Seven ways to grow baskets, ranked by how long each keeps working.
Acquiring new customers keeps getting more expensive. Shopify put average ecommerce customer acquisition cost at $226.38 in 2024, a 7 percent rise in a single year, and the paid channels that deliver that traffic have not gotten cheaper since.1 The shoppers already on your site, meanwhile, are spending whatever your product discovery experience allows them to spend.
So teams go looking for ways to raise average order value (AOV), and nearly every list names the same seven things. You have probably read three versions this year, and you may already pay for five or six of them.
Most retailers who want bigger baskets have already licensed what they need to get them. Whether they see the 20 to 38 percent AOV increases Athos Commerce customers report, or nothing at all, comes down to a single question. Can those seven tactics see each other’s decisions, or does each one optimize against its own private copy of the product data?
When a merchandiser pins four products to the top of a category page, the recommendation carousel three rows down usually has no idea it happened. It promotes two of the same four, and the shopper sees one product twice on a single screen. When search suppresses an item because stock dropped overnight, the bundling logic on the product detail page often keeps pairing it, so shoppers build a basket that cannot ship complete.
Neither of those is a configuration mistake. Both happen because each tool holds its own copy of the product data and refreshes on its own schedule. Someone has to reconcile them by hand, and that someone is one of the two or three people who also run promotions, brief creative, and prepare for peak season.
Manual reconciliation ends most AOV programs. A tactic that someone has to correct every week works until attention moves to the next launch. A tactic that reads the same product data as everything around it does not need weekly correction, so its lift holds through peak season instead of decaying between launches. Most lists start with the tactics that need the most hand-holding, and this one ends with them.
On-site search needs the least maintenance of anything here, since it runs on every session untouched and gates everything below it. No tactic grows a basket around a product the shopper never found. MetroKitchen’s search resolves size variants without manual synonym management, so “12 inch,” “12in,” and “twelve-inch” all return the same products.
We’re amazed to see how much revenue can be driven by search. Even though about 90% of our shoppers prefer to browse our categories, almost half our revenue now comes from search.
Romane Vernet
Senior Designer/Developer, MetroKitchen
At ITS, average order value among search shoppers rose from $239 to $250 year over year, and Fabletics saw a 102 percent increase in revenue from search. Zero-results pages deserve separate attention. Shopper intent runs highest there, and the default page offers no next step. HealthPost places product recommendations on its zero-results pages, and 6.03 percent of those sessions convert.
Behavioral ranking reorders results using what a shopper has browsed and bought before. Real-time intent adaptation adjusts within the session as the shopper clicks, filters, and adds to cart. Personalized reranking keeps working because the model updates from shopper behavior without anyone scheduling it. Laura Mercier recorded a 38 percent increase in average order value alongside a 29 percent decrease in bounce rate from paid search. McKinsey puts typical personalization revenue lift at 5 to 15 percent across sectors, which makes the Laura Mercier result a strong outcome rather than a baseline expectation.2
Behavior-based and session-based recommendation carousels surface complementary and similar items on product detail, cart, and category pages, where a shopper is deciding what to add. One customer reported a 21 percent increase in average order value among shoppers who used search, against 3 percent for those who did not. Another customer reported a 20 percent AOV increase from search-driven sessions, and 1,264 percent higher revenue per visit from search compared with sessions without it.
Bundles work when the pairing reflects what genuinely goes together, which depends on the product data carrying enough attribute detail to know. HealthPost was an early adopter of product bundling and sees a 6.97 percent click-through rate on product-page bundles, which the team credits with incremental revenue and higher average order value. Bundling logic that reads live stock and margin signals avoids pairing an item that search has already suppressed for low stock.
A rules engine applies promotion and badging logic by inventory level, margin, or shopper segment, and keeps applying it without anyone reopening the tool. Early Settler moved merchandising processes that took up to 15 hours a week down to roughly three to five hours. Aje Collective reports two to three full days a week returned to the team through merchandising automation. Those recovered hours matter twice. The rules keep applying promotions and badges while no one is watching, and the time that comes back goes to the assortment and pricing calls that need a person.
Hand-built campaign pages carry the highest ceiling of any tactic here and the shortest useful life, since each one goes out of date as soon as the assortment changes. They pay off on a product launch or a seasonal peak, where the page is current by definition.
Being able to use merchandising campaigns and landing pages has been one of the top drivers of growth in our online business. As well as increasing sales, we’ve also seen improvements in our online engagement.
John Mueller
Ecommerce Manager, Hartville Hardware
Campaign pages stop working when they become the default for the entire catalog. A team hand-curating 40 category pages will keep six of them current, which is the manual merchandising ceiling in practice. The next tactic is how teams work above that ceiling. Curate the pages and rows that carry the campaign, and let automated ranking keep everything else current.
Seasalt Cornwall found that however carefully the team merchandised a category page, shoppers used filters and refined their own way regardless. So the team pins 16 products to keep the top four rows aligned with current marketing campaigns, and lets automated ranking order everything below. Human judgment goes where shoppers look and brand presentation matters, and the long tail stays current without anyone maintaining it. That split is the closest thing here to a rule you can apply to any category page.
Every tactic above improves when the others can see what it decided, and three handoffs do most of that work.
A pinned product tells the recommendation carousels what has already been promoted, so the shopper stops seeing the same item twice in one screen. A/B test results feed the ranking model rather than a slide deck, so each test becomes a permanent improvement instead of a finding someone has to remember to apply. Stock and margin signals reach promotions, bundling, and search suppression on the same cycle, so a product that sold out overnight leaves every placement at once.
Reaching that state is a consolidation problem. It does not require an eighth tactic. That is why Athos Commerce runs search and personalization on the same product data model as merchandising and product feed management. A decision made in any of them stays visible to the rest. That coordination is what the Intelligent Discovery Platform provides: one view of product data serving every place a shopper looks.
The same product data reaches offsite channels as well, so the attribute work improving your on-site recommendations also improves how products appear in email and SMS campaigns, Google Shopping, marketplaces, and AI assistants.
Athos Commerce provides Customer Success, Support, and Education at no charge, so the first two moves cost time only.
Most retailers already own five or six of these seven tactics. Getting more out of them is a question of whether they behave as one system, and that depends on whether they read a single view of the product data.
Seasalt Cornwall’s split is the practical version for most teams. Decide where human judgment genuinely changes the outcome, spend your hours there, and let automation keep the rest of the catalog current. Any team willing to consolidate the product data underneath their discovery tools can operate that way.
There are roughly six weeks before peak-season code freezes. Consolidation work done in that window keeps paying through Black Friday and Cyber Monday, while anything left fragmented puts someone on manual reconciliation duty across the two weeks when their attention is worth the most.
Average order value (AOV) is total revenue divided by the number of orders over a given period, so it measures how much a shopper spends per transaction. It matters most to retailers whose traffic costs keep climbing. Shopify put average ecommerce customer acquisition cost at $226.38 in 2024, a 7 percent rise in a year, which makes raising basket size from existing sessions cheaper than buying more visits. Athos Commerce customers report AOV increases between 20 and 38 percent from on-site product discovery work.
Average order value measures spend per completed order, while revenue per visit measures spend across all sessions, including those that produce no order at all. AOV can rise while total revenue falls, because a smaller number of larger orders lifts the average. Revenue per visit captures both basket size and how often a visit converts. Michael Stars recorded $32.92 revenue per visit from shoppers who used search against $3.13 from shoppers who did not, a difference average order value alone would hide.
Most AOV tactics stop working because they need manual reconciliation. In most stacks, product recommendations and bundling logic hold separate copies of the product data from merchandising pins and promotion rules. They refresh on separate schedules, so a person has to keep them consistent. That work lands on the same few people who also run promotions and prepare for peak season. When their attention moves, the tactic drifts. Tactics that read shared product data stay consistent without weekly correction.
Start with on-site search relevance and zero-results pages, since no other tactic can grow a basket around a product the shopper never found. Then baseline revenue per visit so you can see which tactics already earn and which sit dormant. Before adding anything new, test whether your existing tactics can see each other. Pin a product in a category, then check whether the recommendation carousel on that page reflects the pin.
Product bundles raise average order value when the pairing reflects what genuinely goes together, which depends on the product data carrying enough attribute detail to identify it. HealthPost sees a 6.97 percent click-through rate on product-page bundles and credits the capability with incremental revenue and higher AOV. Bundling logic that reads live stock and margin signals avoids pairing items that search has already suppressed, which is a common reason bundle programs underperform after launch.
McKinsey puts typical personalization revenue lift at 5 to 15 percent across sectors, with individual results varying widely by execution. Laura Mercier recorded a 38 percent increase in average order value alongside a 29 percent decrease in bounce rate from paid search, which sits well above the typical range. Personalized reranking tends to hold up over time because the ranking model updates from shopper behavior between merchandising cycles.
