Personalisation · Recommendations and merchandising
Show shoppers the right products as they browse.
Collection order and recommendations that follow each visit.
Akiso · +73% conversion
Mamaearth · shopping assistant
Sudathi · +25% conversionTrusted by 50+ brands


















Two shoppers
One store. Two shoppers.
Sample store. Switch shoppers to see the collection re-order.
Shopper A
Arrived from a saree reel, on a phone, in wedding season.
Shopper B
Searched “cotton kurta”, came back for a second look.
Only the order and pairings change. Sample store, Helium-made images.
1Matches her reel
2Her city's wedding bestseller
3Completes the look
4Often bought with sarees
5Festive alternative
6Lower: everyday wear
7Lower: everyday wear
8Not her intentHow it reads a visit
Three things known before it re-sorts.
Who the visitor is, what each product is, and what they are doing now.
Who arrived.
About 20 variables the moment they land, before the first click.
- City and city tier
- Device and network speed
- Signs of price sensitivity
- Source and time of day

What you sell.
Vision models read every product: how it looks, what it is for, what sets it apart.
- Construction and attributes
- Occasion and style intent
- What shoppers buy alongside it

What they do.
Every element on the page is mapped, so each tap and scroll carries meaning.
- Add to cart, and what came before it
- Collection and product tags
- Filters, sorts and searches

Results
Personalisation, measured.
Client work against control, and results across the platform.
Higher order values
Cross-sells and upsells matched on product attributes. Across the Helium platform.
More converted sessions
Products and content merchandised for each visit. Across the Helium platform.

Two audiences, one store.
Intent read in the first clicks. +73% conversion, AOV up 25%.
Read the case →
Sarees surfaced by intent.
+25% conversion on nearly half of traffic, with up to 5× lower retargeting CAC.
Read the case →
Store, catalogue and paid in sync.
+27% revenue per visit in 4 weeks.
Read the case →“We saw a clear improvement in conversions once personalization went live. The impact on revenue per visitor was noticeable within weeks.”
Festive kurtas
Curator AI ranks the collection by city, persona, margin and stock.
| # | Product | Score | Why | Move |
|---|---|---|---|---|
| 1 | Saffron silk kurta setKurtas | Mumbai · festive intent | ↑ 8 | |
| 2 | Ivory chikankari kurtaKurtas | High margin · in stock | ↑ 3 | |
| 3 | Banarasi zari sareeSarees | Wedding persona | ↑ 12 | |
| 4 | Indigo block-print kurtaKurtas | Sell-through 81% | — | |
| 5 | Rose organza sareeSarees | Low stock: pushed down | ↓ 6 |
The software
Rules in plain English. Staging first.
Collections ranked by sales, sessions, city and persona. 12 recommendation strategy templates, plus your own.
For your ads
The same signals, sent to ad platforms.
40+ live signals per visit, so ad budgets follow the shoppers likely to buy.
Intent your ads can use.
Each session gets an intent score. Audiences built from it go to Meta and Google.
- Browsing apart from buying
- Hesitation apart from readiness
- Fit with your category's demand

Behaviour
- Add-to-cart frequency
- Dwell time
- Product views
- Repeat visits
- Scroll depth
Context
- Device
- Location
- Recency of visit
- Time of day
Intent
- Brand affinity
- Discount affinity
- Discovery behaviour
- Hesitation patterns
“We hit ROAS ~ 10 on our retargeting ads in just 2 days. I'm still not sure how it works but it's been 3 months and we've already achieved our yearly growth goals.”
Ways to buy
Run it yourself, or let us.
Run it yourself
Free to start
The Helium app. Your team configures it.
- 7-day trial on paid plans
- Install from the Shopify App Store
- Paid plans from $100/month
Managed
Quoted monthly after scope
We run it and report monthly. You approve every change.
- Monthly readout against control
- Read-only access
- Strategies and rankings set for you
Questions
Before you ask.
Can I keep products on top?
Yes. Pin products; the rest re-sorts on schedule.
Do different cities see different orders?
Yes. Orderings are precomputed per city tier and persona.
Does every change go live straight away?
No. Staging first, then live after your approval.
Does it keep learning?
Yes. Every session feeds the model, so rankings improve over time.
How is revenue measured?
Against a control group.
How long does setup take?
The script installs in about two minutes. Personalisation starts after about 15 days of learning.
Software or managed?
Either. Run the app yourself, or we run it.
What about shopping assistants?
We build them too, like Mamaearth's assistant.
Will it slow my store?
We check page speed on staging first.
Book a call
Show shoppers what they came for.
Send the store. We reply with three placements we'd personalise.
Want to test us on content first? Get a free sample
