Research report
2027 B2C Growth Predictions
10 shifts reshaping retention, lifecycle and predictive marketing
Every number a B2C growth team reports today describes something that already happened. Last month’s churn. Last week’s ROAS. The segment that was accurate the day it was built. In 2026, teams got faster at reacting to those numbers. In 2027, reacting stops being enough, because the brands that have moved to prediction act on behavior before it shows up in anyone’s report, and they’re pulling ahead at a pace the quarterly review can’t see.
The ten predictions below are the same shift seen from ten different seats: the media buyer, the lifecycle manager, the retention lead, the analyst, the CRM team. Each one comes with a number that shows it’s already underway, and two or three things you can do about it this quarter. None of them require a reorg. Most of them require a holdout.
We’ve ranked nothing. They arrive together.
- 30% top-performing Braze customers more likely to use AI to anticipate purchase intent Braze’s 2026 Global Customer Engagement Review
- 12% higher average ROAS with Meta’s “maximize value of conversions” goal Meta, June 2025
- 41% of the base lost in a year when you compound 4.25% a month Recurly’s July 2026 benchmark
- 75% of marketers cannot estimate what proportion of their promotions are profitable Kantar and Les Binet’s October 2026 study
1. Predicted LTV replaces ROAS as the number growth teams answer to
The shift. By the end of 2027, the growth teams that matter will report predicted lifetime value by cohort as their headline number, and ROAS will be a diagnostic.
The signal. Karl Thumm, who runs data and analytics at the wedding brand Zola, says it in one line in Zola’s story: “Marketing’s North Star metric is lifetime value over cost to acquire a customer.” He has company. In Braze’s 2026 Global Customer Engagement Review, top-performing Braze customers are 30% more likely to use AI to anticipate purchase intent than the rest, and anticipated intent is what a predicted LTV is made of.
What to do now.
- Pick one acquisition channel and calculate 90-day revenue per customer by cohort, next to the ROAS you reported. The gap is your case.
- Define the day-one signals you already capture (first order value, category, device, promo used, first-week engagement) and check how well they predict 12-month value. They usually predict it better than you’d expect.
- Agree with finance on one pLTV definition before the model exists, so nobody relitigates the number after it does.
2. Ad platforms stop paying for conversions and start paying for value
The shift. Value-based bidding becomes the default on Google, Meta and TikTok for any B2C brand with repeat purchase, and brands that send the platforms a first-order value will lose auctions to brands that send a predicted one.
The signal. The platforms have already built the plumbing and are publishing the results. In June 2025 Meta reported that advertisers using its “maximize value of conversions” goal, instead of optimizing for conversion volume, achieved an average ROAS 12% higher, and said it’s testing profit-based ROAS fed through the Conversions API. Google’s own data on the switch from target CPA to target ROAS showed conversion values rising 14% at a similar ROAS. The auction will optimize for whatever number you send it. In 2027 the number is value.
What to do now.
- Audit what value you’re passing back today. If it’s first-order revenue, you’re teaching the auction to find bargain hunters.
- Run one campaign with predicted value passed as the conversion value and the current setup as the control, for four weeks. Compare 90-day revenue, not ROAS.
- Make sure the value signal reaches the platform inside the attribution window; a prediction that lands a week late is a prediction the auction never sees.
3. Lifecycle teams get a prediction layer, not just a messaging calendar
The shift. The lifecycle calendar survives, but every journey on it gets gated by a score: who enters, who’s suppressed, which branch they take.
The signal. There’s a 40-point gap between what lifecycle teams think they’re doing and what customers feel. In Braze’s 2026 review of 2,200 B2C marketing executives, 93% of marketing leaders believe AI helps them accurately understand customer needs, while only 53% of consumers feel brands are successfully predicting their wants. A calendar of well-timed messages can’t close that gap. A prediction about each customer can.
What to do now.
- List your ten highest-volume journeys and write down the rule that triggers each. Any trigger that fires on something that already happened (a cancellation, 30 days of silence) is a candidate for a prediction instead.
- Pick one journey and add a suppression branch for customers predicted to act without it. Measure what you saved.
- Ask your ESP vendor how a score field gets in and how often it can refresh. Daily is the bar.
4. Churn work moves from monthly reports to weeks-ahead early warnings
The shift. The retention team’s main artifact stops being the monthly churn report and becomes a weekly ranked list of customers predicted to leave in the next 60 days, with the reason next to each name.
The signal. Recurly’s July 2026 benchmark across its subscription network puts monthly churn at 3.27% on average, and 4.25% for ecommerce subscriptions, with 1.25 points of that involuntary. Compound 4.25% a month and you’ve lost about 41% of the base in a year. A monthly report tells you that after it happened, twelve times. The brands moving to weekly early warnings are the ones that get to do something about it.
What to do now.
- Define churn once per business line and write it down, including the inactivity window for businesses where nobody cancels.
- Rebuild your churn report by acquisition cohort. If growth has been hiding a worsening leak, this is where it shows.
- Decide what your team can act on each week (calls, offers, journeys) before you build a model, so the threshold matches your capacity.
5. Discounts get rationed by uplift instead of blasted by segment
The shift. Promo budgets get allocated by who the discount would change, and the brands that keep sending 20% off to whole segments will be funding their most loyal customers’ margin for nothing.
The signal. Kantar and Les Binet’s October 2026 study of consumer-goods marketers found that 75% of marketers cannot estimate what proportion of their promotions are profitable, even though promotions take 36% of marketing and trade spend against 17% for paid media. Uplift modeling answers that question customer by customer. Once a brand can see which quarter of a discount list the coupon moved, the other three quarters stop getting it.
What to do now.
- Add a 10% holdout to your next promotional send. You’ll have uplift training data in one cycle and a real incrementality number for the first time.
- Pull the list of customers who redeemed your last retention offer and check how many showed any risk signal beforehand. The share that didn’t is your discount waste.
- Create a “no promo” tier for customers predicted to buy at full price in the next 30 days, and suppress them from every discount journey.
6. Static segments die; segments refresh daily from predicted behavior
The shift. The segment library gets replaced by a handful of predictive segments (“will churn in 60 days,” “will buy category X in 30,” “will respond to a reactivation offer”) that recompute every night for every customer.
The signal. Shoppers have stopped waiting for the Tuesday send. In Amperity’s January 2026 survey of 1,000 US consumers, 69% said they’re more likely to buy when retailers adjust offers instantly while they browse, and 57% said shopping experiences still feel generic. A segment rebuilt quarterly can’t adjust anything in the moment. A score recomputed nightly from predicted behavior can.
What to do now.
- Count your active segments and the date each was last edited. Anything older than a quarter is a photograph.
- Pick the three segments that drive the most spend and define the predicted behavior each one is meant to capture. That’s the model spec.
- Set an expiry rule: any segment not refreshed from live data in 30 days gets archived.
7. AI agents build the models, analysts own the questions
The shift. In 2027, the scarce skill on a growth team stops being model building and becomes question framing: what to predict, over what window, for which decision, measured how.
The signal. Gartner expects 40% of enterprise applications to feature task-specific AI agents by the end of 2026, up from less than 5% in 2025. Model building is exactly the kind of bounded, repeatable task those agents absorb first: data preparation, feature engineering, training, validation. The people don’t go away. Their job moves from the pipeline to the question, which was always the valuable part.
What to do now.
- Write down the five predictions your team would act on tomorrow if they existed, each with its window and its action. Most teams find that four of them share the same data.
- Move your analysts’ time from building and maintaining to defining labels, checking drivers and designing holdouts. That’s where their judgment compounds.
- Pick one prediction and let an agent build it on your historical data as it is, without a cleanup project first. Judge it on the holdout, not the demo.
8. First-party data becomes the moat as signal loss keeps growing
The shift. The brands that win 2027’s auctions and journeys are the ones whose own order, session and subscription history is complete enough to train on, because the borrowed signals keep getting thinner.
The signal. Signal loss turned out to be a plateau, and the plateau is low. The industry-wide iOS App Tracking Transparency opt-in rate sits at 35%, flat for three years, so two-thirds of iPhone users are invisible to borrowed signals. Chrome kept third-party cookies in April 2025, then Google retired the Privacy Sandbox APIs that were supposed to replace them in October. Nothing new is coming to fill the gap. Your own order, session and subscription history already fills it.
What to do now.
- Inventory the first-party events you capture per customer (orders, sessions, subscription changes, payments, support contacts, campaign history) and where each one lives. Gaps in campaign history hurt most, because that’s what uplift learns from.
- Join the tables you have before you buy any new data. Most brands discover they’ve been sitting on the signal.
- Stop treating data consolidation as a prerequisite for prediction. Models train on real, imperfect history; the consolidation can follow the first use case.
9. Incrementality becomes the price of admission for every retention budget
The shift. By 2027, a retention program without a holdout doesn’t get funded, because finance has learned that “customers saved” counts people who were never leaving.
The signal. Incrementality has crossed from the measurement team’s hobby to the majority practice. An EMARKETER and TransUnion survey found 52.0% of US brand and agency marketers now use incrementality testing to measure their campaigns, with another 36.2% planning to invest in it within a year. Haus’s 2026 index of 500 senior decision-makers found 60% trust independent incrementality testing most, ahead of media mix modeling at 40% and in-platform reporting at 37%. Retention budgets are simply the last ones to be asked for a control group.
What to do now.
- Put a holdout on your largest retention journey this quarter: a random 10% of eligible customers who get nothing. Report lift, not save rate.
- Size the holdout for the decision you need to make, not for comfort. A 10% holdout on 200,000 customers is plenty; on 5,000 it isn’t.
- Write the incrementality result into next year’s budget request, as the number the whole program answers to.
Go deeper: incrementality.
10. Next best action leaves the bank and goes to every B2C brand
The shift. Per-customer decisioning, which banks and telcos have run for a decade, becomes standard for ecommerce, subscription, gaming and fintech, because the models that power it no longer need a data science team.
The signal. Twilio’s 2025 State of Customer Engagement report puts the demand side in one pair of numbers: 88% of consumers are more likely to buy when engagement is personalized in real time, and only 44% of brands say they’re executing at that level. Banks closed that gap years ago with decisioning layers that cost millions. The same per-customer decision now runs on a churn score, an uplift estimate and an arbitration rule a lifecycle team can write in an afternoon.
What to do now.
- Pick one decision you’re currently making with a segment and a blanket offer. “Who gets the retention discount” is the usual answer.
- Write the arbitration rule in plain language, including the option of doing nothing, before you build anything.
- Keep a global 10% holdout from day one so the program’s impact is never in doubt.
How to get ready
Three steps cover most of the distance, and none of them is a platform migration.
-
First, get your first-party history in one reachable place: orders, sessions, subscriptions, payments, support contacts and the campaign log of who got what. It doesn’t need to be tidy. It needs to be joinable, because every prediction above is trained on it, and the campaign log in particular is what uplift learns from.
-
Second, pick one prediction from this list and put it to work this quarter. Choose the one attached to the largest line in your budget, which for most B2C brands is retention offers or acquisition bidding. One model, one decision, one owner.
-
Third, set up the holdout before the model goes live. A random slice of eligible customers who get no action is the only thing that turns “we ran a program” into a number finance will fund again.
If you’d like the prediction part handled, that’s what we do. You bring your historical customer data and the question, the Predictive AI Agent reads that data as it is, trains and validates the model, and we own the outcome, with scores live in the CRM, ESP and ad platforms your team already uses.
See which of your customers will churn, buy or grow next:
Book a demo
