Case study
Luxury home and handcrafted rugs
Jaipur Rugs
Reducing cognitive load on a luxury craft marketplace without reducing its richness.
Continuous product work on a live e-commerce experience for a handcrafted rug house: navigation, designer discovery, filters, product pages, live chat and mobile. The research here is a framework built from working evidence rather than a formal study, and it is labelled that way throughout.
Three ways to read this. Pick the one that fits the time you have.
- Role
- Product and Brand Designer
- Duration
- Jaipur
- Team
- Merchandising, retail, marketing and engineering
- Scope
- NavigationDesigner discoveryFiltersProduct pagesRoom viewerLive chatStore locatorCampaign and editorialMobile responsiveSEO-aware structure
In four lines
The product
Handcrafted rugs as design objects, supported by decor categories including paintings, Pichwai paintings and tapestries.
The central problem
The catalogue is rich, and richness becomes friction. Too many collections, too many attributes, long designer journeys, mobile carrying desktop complexity.
How I worked
Not a formal research study. Cross-functional knowledge, support conversations, search data, analytics, behavioural tools and merchant expertise, applied continuously to a live product.
Evidence status
Read this one as a framework. Most rows below are hypotheses awaiting validation, and each one says which it is.
01 / 10Overview
01 · Overview
A showroom that has to work as a shop
Where craft storytelling and commerce mechanics meet, and why that is harder than either alone.
Jaipur Rugs sits at the intersection of handcrafted product, luxury home design, artisan storytelling and commerce. The company's public materials put craft, artisans and social impact at the centre of the brand, and the Foundation's Manchaha initiative goes further: artisans become the designers, working with leftover material, since 2010.
For an e-commerce product designer that creates a specific tension. The site has to support discovery and purchase of a high-consideration item, while carrying the emotional and cultural weight that justifies the price. Lean too far toward commerce and it becomes an anonymous catalogue. Lean too far toward story and the customer cannot find a rug.
So the opportunity was never to make rug shopping easier. It was to make the customer feel more informed, inspired and confident, without flattening the premium character of the brand.
Which reduces to one line: reduce cognitive load without reducing richness.
The customer should experience the richness of the brand without having to process the complexity of the organisation behind it.
Content needed
Sources for the company's public figures: 40,000+ artisans, 600+ villages, 45+ countries, 34+ awards.
These come from Jaipur Rugs' own trade site and are quoted in my research document, but every citation in that document is a placeholder token rather than a URL, so I cannot point at the source. They are also company scale figures rather than outcomes of my work, so they carry no weight as evidence for anything on this page. Included for context, flagged rather than asserted.
02 · Approach
How the research actually happened
Not a study with a sample size. A live product, watched continuously from several angles.
The honest version of this chapter matters more than an impressive one. I did not run a formal research programme with recruited participants and a documented method. What I had was better suited to a live commercial product and worse suited to a tidy narrative: a set of evidence streams that each showed a different face of the same customer.
Cross-functional knowledge from merchandising and retail, who field customer questions constantly. Support and live-chat conversations. Search behaviour. Product analytics and session recordings. Product constraints from engineering. And continuous observation of a site that was always live and always changing.
Merchant and retail teams
OngoingThe strongest qualitative source available. These teams meet the same customer questions week after week and know the assortment commercially, not just structurally.
Live chat
OngoingEvery conversation is a customer telling you what the site failed to answer. Treated as a data source rather than a queue of tickets.
Search and SEO data
AvailableWhat people typed to arrive, and what they searched for once inside. Language the customer chose, not language the business chose.
Analytics and recordings
AvailableWhere people went, where they stopped, where they hesitated. Useful for locating a problem, weak at explaining it.
Product constraints
OngoingWhat could actually be built, surfaced early enough to shape the design rather than to invalidate it late.
Continuous observation
OngoingA live site with real traffic is an instrument. Watching it over time replaces a snapshot study with a moving picture.
That combination changed the unit of work: from designing pages to improving a product continuously.
Content needed
The actual evidence behind every claim in this chapter: Search Console queries, GA4 journey data, Clarity recordings and heatmaps, and the live-chat export.
I named the streams because I genuinely used them. I am not quoting a single figure from any of them here, because I do not have the exports in front of me, and a number I cannot produce on request is worse than no number.
03 · Journey
A confidence journey, not a funnel
For a purchase this considered, the useful model is how sure the customer feels, not how far down they are.
What is here, and where do I start?
The customer may arrive through a campaign, a designer name, a search result or the homepage, often without a specific product in mind. Discovery needs several front doors rather than one.
- Discovery
- Orientation
The job at every stage is the same: reduce uncertainty without removing the sense of discovery.
Those pull against each other. Strip out enough ambiguity and you also strip out browsing, which for this category is most of the pleasure and a good deal of the basket.
04 · Customer
Who is actually shopping
One primary model and five secondary ones, all working, none formally recruited.
The considered luxury-home shopper
Persona
Browses before deciding, often arriving without a specific product in mind, leaning heavily on visual inspiration and needing real confidence about size and fit before spending.
- Behaviour
- Browses, compares, returns
- Sensitivity
- Quality and perceived value
- Support
- May need a human before buying
- Uses inspiration heavily
- May care about designer identity
- Needs confidence in size
- Compares multiple options
- Expects the promise to outlast checkout
The tensionShe wants to browse and she wants to be sure, and those need different interfaces. The same session swings between wandering and scrutinising, often several times, and a site tuned for only one of the two loses her at the other.
The question she changed
Before this persona, the team asked: What does this customer want to buy? After it, the team asked: What does this customer need to feel before she is willing to buy it?
A working persona built for this project from cross-functional knowledge and support conversations. It is not the output of a recruited persona study, and it should not be read as one. The device-split assumption in particular, mobile for discovery and desktop for evaluation, is plausible and unverified: it needs checking against analytics before anyone designs against it.
- 01
Designer-led
Arrives because a designer's name means something to them. Wants that designer's work, quickly, without a detour through general browsing.
- 02
Inspiration-led
Comes in through a campaign, social content or editorial imagery. May not know the category or the attributes yet.
- 03
Specification-led
Already knows roughly the size, colour, material or budget. Wants efficient filtering and comparison and nothing in the way.
- 04
Assisted
Needs help, and reaches for live chat or support to resolve uncertainty the site did not.
- 05
Returning
Comes back for a new collection, another room or a different designer, and is served by the post-purchase and relationship experience.
- 06
What this implies
Five different first moves. A single funnel optimised for one of them quietly fails the other four, which is the argument for multiple entry points rather than one hero path.
05 · Load
Richness becomes friction
The central problem, and the nine places it comes from.
Too many collections
Each meaningful to the business, and collectively a wall to a first-time visitor.
Many product attributes
Every one useful to someone, and all of them present at once.
Long designer journeys
Several steps between recognising a name and reaching that person's work.
Competing visual surfaces
Large imagery everywhere means nothing is emphasised.
Complex navigation
Structure that reflects the organisation rather than the shopper.
Filters that miss the mental model
Built from the database schema instead of the customer's own language.
Unclear size and room context
The hardest thing to judge, and the most expensive to get wrong.
Story before the next action
Craft narrative is an asset until it stands between the customer and the product.
Mobile carrying desktop complexity
The same density on a tenth of the screen.
Filters are the clearest case. The best filter system is not the most detailed one. It is the one that reduces the search space in the customer's own language.
06 · Signals
What the session data actually shows
Twelve months of Microsoft Clarity on the live site. It validates some of the framework, contradicts a piece of it, and surfaces two problems the framework never mentioned.
Everything above this chapter was reasoning. This chapter is measurement: a Clarity export covering 1 June 2025 to 31 May 2026 on jaipurrugs.com. It changes the status of several rows in the matrix that follows.
One caveat has to come first, because it reframes every percentage in the file. Almost exactly half the recorded sessions are bots: 588,720 of 1,177,890, or 49.98%. Clarity computes its session percentages over the full total, so a metric reported as 0.36% of sessions is 0.73% of human ones. Where the distinction matters below, both numbers are given. Anyone quoting a single figure from this dashboard without saying which denominator they used is quoting a number that is roughly half wrong.
589170 Human sessions in twelve months
Clarity: 1,177,890 total minus 588,720 bot sessions, 1 June 2025 to 31 May 2026.
50 % Of recorded sessions were bots
Clarity bot sessions, 588,720 of 1,177,890. Reported separately from the headline total.
64 % Mobile and in-app browsers
Sum of Clarity browser shares: Chrome Mobile 29.87, Mobile Safari 19.47, Instagram 7.06, Google App 4.24 and six smaller mobile browsers.
25 % Of sessions hit a JavaScript error
Clarity: script error on 24.72% of sessions, 84,407 in total.
47 % Average scroll depth
Clarity scroll depth average, 46.9%. Most visitors never reach the lower half of a page.
2.5 Pages per session
Clarity average, 2.505. Shallow for a browse-led category.
Finding one
The carousels are broken for a quarter of visitors
The single largest thing in this export is not a design problem, and the research framework never anticipated it. Clarity logged 341,418 JavaScript errors. The top entries name the cause directly: `$(...).slick is not a function` on 15.55% of sessions, and `$(".lazy").slick is not a function` on a further 10.08%.
Slick is the jQuery carousel library. When it fails to initialise, carousels do not become static, they become unstyled: every slide stacks, lazy-loaded images do not swap in, and the layout collapses. On a site whose product galleries and homepage merchandising are built from sliders, roughly a quarter of visitors are seeing a fundamentally different page from the one that was designed.
This also explains the second number nobody would design for. Cumulative Layout Shift measures 2.46, where anything above 0.1 is considered poor. That is twenty five times the threshold, and it is what a page does when image containers have no reserved height and a carousel script dies partway through boot. Largest Contentful Paint of 5.1 seconds is roughly twice the 2.5 second target, and the overall performance score is 52 out of 100.
Which reorders the whole backlog. No amount of filter design helps a visitor whose gallery never rendered.
I had been treating cognitive load as the central problem. For a quarter of sessions the problem is more basic than that, and it is an engineering fix rather than a design one.
Finding two
The store locator is the second most visited page
Top pages by sessions
- Home156488
- Stores38542
- Clearance29308
- Rugs29074
- Search27583
- New arrivals21627
- Summer rugs21188
- Bestsellers18158
Clarity top pages, twelve months, India storefront unless noted. This is a count of sessions reaching each page, not a performance measure. Note the second row: store discovery outranks the rugs listing and the search page.
The store locator drew 38,542 sessions, more than the rugs listing at 29,074 and the search page at 27,583. For a business selling a high-value tactile product that is not surprising in hindsight: people want to see and touch a rug before spending. It was surprising to me, because the research framework filed store locator last of ten areas.
The interaction data makes it sharper. The Find a location event fired in 884 sessions. That is one interaction for every 44 visits to the page. Either the page answers the question without needing interaction, which would be fine, or people arrive wanting to search and find something static, which is what the framework guessed and this now supports.
Finding three
One session in five shows an interaction failing
Dead clicks, 9.84%
Clarity115,846 sessions where someone clicked something that did nothing. The clearest possible signal that an element looks interactive and is not.
Quick back clicks, 10.44%
Clarity122,933 sessions where someone arrived on a page and immediately reversed out. Wrong page, or the page did not look like what was promised.
Rage clicks, 0.21%
Clarity2,474 sessions. Notably low, which matters: people are not fighting the interface in frustration, they are quietly finding it inert and leaving.
Outbound clicks, 26.98%
Clarity317,789 sessions include a click leaving the site. Over a quarter. Worth understanding before anything else on this list is optimised.
Active time, 78s of 316s
ClarityA quarter of time on site is active. The rest is idle, which for a browse-led category may be reading, or may be waiting.
Mobile, 63.7%
ClarityOf which 8.4% arrives inside Instagram and Facebook in-app browsers, the most constrained rendering environment on the list.
The mobile figure settles an open question in the persona. Nearly two thirds of sessions are mobile, and social referrals bring roughly 122,000 of them.
That supports the inspiration-led shopper as a primary rather than secondary model. It does not confirm the device-split assumption, that people discover on mobile and evaluate on desktop, which needs cross-device analysis this export cannot provide.
Finding four
The chat data contradicts my own framework
My research document treats live chat as a rich source of qualitative behavioural data, and the matrix below builds a whole row on it. This export does not support that. Clarity recorded chat as available in 515 sessions, 0.04% of the total, with 448 engaged users, an average conversation duration of zero seconds and zero complete turns.
I am not going to reconcile that in my own favour. Either Clarity was measuring a different chat product from the one the support team actually used, or the integration was not configured, or the live-chat work sits outside this twelve month window. All three are plausible and I cannot tell which from this file alone. Until that is resolved, the chat row in the matrix is a row I would not defend in an interview.
Content needed
Reconciliation between the Clarity chat metrics and the actual live-chat platform export, plus the funnel discrepancy: Clarity reports 401 purchase events but its own funnel reports 120 conversions from 4,162 add-to-carts.
Two internal inconsistencies in one dashboard. Neither is fatal and both need explaining before any number here goes in front of a hiring manager, because the first person to check will find them.
07 · Matrix
Signal, response, and how far it got
Ten areas of the product, each with its evidence status stated rather than implied.
- 1 shipped
- 2 designed
- 8 hypothesis
- Signal
- Desktop complexity carried onto a small screen
- Where it came from
- My own work on the live site. Clarity now quantifies the need: 63.7% of sessions are mobile or in-app, including 8.4% inside Instagram and Facebook browsers
- Design response
- Responsive interaction redesign and a reworked mobile navigation
- How it would be validated
- Mobile analytics and session recordings
- Signal
- Choice overload across a large assortment
- Where it came from
- Collaboration with merchandising and retail teams
- Design response
- A filter model built from merchant and retail knowledge rather than the database schema
- How it would be validated
- Filter usage data and chat questions about attributes
- Signal
- A long journey between recognising a designer and reaching their work
- Where it came from
- Opportunity I identified while working on the live site
- Design response
- Collapse it to a direct designer to products route, roughly a two page experience
- How it would be validated
- Journey analytics and task-based testing
- Signal
- Carousels failing to initialise for roughly a quarter of visitors
- Where it came from
- Clarity: 341,418 JavaScript errors. Slick failures on 25.6% of sessions across two variants, CLS 2.46 against a 0.1 threshold, LCP 5.1s, performance score 52. This row did not exist in my research framework at all
- Design response
- Fix the slick and lazy-load failures, and reserve image dimensions to stop layout shift
- How it would be validated
- Error rate, Cumulative Layout Shift and Largest Contentful Paint after the fix
- Signal
- The same questions recurring, each one a gap in the site
- Where it came from
- CONTRADICTED by the Clarity export, which records chat in 515 sessions with zero complete turns. Downgraded from designed until the chat platform export explains the gap
- Design response
- Treat chat as a structured insight repository rather than isolated tickets, taxonomised by question type
- How it would be validated
- Chat data, with recurring questions promoted into FAQ, filters or product page content
- Signal
- Discovery friction and unclear grouping
- Where it came from
- Signal now measured: dead clicks in 9.84% of sessions, quick back clicks in 10.44%, average scroll depth 46.9%. The response is still untested
- Design response
- Simpler grouping with visual cues, and separate entry points per intent
- How it would be validated
- GA4 navigation events, Clarity recordings, usability testing
- Signal
- Purchase uncertainty on a high-value item
- Where it came from
- Reach is measured, 28.6% of sessions hit a product page. The uncertainty itself is still inferred, and the chat evidence it relied on is now in doubt
- Design response
- Improve hierarchy and answer the high-value questions in context, before the customer leaves to ask
- How it would be validated
- Product page behaviour and the questions arriving through chat
- Signal
- Spatial uncertainty: rugs are judged at scale, against furniture and light
- Where it came from
- Category reasoning. My involvement in this surface needs confirming
- Design response
- Improve visualisation so an abstract product image becomes a spatial decision
- How it would be validated
- Tool engagement, and whether it reduces chat questions about fit
- Signal
- Competing objectives across brand, campaign and commerce
- Where it came from
- Signal partly measured: the homepage is the top page at 156,488 sessions, and 26.98% of all sessions include an outbound click
- Design response
- A curated editorial hierarchy with an obvious route into relevant products
- How it would be validated
- Engagement, click-through and conversion
- Signal
- Order uncertainty after a large purchase
- Where it came from
- Reasoning about the category. No data gathered
- Design response
- A clear order status experience, so the premium relationship survives checkout
- How it would be validated
- Support contact volume for tracking and status
- Signal
- Store discovery is a major, under-served need
- Where it came from
- Signal strongly measured: the store locator is the second most visited page at 38,542 sessions, ahead of the rugs listing, yet Find a location fires in only 884. One interaction per 44 visits
- Design response
- Interactive search connecting digital browsing to the showroom
- How it would be validated
- Search usage on the locator
A row marked hypothesis means exactly that: a reasoned proposal, not a validated finding and not a shipped change. Where my source document was ambiguous about whether something shipped or was only proposed, I marked it down rather than up. Note what the Clarity export did to this table. It put measured signals behind four rows that previously had only observation, it added a front-end reliability row the framework never contained, and it forced the live chat row down from designed to hypothesis by contradicting the evidence that row was built on. That is what evidence is supposed to do, and a matrix where new data only ever promotes rows is not being read honestly.
08 · Questions
What I would need to know
The questions this work is still waiting on, kept as questions.
- 01
How do customers discover a rug when they do not know what they want?
The inspiration-led path is the one least served by conventional filtering, and the one this category depends on most.
- 02
What information is genuinely necessary before buying?
Not what is available. What actually changes the decision, and in what order.
- 03
Do customers use designer names as a discovery route?
The designer flow work assumes yes. That assumption is worth checking before more is built on it.
- 04
Which filters do customers reach for first?
Size, colour, room, style or price. The answer determines the order and the defaults.
- 05
Does the room viewer change confidence?
Engagement is easy to measure and not the question. Whether it reduces hesitation is.
- 06
Which questions arrive through chat most often?
The recurring ones are a specification for missing site content, written by customers.
- 07
Where does the mobile journey actually diverge?
The device split in the persona is assumed. It needs analytics behind it.
- 08
How much craft story is enough?
Storytelling is the brand's advantage right up until it delays the next action. The threshold is unknown.
The synthesis that holds all of it together: the best experience is not the one with the most information. It is the one that reveals the right information at the moment the customer needs it.
09 · Evidence
What is missing, stated plainly
The gaps between this framework and a case study I would defend line by line.
Asset needed
Every screen. There are no exported designs for this project: no navigation before and after, no designer journey, no product page audit, no filter taxonomy, no room viewer flow, no mobile audit, no Figma exploration frames.
This is the largest gap on the page and the reason this case study reads as an argument rather than a portfolio piece. Nothing has been borrowed from the sibling Jaipur Rugs group brands to fill it, which would be the easy and dishonest fix.
Content needed
The remaining quantitative evidence: GA4 journey data, Search Console queries and landing pages, the live-chat platform export, and campaign analytics.
The Clarity dashboard is now attached and did most of the work in the Signals chapter. What it cannot answer is anything cross-device, anything about search intent before arrival, and anything about what customers actually asked. The chat export is the most urgent of the three, because Clarity has cast doubt on a row that was built on it.
Content needed
Documented competitor benchmark sessions across homepage merchandising, navigation, search, filtering, designer discovery, product page hierarchy, room visualisation, craft transparency, assistance, post-purchase and mobile.
The benchmark dimensions are defined and the sessions are not recorded. Naming competitors I have not formally reviewed would be inventing research, so this page names none.
The rule I am holding this page to, taken from my own research notes: never label a design assumption as a user insight.
If it came from observation, call it an observation. If it came from analytics, cite the metric. If it came from an interview, give the sample and the method. This page has observations and reasoning, and it says so.
10 · Reflection
What this project taught me
Mostly about working on something live, and about the difference between evidence and confidence.
The most useful shift here was from page-by-page design toward continuous product improvement. On a live commercial site with real traffic, the question stops being what should this page look like and becomes what is this product failing to answer this week. That reframing came from having several evidence streams running at once rather than one study concluding.
It also taught me where that approach is weak. Continuous observation is excellent at locating problems and poor at proving solutions. I can tell you with some confidence where customers hesitated. I cannot yet tell you that any change I made fixed it, because the validation column of that matrix is still empty.
Writing this case study honestly was the more uncomfortable lesson. It would have been easy to present the hypotheses as findings, invent a percentage or two, and produce something far more impressive than what you have just read. That version would collapse under the first follow-up question in an interview, and it would be worth less than a page that says clearly which parts are known and which are argued.
The central opportunity, unchanged from the first week to the last: let customers experience the craft, design and luxury of Jaipur Rugs while making discovery, evaluation, purchase and everything after feel effortless.
Outcome
What changed, stated honestly.
- 01
Twelve months of Clarity session data analysed and reconciled against the research framework, promoting four rows to measured signals, adding one the framework had missed and downgrading one it had overstated
- 02
A responsive redesign of the mobile experience and mobile navigation on a live luxury e-commerce site
- 03
Live chat established as a structured source of customer insight rather than a support queue
- 04
A filter model informed by merchandising and retail knowledge instead of the database schema
- 05
A designer discovery route collapsed from a long journey to roughly a two page experience
- 06
A working model of the customer as a confidence journey rather than a funnel, with five distinct entry intents
- 07
Cognitive load named as the central product problem, with nine specific sources identified
- 08
A research to design matrix covering ten product areas, each carrying its evidence status
On metrics
These are design outcomes and working artefacts, not results. The Clarity figures throughout describe the state of the live site over twelve months; they are diagnostic evidence for the problems, and none of them measures the effect of anything I changed. I have no verified figures for conversion, engagement, task completion, support deflection or revenue attributable to any of this work, and no exported screens to show. The statuses in the matrix above need confirming against my project record: where my source document was ambiguous about whether something shipped or was proposed, I marked it down rather than up. This case study is a framework built from real working evidence, and it should be read as one until the analytics, chat exports and design files are attached to it.
The best experience is not the one with the most information. It is the one that reveals the right information at the moment the customer needs it.