4.1 (constructed: 1.56). The ask "Meta ROAS 4.1: double budget?" is ask 6 on the Partnership session 2 triage desk, where data is consulted and the marketing lead decides: whoever lives with the outcome owns the decision.

Attribution asks: of the sales that happened, which ads were nearby? Incrementality asks: how many of these sales would not have happened without the ads? The first is answered by counting inside a window; the second needs a comparison group that did not see the ads. Meta also offers an incremental attribution setting that, in its own description, uses machine learning models to predict whether a conversion was caused by an ad. It is a model of incrementality, not a holdout, and worth reading as such.

Attribution models, marketing mix modelling and geo-lift design have their own home in Marketing Attribution ↗. Writing the ads themselves with AI is taught in AI + Marketing ↗.

1 · What it is. The brand sends a message to people who agreed to hear from it, as a one-off campaign or an automated lifecycle flow (welcome, reorder reminder, win-back). The message lands in the inbox or in spam, a person may click, and some clicks become orders.

2 · What marketers watch. Open rate, click rate, unsubscribes, and revenue per send.

3 · The vanity trap. The open rate after Mail Privacy Protection. Apple announced it in its newsroom on 7 June 2021, alongside iOS 15, saying it stops senders from knowing when a person opens an email and masks the person's IP address. Apple's legal privacy page explains the mechanism: Mail downloads remote content in the background by default, "regardless of whether you engage with the email" (both read at source, October 2026). An open is counted when a tiny invisible image loads, so for Apple Mail readers the image now loads whether or not a human ever looked. Constructed: Kettle & Co's marketing lead reports that opens "jumped to 48 percent" and asks whether the new subject lines are genius. The likelier story is the background download. You will also meet a figure of "around 75%" for Apple opens from the vendor Litmus in 2021; it was a forecast of what MPP would do, not a measurement, so do not quote it as data.

4 · Where data adds value. Replace opens with clicks and orders per send, both of which need a human. For the question "is email working at all", run a holdout: keep a random slice of the list out of a send or a flow and compare orders per customer between the two groups. Angles session 6 reads a constructed email holdout honestly, including what to say when the difference cannot be told from zero. The open-rate ask is ask 7 on the Partnership session 2 triage desk, owned by data because the answer is in the send logs and the order table.

Campaign sends are visible: someone writes them, schedules them, and reports on them. Lifecycle flows, the welcome series, the "running low on tea" reminder, the win-back after ninety quiet days, run silently and are often never re-read after launch. They are also the sends most likely to be credited with orders that would have happened anyway, because they target people who are already about to buy. The data team's useful move is the same as for campaigns: orders per recipient against a random holdout, flow by flow, read once a quarter. A flow that cannot beat its own holdout is a cost, not a channel.

Using AI to write or personalise the emails themselves belongs to AI + Marketing ↗.

1 · What it is. A publisher, a review site, a cashback or coupon site, promotes the brand with a tracked link or code. When an order follows, the affiliate network usually pays the partner whose link was clicked last.

2 · What marketers watch. Affiliate-attributed revenue, commission paid, and the cost per order that commission implies.

3 · The vanity trap. Affiliate-attributed revenue when coupon sites intercept buyers who were already at checkout. A buyer with a full basket sees a "discount code" box, opens a new tab, searches for a code, clicks a coupon site's link, and comes back. The order was going to happen; the coupon site supplied the last click and collects the commission, and the brand also gives away the discount. On the affiliate report this looks like efficient revenue.

4 · Where data adds value. Split each affiliate's orders into new and returning customers using the company's own customer table, and look at what happened next. Constructed: Kettle & Co's affiliate channel shows the cheapest paid acquisition cost in Q3, 15.0 pounds per new customer, and the worst 6-month repeat of any channel, 14 percent. That pairing is the coupon-site signature worth a meeting. The BD lead's ask "Affiliates that are coupon sites" is ask 8 on the Partnership session 2 triage desk, owned by data, because the evidence is the order table.

1 · What it is. The brand signs a contract with a creator and pays a fee for posts, usually with a discount code or tracked link. The audience is the creator's, not the brand's. Organic creator work and how creators reach people are read in Angles session 2; this card is only the paid contract.

2 · What marketers watch. Reach, views and engagement per post, and code redemptions.

3 · The vanity trap. Reach and engagement. They describe the creator's audience, which the brand paid to borrow for a day; they do not say whether anyone in it became a customer.

4 · Where data adds value. Cost per new customer by partner: the fee divided by first-time buyers who used that partner's code or link, checked against the customer table so returning customers using a code do not count as new. Code tracking undercounts, so treat it as a floor and say so. Constructed: Kettle & Co's creators channel cost 36.0 pounds per new customer in Q3, the most expensive channel on the table, with 31 percent 6-month repeat.

1 · What it is. The brand takes a stand at a fair, staff scan the badges of people who stop by, and the BD team follows up. Some of those people belong to businesses that open an account and order, often weeks later.

2 · What marketers watch. Leads collected and meetings booked, sometimes cost per lead.

3 · The vanity trap. The badge-scan count. A scan records that someone stopped at the stand, which includes competitors, students, other exhibitors and people who wanted a free sample. A big number says the stand was busy, not that it was worth its cost.

4 · Where data adds value. Match the scans to accounts in the company's own customer and order data, and count the accounts that ordered within an agreed window, say 90 days, separating new accounts from existing customers who would have ordered anyway. The BD lead's ask "Was the trade fair worth it?" is ask 9 on the Partnership session 2 triage desk, tagged as needing both kinds of knowledge: data supplies the matched orders, and the BD lead, who knows which conversations mattered and what the fair did for relationships, makes the call on next year's stand.