Hydro Flask US · Digital Analytics

What removing the PLP filters cost us

The Bottles listing page lost a quarter of its shoppers at exactly one step — the step filters exist to serve.

Page analysed
/shop/bottles
Change date
~11–12 Aug 2026
Before window
22 Jul – 11 Aug 2026
After window
12 – 25 Aug 2026
Sources
Adobe Analytics · Magento DB
Status
Directional
−25.9% Fewer visitors reach a product page 1 in 4 shoppers who land on Bottles no longer get as far as opening a product.
77% Leave without opening a single product Up from 69% before the filters were removed.
−24.3% Drop in daily orders from this page From 142.9 to 108.1 orders per day, while the site overall grew.
01

The short version

We removed the left-hand filter panel from the Bottles product listing page. Since then, that page has become materially worse at turning visitors into shoppers.

The damage is concentrated at one specific step: helping a shopper find a product worth looking at. Once a shopper reaches a product page, they convert better than before. Our products, our pricing and our checkout are not the problem. Product discovery is.

Gross annualised revenue at risk: ~$633,000

On this single page, holding traffic constant so a seasonal dip is not counted as damage.

Read this honestly

Total site revenue rose 6.4% over the same period. This is not a visible hole in the P&L — demand appears to have rerouted through other paths. The cost is opportunity and fragility, not money already gone. Section 06 gives the defensible range.

02

What actually changed

The Bottles page used to carry a filter panel down the left-hand side — size, colour, lid type, mouth size, price. A shopper looking for a 32 oz wide-mouth bottle in blue could narrow 83 products down to a handful in two clicks.

That panel was removed. The page now shows sub-category tiles at the top, then a flat grid of 83 products — 48 of which load before the shopper has to paginate. There is no filter, and no sort control either.

The shopper now has to do the matching work manually.

03

Where the funnel broke

This is the whole argument in one picture. Two steps got worse. The middle step got better.

Before — filters present After — worse After — better
PLP → Product ViewNEXT HIT −7.94pp  /  −25.9%
Before
30.60%
After
22.66%
Product View → OrderEVENTUAL PATH +0.96pp  /  +8.8%
Before
10.82%
After
11.78%
PLP → OrderOVERALL −0.641pp  /  −19.4%
Before
3.311%
After
2.670%

Each step is scaled against its own maximum so the change is readable; bars are not comparable between steps. Metric is unique visitors, Adobe Analytics Fallout.

Why the middle step matters most

Product View → Order improved 8.8%. Shoppers who reach a product still buy — slightly more readily than before. That single fact rules out pricing, product, photography and checkout as explanations, and isolates the damage to product discovery.

04

Against a growing business

Context matters. The site was not having a bad month — it was having a good one. The Bottles page shrank against a rising tide.

Metric — matched 14-day windowsBeforeAfterChange
Sitewide orders per day986.61,014.5+2.8%
Sitewide revenue per day$65,805$69,998+6.4%
Average order value$66.70$69.00+3.4%
Average units per order2.152.36+9.8%
Orders from Bottles PLP per day142.9108.1−24.3%

That is roughly 30 points of relative underperformance on our highest-traffic listing page. Its share of total orders fell from about 15% to about 10.7%.

05

Why filters drive conversion

Filters are not a convenience feature. They do three commercial jobs:

They shorten the path to relevance

A shopper who wants a specific size and colour can say so in two clicks. Without filters they must visually scan 83 products. Most will not.

They let the shopper state their intent

A filter panel is how a customer tells us what they came for. Remove it and we are asking a motivated buyer to do our merchandising work for them.

They prevent the too-much-choice exit

When the effort of choosing exceeds the motivation to buy, shoppers leave. Eighty-three products, no sort, no filter, pagination after 48 — the 77% who now leave without opening a product are the measurable result.

The sub-category tiles we kept are a navigation aid, not a filtering aid. They serve a shopper who wants “Mini bottles”. They do nothing for a shopper who wants “32 oz, wide mouth, blue” — which is how most repeat customers shop.

06

Revenue at risk

We hold traffic constant and isolate the conversion effect, so a seasonal traffic dip is not counted as damage.

After-period PLP visitors 4,051 / day Expected orders at before-CVR 4,051 × 3.311% = 134.1 / day Actual orders 108.1 / day Shortfall 26.0 / day Annualised 26.0 × 365 = 9,485 orders × AOV $66.70 = $632,675

Unit economics

Every 1,000 visitors to the Bottles PLP now produce 6.4 fewer orders, worth $428.

A range, not a single number

This figure is an upper bound: some lost orders were clearly recovered through other paths — sitewide revenue rose over the same period.

If the true conversion loss is…100% truly lost50% recovered75% recovered
−19.4% — as measured$633k$316k$158k
−14.5% — if the effect is smaller$475k$237k$119k
−9.7% — if the effect is smaller$316k$158k$79k

How to present this

A $150k–$633k annual opportunity on this page, most likely in the $250k–$450k band. Do not present $633k as money already lost — total revenue is up, and that claim will not survive scrutiny.

07

What we do not yet know

There is no control page

Sitewide trend is a weak control — it includes checkout and homepage effects. /shop/bottles-drinkware/coffee-tea still has filters and is the natural comparison. Running the same fallout on it turns “a drop happened” into “the change caused the drop”.

We do not know where the volume went

If those shoppers now reach the same products via search, the real cost is small. If they leave the site, it is large. Adobe’s next-page flow answers this — and it is the difference between annoying and urgent.

08

This is one page of many

The analysis by Alex Weisbecker covers /shop/bottles only. A review of the remaining categories has been requested and is outstanding.

That matters, because a live audit found 15 of 25 sampled listing pages have no left-side filters — and they are primary navigation destinations, not minor corners.

No filters — exposed

/shop/bottles/shop/bottles-drinkware/cups-tumblers /shop/accessories/shop/coolers/coolers/bags-packs /coolers/insulated-totes/coolers/lunch-boxes/coolers/soft-coolers /shop/kitchenware/kitchenware/food-jars/shop/bundles /lightweight-water-bottles/mini-water-bottles/accessories/boots

Filters still present

/coffee-tea/beer-wine-spirits/kids-collection /new-arrivals/sale/caps-and-lids /cleaning/bottle-slings

Do not multiply $633k by fifteen

Traffic and shopping behaviour vary enormously by category. But if even a portion of that traffic behaves like Bottles, total exposure is materially larger than this single-page analysis suggests. Each affected category needs its own fallout report before any aggregate number is quoted.

09

The technical cause

The filter panel is not broken. Algolia is healthy on these pages — the search index returns correct results and all nine facets build with live product counts. The panel is rendered, then hidden by a single stylesheet rule that applies to any category using the Segmented Category page layout. It also hides the sort dropdown and the mobile “Filter” button.

Overriding that one rule restores the full filter and sort interface immediately.

Restoring attribute filtering is already scoped as HOHFUS-4401, currently P4 – Low, not started. Its dependency HOHFUS-3428 is already resolved. The recorded design intent was for visual filters and attribute filters to coexist — the second half was never built.

This analysis is the case for re-prioritising HOHFUS-4401.

10

Recommendation

  1. Close the two validation gaps in section 07 — roughly one day of analyst time.
  2. Extend the analysis to the other affected categories listed in section 08.
  3. Re-prioritise HOHFUS-4401 off P4 on the strength of the corrected numbers.
  4. Ship it as an A/B test on /shop/bottles rather than a blind rollout. That converts an argument into a measured result, and gives a defensible figure for every remaining category.