Most trend pieces hand you a list and make every item on it look equally urgent. For an online store the real question is not which trends exist, it is which of them will pay back in your catalog on your budget. This piece starts with a filter for making that call, then takes the shifts that have proved durable one at a time, and ends with three things nobody calls a trend that work in almost every store.
1. Three questions that tell you whether a trend is yours
Answer three questions before trying anything new. If all three are not yes, that trend is early for you.
- Can you measure the result? An investment you cannot measure cannot be repeated even when it works. That is why this is the first question.
- Is your current infrastructure ready for it? Building personalisation on incomplete product data is building a floor with nothing under it.
- What are you giving up? Every new initiative is time taken from something that already works. A project started without writing that down usually damages both.
2. AI-driven personalisation
Personalisation has been discussed for years, and in practice the same word covers two different things with very different costs.
- Rule-based personalisation: “show this banner to anyone browsing that category”. Cheap, legible, and in most stores it delivers the larger share of the gain.
- Model-based recommendation: systems that learn from behaviour. To work meaningfully they need both enough traffic and clean product data. On a small catalog the extra return rarely covers the cost.
In both cases the limiting factor is the data, not the algorithm. A product with the wrong category and empty attributes is a product no recommendation engine can place correctly.
3. Search no longer happens in one place
Product search has been drifting from the search engine to marketplaces for years. Generated answers and chat interfaces have now been added to that. The practical consequence: your product name and attributes are read for many independent surfaces rather than one.
- Every surface draws on the same data and weights different fields. A marketplace leans on the category, a shopping result on the title, and a generated answer on how complete the attributes are.
- Inconsistency is punished. When the product name is written one way on the site, another in the feed and a third on the marketplace, the system that has to pick a winner usually picks none.
- You cannot measure this traffic separately. Visitors arriving from a generated answer cannot currently be isolated in analytics, so budget directed here is a bet. The one thing you can measure is whether your data is machine-readable.
4. Short video is a creative factory, not a reach channel
The value of short video is in production speed far more than in any promise of organic reach. A brand that can shoot three videos a week has material to test in paid, and that is the real lever on performance.
- The first three seconds decide it, not the production value. An expensive shoot does not rescue a weak opening.
- Move what holds organically into paid. It is the most practical way to run creative testing for free.
- Show the product in the first second. The brand story comes second, because the viewer is already poised to scroll.
5. Measurement got harder, and that is permanent rather than a trend
Browser-based tracking loses more data every year. What that looks like from the outside is a widening gap between the conversions in the ad platform and the orders in the accounts.
- Server-side event sending is no longer optional. The loss varies by device and browser, so it also breaks comparison between channels.
- Verify deduplication. The same purchase sent from both the browser and the server is the most common reason a platform reads higher than the ledger.
- Stay on one attribution window. Comparing two periods measured on different windows is comparing two different questions.
6. Product data became infrastructure
A few years ago the feed was a technical file prepared for shopping ads. Today the same data feeds the organic product page, the shopping result, the marketplace listing, the catalog ad and the source behind a generated answer. Product data became a marketing asset, and in most teams it is still filed under IT.
What that means in practice: a product with a weak title, the wrong category or empty identifiers goes invisible on five surfaces at once, and no campaign setting fixes it. To see where your catalog stands, use the free feed analysis: more than 30 checks run per product, the catalog is graded A to F, and your source feed is never touched.
7. As acquisition gets dearer, retention moves up
As ad costs rise, the value of the second order rises with them. Despite that, retention work at most brands is one weekly newsletter.
- Segmentation matters more than list size. Sending everyone the same campaign is the fastest way to train your best customers to wait for a discount.
- Measure time to first repeat purchase. It degrades weeks before monthly revenue does, which makes it an early warning.
- Treat the post-order experience as marketing. Delivery time and how easy returns are affect the second order more than any campaign.
8. Three things nobody calls a trend that almost always work
- Reshooting the product images. It moves click-through directly in both organic and catalog ads, and it is usually cheaper than opening a new channel.
- Reading the site search report. The top ten searches with no results is next month’s content plan and stock plan at once.
- Keeping the negative keyword list current. One of the rare pieces of work that saves money the day it is written.
What the three have in common is that none of them needs a new tool or a new channel. They are better use of assets you already own, which makes their return more predictable than any trend investment. Planning a quarter around a new channel before these three are done usually ends with both left half finished.
Frequently asked questions
How much budget should go to new approaches?
A percentage would mislead, but there is a useful rule: the budget for a new test should be large enough that the result is measurable and small enough that losing it does not disturb the plan. A test that fails either condition teaches you nothing or damages the core business when it goes wrong.
Does a small brand need to follow all of them?
No, and trying usually hurts. On a small team the right strategy is to do two channels well and deliberately postpone the rest. A team that opens six at once never gathers enough data to learn from any of them.
Where should I start?
By verifying measurement and looking at your product data. Both are preconditions for every trend on this list, and both can be done inside a week. If you want someone to set the two up together, see our consultancy side or write to us.
