10 Turtle

Your Science. Their Needs.One Smarter Granions Experience.

Bringing personalized skincare guidance, intelligent product discovery, and effortless shopping into your existing Granions website.

The Granions customer journey

From the first question to the right product

AI Skin Analysis, in motion

A 90-second walkthrough of the Skin Analysis studio, from the capture to the routine. Sound on.

One connected Granions experience.

“Why does my skin look uneven?” “This serum fits my skin, and I know why.” Purchase
The flagship experience

Their Skin. Your Science.A Routine Made Personal.

One selfie, three simple questions, and personalized skin insights, connecting customers with relevant Granions products and a clear reason behind every recommendation.

What we created

A guided facial assessment built into Granions.fr. Customers explore their visible skin concerns, receive an understandable visual report, learn about the relevant ingredients and discover suitable products from the Granions Dermatologie range.

From generic product browsing to a personalised starting point.

Visible skin observations in plain words: uneven tone, fine lines, texture, visible blemishes. Simple explanations of the results and the ingredients. A curated selection of appropriate Granions products with usage and suitability information. Optional follow-up assessments for customers who want to compare over time.

A customer should leave the analysis understanding their skincare options, not simply receiving technical scores.

The principles behind it

GRANIONS PRINCIPLE

Simplicity

Already central to the Granions identity, now guiding the digital experience

Complex product exploration becomes clear guidance, intuitive navigation and easy-to-understand recommendations.

HOW WE APPLY ITFive simple steps, a report that explains instead of scoring, and products presented with the reason they fit.
GRANIONS PRINCIPLE

Safety

Transparent information, responsible recommendations, consent and privacy

Cosmetic observations rather than medical diagnoses. Facial photographs handled with appropriate consent, privacy protection and data minimisation.

HOW WE APPLY ITCamera consent before the selfie; only measures validated for the website capture method; supplements suggested only after a suitability questionnaire, never from a selfie alone.
GRANIONS PRINCIPLE

Efficacy

Customer needs connected to intended uses, formulations and approved evidence

Every recommendation uses Granions-approved product rules, suitability information and accurate cosmetic claims.

HOW WE APPLY ITThe analysis connects visible concerns and questionnaire answers to approved products, with a clear explanation of why each product may be relevant.

How it works, step by step

Click any step, or swipe the panel for the next one.

The demo, step by step

Swipe for the next step

Swipe the demo or use the arrows for the next step. Click a step above to jump to it.

How it comes together

Start→Selfie→Analysis→Product recommendations→Shop

Everything happens within the Granions experience, without downloading a separate application.

More than a skin score. A clearer understanding of what comes next, and the Granions products that fit.

Product discovery

AI Discovery Booster GEO / AEO

How we make Granions visible in AI answers, before the customer ever arrives.

Competitor + Google research tells us what the Granions pages are missing. The GEO method tells us how to improve them.

The question behind the method

CUSTOMER QUESTION

“What is the difference between niacinamide and hyaluronic acid for skincare?”

The search journey increasingly begins with questions about ingredients, skin concerns and suitability, not brand names

Granions already has the products and the ingredient information relevant to that question. The GEO/AEO strategy organises and strengthens that content so search engines and AI answer systems can more easily understand its relevance.

HOW WE APPLY ITReview priority product and ingredient pages; research customer questions and competing sources; add the missing explanations and FAQs; improve clarity, credible citations and evidence; check technical discoverability; track search and AI-answer visibility over time.

The method, step by step

Click any step, or swipe the panel for the next one.

The demo, step by step

Swipe for the next step

Swipe the demo or use the arrows for the next step. Click a step above to jump to it.

How it comes together

Audit→Select→Competitors→Google Trends→Find the gap→Apply GEO methods→Improve page→Technical check→Re-audit

Rankings, citations and placement inside AI-generated answers cannot be guaranteed, this is the discovery opportunity, and we measure it over time.

Granions’ formulation expertise can support product discovery even before customers know which brand to search for.

Product recommendation

Decision Recall Personalised Recommendations

Your products become part of each customer’s skin journey, with a clear reason why, never just a product name.

What we created

The analysis connects relevant visible concerns and questionnaire answers with approved products from the Granions Dermatologie range: Niacinamide 12% for uneven tone and visible spots, Hyaluronic Acid 4% for hydration, Azelaic Acid 10% for visible blemishes, the Anti-Age Lifting Serum for fine lines, SPF30 Face Cream for daily protection.

Granions already combines topical skincare and beauty supplements through its In & Out approach. The experience makes those complementary routines easier to discover, the Hyaluronic Acid 4% serum on the OUT side, Hyaluronic Acid 600 mg on the IN side, together as the existing Hydration In & Out Pack.

Every suggestion includes a clear explanation of why the product may be relevant. Recommendations are illustrative and require Granions-approved product rules, contraindication screening where applicable and accurate cosmetic claims.

Example recommendation pathways

PATHWAY

Uneven skin tone and visible spots → Niacinamide 12% Serum

Positioned around helping improve the appearance of uneven tone and pigmentation marks

The customer sees the observation on her own face map, the ingredient explained in plain words, and the product with its usage and suitability information.

HOW WE APPLY ITApproved product rules decide when the serum is suggested; the explanation always states why.
IN & OUT ROUTINE

Hydration → Hyaluronic Acid 4% Serum + Hyaluronic Acid 600 mg

OUT: topical skincare · IN: beauty supplementation · Together: the existing Hydration In & Out Pack

A product journey organised around customer needs, using a routine philosophy Granions already owns. The same principle supports the anti-ageing and other In & Out bundles.

HOW WE APPLY ITSupplement suggestions require an additional suitability questionnaire and Granions-approved guidance, they are never inferred from a selfie alone.

How it works, step by step

Click any step, or swipe the panel for the next one.

The demo, step by step

Swipe for the next step

Swipe the demo or use the arrows for the next step. Click a step above to jump to it.

How it comes together

Visible concern→Questionnaire answers→Approved product rules→Product + the reason why→In & Out routine

From individual products to thoughtfully connected routines, the Granions products become part of each customer’s skin journey.

Product exploration

Decision Booster AI Shopping Assistant

A personal product guide inside Granions.fr that understands the need, shows the product, and explains why it fits.

What we created

Granions’ products contain detailed formulation, ingredient and usage information. The AI Shopping Assistant makes that knowledge easier to access through a natural conversation: it understands what the customer is trying to achieve, finds relevant products from the catalogue, displays their images and details, explains why each may be suitable, supports follow-up questions and comparisons, and directs customers to the existing product pages.

The assistant itself is flexible: customers can resize it and move it around the screen, so they can keep browsing Granions.fr while continuing the conversation.

A real conversation

A CUSTOMER ASKS

“My skin feels dry, but I also have uneven skin tone. Which Granions serum should I explore?”

The assistant responds

“Granions has different serums for those concerns. Hyaluronic Acid 4% focuses on hydration, while Niacinamide 12% is designed for uneven-looking tone and spots. I can explain their ingredients and help you compare them. Is your skin sensitive, and are you currently using other active skincare products?”

HOW WE APPLY ITGrounded only in Granions-approved product information; the two real products appear inside the chat with image and key facts; a healthcare professional is recommended whenever a question needs one.

How it works, step by step

Click any step, or swipe the panel for the next one.

The demo, step by step

Swipe for the next step

Swipe the demo or use the arrows for the next step. Click a step above to jump to it.

How it comes together

Understand the need→Find the matching product→Show the product visually→Explain why it fits→Follow-ups and comparisons→Open the product page

All while the customer can move, resize, or keep the assistant open alongside the store.

Granions’ product expertise becomes available in a format customers can easily understand and use.

Purchase confidence

Confidence Booster Smart Reviews

Real customer experiences at the right moment, authentic feedback made easier to discover.

What we created

Granions already displays verified customer reviews across its product pages. Smart Reviews helps shoppers encounter relevant feedback while they are considering a product: a small, unobtrusive review card shows an authentic customer experience related to that product, and other useful reviews can follow.

Verified Customer Reviews and AI Product Insights have separate labels and sources. The technology never invents testimonials or rewrites AI content as if it were a customer review.

The rule behind it

TRANSPARENCY

Verified Customer Review · AI Product Insight

Two labels, two sources, always visually distinct

Customers discover useful experiences without searching through an entire review section, while existing feedback stays authentic and transparent.

HOW WE APPLY ITReal feedback from a real customer stays a Customer Review; an AI interpretation of approved product information is labelled an AI Product Insight.

How it works, step by step

Click any step, or swipe the panel for the next one.

The demo, step by step

Swipe for the next step

Swipe the demo or use the arrows for the next step. Click a step above to jump to it.

How it comes together

Customer explores a product→A relevant review appears→Other useful reviews follow→Insights clearly distinguished→An informed decision

For Granions, existing customer feedback becomes a more accessible part of product discovery, while preserving authenticity and transparency.

Frictionless purchase

Your Website. Their Convenience.A Better Way to Shop.

Give Granions customers the convenience of a mobile app, with one-handed navigation, easier product discovery, and effortless shopping, all within your existing website.

What we created

Granions already offers a mobile-accessible website. But displaying a website on a smaller screen and designing the shopping journey around mobile behaviour are two different things. Customers often shop holding the phone in one hand.

There is nothing to download and no separate app to maintain. The existing website and domain serve the experience; products, cart and checkout stay connected to the existing store.

Left-side category navigation (Skin Care, Collagen, Hair, Vitamins, Wellbeing, Sleep). Instant filtering within Skin Care: All Products, Serums, Creams, Routines. A two-column visual product grid. Fixed bottom navigation: Home · Search · Categories · Cart · Account. Sticky Add to Cart, integrated assistance and relevant reviews on every product page.

The objective is not to make the website look like an app. It is to make shopping feel easier.

The screen behind it

ONE-HANDED BROWSING

Every product category, just a thumb away

Category rail · instant filters · visual grid · fixed bottom navigation

A customer looking for a specific serum moves between categories and product options without constantly returning to the top menu.

HOW WE APPLY ITThe category layout from 10 Turtle’s mobile shopping experience, with official Granions products in place of the reference products.

How it works, step by step

Click any step, or swipe the panel for the next one.

The demo, step by step

Swipe for the next step

Swipe the demo or use the arrows for the next step. Click a step above to jump to it.

How it comes together

Open Granions→Find the category→Filter→View the product→Add to cart→Existing checkout

Fewer unnecessary steps between opening the store and reaching what the customer wants.

One digital storefront. An experience designed for the device in their hand.

Roadmap

Start with a focused experience. Learn from real customer behaviour.

Every capability runs on the existing Granions website: same domain, same catalogue, same cart and checkout. Begin with the strongest customer opportunity, validate it, expand on measurable results.

The plan, step by step

1. Discover & configure

Review the website integration options, map the relevant Granions products, define approved recommendation rules, establish privacy and measurement requirements.

2. Launch the flagship experience

Pilot AI Skin Analysis within the Dermatologie journey, connect product recommendations, introduce the AI Shopping Assistant where it supports customer questions.

3. Enhance mobile shopping

App-like category navigation, simplified mobile product discovery and convenient purchase actions on selected journeys.

4. Extend discovery & trust

Smart Reviews on selected product pages; GEO/AEO improvements on priority educational and product content.

5. Measure & expand

Assessment completion, recommended-product clicks, mobile navigation behaviour, assistant engagement, add-to-cart, purchases and repeat visits.

The approach is modular: begin with the strongest customer opportunity, validate the experience, and expand based on measurable results.

How everything integrates

New capabilities, existing infrastructure. The website continues as the central commerce platform; each capability can be introduced and tested independently.

Layer 1: Existing Granions website

Product pages, navigation, catalogue, account, cart and checkout.

Layer 2: Customer experience enhancements

Embedded AI Skin Analysis, mobile interface improvements, AI Shopping Assistant, Smart Reviews.

Layer 3: Granions product intelligence

Approved product information, ingredients, usage guidance, suitability rules and product images.

Layer 4: Connected customer journey

AI recommendations link back to existing product pages and shopping actions.

Layer 5: Privacy, controls & measurement

Customer consent, appropriate data handling, review authenticity, analytics and performance monitoring.

Every step includes

Implementation

Embedded components and front-end enhancements on the existing site. No rebuild, no platform migration, nothing for customers to download.

Approved content

Recommendations use Granions-approved information, suitability rules and accurate cosmetic claims, cosmetic observations, not diagnoses.

Consent & privacy

Facial photographs and assessment data handled with appropriate consent, privacy protection and data minimisation.

Measurement

A controlled pilot against the current shopping journey, real changes, not an assumed uplift, reported month by month.

Precision in formulation. Personalisation in discovery.

Back to the journey
Value for customers and for Granions

Every improvement has a purpose.

Each customer improvement maps to an opportunity for Granions: more qualified product visits, stronger engagement, relevant routine discovery and repeat customer relationships.

What customers gain

Customers discover Granions more naturally, understand their skin concerns more clearly, find products with less effort, make better-informed decisions, and have a reason to return.

For Granions: more qualified product visits, stronger engagement, relevant routine discovery and repeat customer relationships.

What customers gain

1. Discover Granions more naturally

Relevant educational content reaches people actively researching skincare needs and ingredients.

2. Understand their skin concerns

AI Skin Analysis gives shoppers an engaging starting point for exploring suitable products.

3. Find products with less effort

Mobile-first navigation and intelligent assistance reduce unnecessary searching and comparison.

4. Decide with better information

Clear product explanations and authentic reviews provide useful context before purchasing, and a reason to return.

Capability by capability

Each capability, the journey stage it supports, what it changes for the customer, and what we measure in the pilot.

#CapabilityJourney stageWhat changes for the customerWhat we measure
1GEO / AEOStage 1: Product discoveryGranions’ ingredient expertise is found when customers ask questions, before they know which brand to search forSearch and AI-answer visibility over time; qualified visits to priority pages
2AI Skin AnalysisStage 2: Understanding the skinA personalised starting point instead of generic browsing; a report that explains instead of scoringAssessment starts and completion; recommended-product clicks; optional follow-up assessments
3Personalised recommendationsStage 3: Product recommendationProducts suggested with the reason why; In & Out routines easier to discover and understandEngagement with recommendations; routine (pack) discovery and adoption
4AI Shopping AssistantStage 4: Product explorationQuestions answered with approved product information; comparisons and follow-ups in contextAssistant engagement; questions answered; add-to-cart from conversations
5Smart ReviewsStage 5: Purchase confidenceRelevant verified reviews at the moment of decision, clearly separated from AI insightsReview visibility and interaction; add-to-cart on pages with Smart Reviews
6App-like mobile shoppingStage 6: Frictionless purchaseOne-handed browsing, sticky Add to Cart and the existing checkout, nothing to downloadMobile navigation behaviour; mobile add-to-cart; checkout completion; repeat visits

The next chapter of the Granions experience

A customer who understands her skin

And a product recommendation she can understand.

A mobile experience that feels natural

And a knowledgeable assistant when she needs guidance.

Authentic experiences from other customers

And a clearer path from discovering Granions to choosing its products.

All within the website Granions has already built

Proposed next step: a focused demonstration using Granions’ actual Dermatologie products, followed by a technical integration review and pilot plan.

Your science. Your products. A more personal digital experience.

See the roadmap
Research

The research behind 10turtle

Fourteen peer-reviewed papers and two frameworks. Each of the six tools applies two or three of them, and every one below answers the same four questions.

How to read this

1. What the paper is

Who wrote it, where it was published, what they tested.

2. What we took from it

The finding, in plain words.

3. Where we used it

The tool, the journey stage and the exact feature.

4. How we used it

What the tool does differently because of this finding.

The two frameworks the AI Shopping Assistant also builds on, Jobs to Be Done and the Means-End Chain, are summarised in that section with the same four questions.

Overview: paper to tool

All 14 papers and the two frameworks at a glance, in the order the tools appear in the deck. Click a title to jump to its summary.

#PaperInstitutions · Published inTool (journey stage)What we took from it
1GEO: Generative Engine OptimizationIIT Delhi, Princeton University · KDD 2024 (ACM SIGKDD)GEO
Stage 1: Product discovery		Statistics, citations, quotes, clarity and authority make a page more likely to be cited by AI engines; keyword stuffing does not
2Concrete LanguageYork University (Schulich), Wharton (University of Pennsylvania) · Journal of Consumer Research, 2020AI Shopping Assistant
Stage 3: Product exploration		Specific, concrete words make customers feel understood and buy more
3Processing FluencyPrinceton University · Psychological Bulletin (APA), 2008AI Shopping Assistant
Stage 3: Product exploration		Information that is easier to process is judged more favourably
4Voice Assistant, Buy Coffee Capsules!LMU Munich, University of Münster · The DATA BASE for Advances in Information Systems (ACM SIGMIS), 2023Voice Shopping
Stage 4: Conversational shopping		People use voice shopping for convenience and enjoyment; reliability and control decide whether they keep using it
5Media Equation TheoryUniversity of Minnesota, Cal Poly Pomona · Journal of Global Fashion Marketing, 2023Voice Shopping
Stage 4: Conversational shopping		People respond socially to systems that communicate like humans; human-likeness and trust drive purchase intention
6Social Presence and TrustUniversity of Portsmouth, University of Winchester · Psychology and Marketing, 2021Voice Shopping
Stage 4: Conversational shopping		Consistent speaking and listening behaviour builds social presence, which builds trust and acceptance of recommendations
7Website MorphingMIT Sloan School of Management · Marketing Science (INFORMS), 2009Mobile Experience
Stage 6: Frictionless purchase		A site does not have to show everyone the same interface; adapting it raises purchase intention
8Search Cost Theory (WeStore vs. AppStore)Vanderbilt University (Owen), Ohio State University (Fisher) · SSRN working paper, JD.com dataMobile Experience
Stage 6: Frictionless purchase		Fewer searches, page loads and navigation steps mean more exploration and more conversion
9Technology Acceptance ModelMIT Sloan School of Management · PhD dissertation, 1985Mobile Experience
Stage 6: Frictionless purchase		People adopt what is useful and easy to use, not what merely looks new
10How Online Reviews Affect Purchase Intention: A Meta-AnalysisWuhan University · Data and Information Management (Elsevier), 2024Smart Reviews
Stage 5: Purchase confidence		Reviews measurably move purchase intention, so they should not sit unread at the bottom of the page
11The Role of Review Structure in Perceived HelpfulnessUniversity of Queensland, University of Cambridge (Judge) · Scientific Reports (Nature Portfolio), 2026Smart Reviews
Stage 5: Purchase confidence		How a review is structured decides how helpful it is, and the best structure depends on the product’s rating
12Consumer Reactions to Perceived Undisclosed ChatGPT Usage in an Online Review ContextWeber State University · Telematics and Informatics (Elsevier), 2024Smart Reviews
Stage 5: Purchase confidence		Reviews that look AI-written lose trust, so AI content must never pose as a customer
13Consideration Set EffectMIT Sloan School of Management · Journal of Consumer Research, 1990Recommendations
Stage 2: Product recommendation		Shoppers narrow to a small set before choosing; help them build that set faster
14How Do Recommender Systems Lead to Consumer Purchases?HEC Paris, University of Cologne, Goethe University Frankfurt · Information Systems Research (INFORMS)Recommendations
Stage 2: Product recommendation		Recommendations raise purchase propensity and basket value by making the store remember what mattered
F1Jobs to Be Done
Framework		Harvard Business School · The Innovator’s Solution (2003); “Know Your Customers’ ‘Jobs to Be Done’”, Harvard Business Review, 2016AI Shopping Assistant
Stage 3: Product exploration		Customers hire a product to get a specific job done; the job, not the customer’s profile, predicts what they buy
F2Means-End Chain Theory
Framework		University of Southern California · Journal of Marketing (AMA), 1982AI Shopping Assistant
Stage 3: Product exploration		A product attribute matters only through the consequence it produces and the personal value that consequence serves
Stage 1: Product discovery

GEO

One paper drives the whole GEO method: it tells us which page changes make AI engines cite a product page, and which do not.

Paper 1 of 14GEO · Stage 1: Product discovery

GEO: Generative Engine Optimization

Indian Institute of Technology Delhi and Princeton University · KDD 2024, the ACM SIGKDD conference, a top peer-reviewed venue in data mining · Tested on a 10,000-query benchmark across 25 domains and on a live generative engine (Perplexity.ai).

What the paper is. Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande (IIT Delhi and Princeton), published at KDD 2024. They built GEO-bench, a benchmark of 10,000 real queries across 25 domains, and tested nine ways of rewriting a web page to see which ones made generative engines (including Perplexity.ai) surface and cite that page more. Three changes work best: adding credible citations, adding quotations and adding statistics. They raised visibility by 30 to 40% on the position-adjusted word count metric and 15 to 30% on subjective impression.

1. What we took from it

  • Three changes work best: adding credible citations, adding quotations and adding statistics. They raised visibility by 30 to 40% on the position-adjusted word count metric and 15 to 30% on subjective impression.
  • Making the text clearer and easier to read also helped, by 15 to 30%. Engines reward presentation, not only content.
  • Keyword stuffing, the classic SEO trick, does not work for AI engines.
  • Which method works best varies by domain, so pages need targeted changes, not one template.

2. Where we used it

The GEO page: the research card, the “How we apply it” line, and step 7 of the ten-step method (“GEO methods”). It also sits behind the Product discovery tile on the Journey page.

3. How we used it

Steps 1 to 6 of our method find what a product page is missing (audit, select pages, understand the page, competitors, search trends, find the gap). Step 7 fixes the gap using only the paper’s proven methods: relevant statistics, credible sources, useful quotations, clearer wording and a more authoritative tone, and never extra keywords. Step 8 applies those changes to the page and adds conversational FAQs, step 9 checks crawl access, sitemap, speed and AI access, and step 10 re-audits so the before-and-after is measured the way the paper measured it. Because the paper found results differ by domain, we audit each page instead of applying one fixed formula.

Engines reward presentation, not only content.

30–40%visibility gain from citations, quotations and statistics
15–30%gain from clearer, easier-to-read text
10,000queries in the benchmark, across 25 domains
Figure. Key numbers from the paper, and from how the tool applies it.
Stage 3: Product exploration

AI Shopping Assistant

Two papers shape how the assistant talks: one says use concrete facts, the other says make the decision easy to process. Two frameworks, Jobs to Be Done and the Means-End Chain, shape what it talks about; both are summarised below the papers.

Paper 2 of 14AI Shopping Assistant · Stage 3: Product exploration

Concrete Language

Schulich School of Business, York University, and The Wharton School, University of Pennsylvania · Journal of Consumer Research, 2020, one of the leading peer-reviewed journals in marketing · Five studies, including more than 1,000 real customer-service conversations from two companies.

What the paper is. “How Concrete Language Shapes Customer Satisfaction” by Grant Packard (York University) and Jonah Berger (Wharton), Journal of Consumer Research, 2020. Five studies, including text analysis of over 1,000 real customer-service conversations (phone calls and emails) from two companies, plus controlled experiments. Customers are more satisfied, more willing to buy and spend more when the employee speaks concretely (“that grey linen shirt”, not “that item”).

1. What we took from it

  • Customers are more satisfied, more willing to buy and spend more when the employee speaks concretely (“that grey linen shirt”, not “that item”).
  • One standard deviation more concreteness raised satisfaction by 9% and actual spending by at least 13% in the field data.
  • The reason: concrete words signal that the employee listened and understood the customer’s specific need.

2. Where we used it

The AI Shopping Assistant page, research card 3, and step 5 of its method (“Explain why”). The vacuum example on the page (“2.5 kg”, not “lightweight”) is this paper in one line.

3. How we used it

The assistant is set up to speak only in verified catalogue facts: in step 1 every spec is turned into a plain fact (2.5 kg, 60 minutes of runtime), and in step 5 the explanation uses numbers instead of adjectives. In step 3 the assistant states the shopper’s need back in their own concrete terms before recommending anything, which is exactly the listening signal the paper found behind the effect.

Concrete words signal that the employee listened and understood.

+9%satisfaction per standard deviation of concreteness
+13%actual spending, at least, in the field data
1,000+real customer-service conversations analysed
Figure. Key numbers from the paper, and from how the tool applies it.
Paper 3 of 14AI Shopping Assistant · Stage 3: Product exploration

Processing Fluency

Princeton University, Department of Psychology · Psychological Bulletin, 2008, the American Psychological Association’s flagship review journal · A review that organises decades of judgment and decision-making research into one framework.

What the paper is. Anuj Shah and Daniel Oppenheimer (Princeton), “Heuristics Made Easy: An Effort-Reduction Framework”, Psychological Bulletin, 2008. A review paper that explains how people judge and decide when they cannot weigh every fact: they cut effort by looking at fewer cues, using cues that are easy to access and process (fluency), simplifying how they weigh them, integrating less information and comparing fewer alternatives. People do not compute; they reach for the easiest-to-process information and judge from that.

1. What we took from it

  • People do not compute; they reach for the easiest-to-process information and judge from that.
  • A fact that is easy to grasp carries more weight in a decision than a fact that is technically better but harder to process.
  • Fewer options and fewer cues make a decision feel easier and get made.

2. Where we used it

The AI Shopping Assistant page, research card 4, and steps 4 and 5 of its method (“Find the match”, “Explain why”). The line “Half the weight of your current vacuum” on the page is this principle.

3. How we used it

The assistant reduces the shopper’s effort the way the paper describes. It shows one product that fits rather than a list of 48 (fewer alternatives), picks the one feature that matters instead of listing every spec (fewer cues), and translates the raw number into something instantly meaningful, such as “2.5 kg, half the weight of your current vacuum” (an easy-to-process cue). Two suggested follow-ups after each answer keep the next decision small too.

People do not compute; they reach for the easiest-to-process information.

1 of 48products shown: fewer alternatives
1feature that matters, not every spec
2follow-ups after each answer
Figure. Key numbers from the paper, and from how the tool applies it.
Framework 1 of 2AI Shopping Assistant · Stage 3: Product exploration

Jobs to Be Done

Clayton Christensen, Harvard Business School · Introduced in The Innovator’s Solution (Harvard Business Review Press, 2003, with Michael Raynor), set out in “Know Your Customers’ ‘Jobs to Be Done’”, Harvard Business Review, 2016, and the book Competing Against Luck, 2016 · A framework built from case studies of product successes and failures rather than a single experiment.

What the paper is. A framework, not a single study. Christensen and his co-authors (Taddy Hall, Karen Dillon and David Duncan in the HBR article) argue that people do not buy products because of who they are; they “hire” a product to make progress in a particular circumstance, the job to be done. A job has functional, social and emotional dimensions. The best-known illustration is the fast-food milkshake hired for a long, boring morning commute, which competes with bananas and bagels, not with other milkshakes. Customers hire products to get a job done in a specific circumstance; understanding the job predicts purchase better than demographics or product attributes.

1. What we took from it

  • Customers hire products to get a job done in a specific circumstance; understanding the job predicts purchase better than demographics or product attributes.
  • A job has functional, social and emotional dimensions, and the circumstance matters as much as the customer.
  • The real competition is anything else that gets the job done, including doing nothing, so a product has to be positioned against the job, not against its category.

2. Where we used it

The AI Shopping Assistant page, research card 1, and steps 2 and 3 of its method (“Buying questions”, “Understand the need”).

3. How we used it

Step 2 maps the real job behind each product category before the assistant ever talks to a shopper: what people are trying to get done when they look for a vacuum, a jacket or a pair of running shoes. Step 3 applies it in the conversation: “my vacuum is heavy” is translated into the job, “easier cleaning upstairs”, and the assistant states that job back to the shopper before it recommends anything. The product it then picks is the one that gets the job done, not the one that scores best on a spec sheet.

People don’t buy products; they hire them to get a job done.

3dimensions of a job: functional, social, emotional
2003 → 2016from The Innovator’s Solution to the HBR article and Competing Against Luck
Figure. Key numbers from the paper, and from how the tool applies it.
Framework 2 of 2AI Shopping Assistant · Stage 3: Product exploration

Means-End Chain Theory

Jonathan Gutman, University of Southern California · “A Means-End Chain Model Based on Consumer Categorization Processes”, Journal of Marketing, 1982, the American Marketing Association’s flagship peer-reviewed journal · A conceptual model that became the basis of the laddering interview method (Reynolds and Gutman, Journal of Advertising Research, 1988), used in advertising and product strategy for four decades.

What the paper is. Jonathan Gutman, Journal of Marketing, 1982. The paper proposes that consumers categorise products by what they lead to: a product’s attributes produce consequences (functional and psychosocial), and consequences are wanted because they serve personal values or end states. Attribute, consequence, value form a chain, the means-end chain, that links what a product is to why a person cares. Reynolds and Gutman later turned the model into the laddering interview, which walks a consumer up the chain by repeatedly asking why an attribute matters. An attribute on its own means nothing to a shopper; it matters through the consequence it produces and the personal outcome that consequence serves.

1. What we took from it

  • An attribute on its own means nothing to a shopper; it matters through the consequence it produces and the personal outcome that consequence serves.
  • The chain runs attribute, then benefit, then personal outcome, and the shopper cares most about the top of the chain.
  • To explain a product convincingly you climb the chain: name the attribute, say what it does for the person, and connect it to what they are trying to achieve.

2. Where we used it

The AI Shopping Assistant page, research card 2, and step 5 of its method (“Explain why”).

3. How we used it

When the assistant explains why a product fits, it does not read out the spec sheet. It picks the one attribute that matters for the job and climbs the chain in one sentence: 2.5 kg (attribute), less to carry (consequence), easier upstairs (the outcome the shopper asked for). Every explanation in the tool follows that attribute, benefit, outcome order, so the number the shopper hears is always tied to the reason they came.

Attribute, then benefit, then personal outcome: the shopper cares most about the top of the chain.

3links in the chain: attribute, consequence, value
1982 → 1988from the model to the laddering interview method
Figure. Key numbers from the paper, and from how the tool applies it.
Stage 4: Conversational shopping

Voice Shopping

Three papers answer three questions: why people shop by voice at all (convenience and enjoyment), what makes them trust the assistant (human-likeness and control) and what builds that trust (social presence and competence).

Paper 4 of 14Voice Shopping · Stage 4: Conversational shopping

Voice Assistant, Buy Coffee Capsules!

Ludwig-Maximilians-Universität München (LMU Munich) and University of Münster · The DATA BASE for Advances in Information Systems, 2023, the peer-reviewed journal of ACM SIGMIS · Mixed methods: 30 user interviews and a 176-person survey.

What the paper is. “Understanding the Determinants of Consumers’ Intention to Use Voice Commerce” by Rzepka, Berger, Koslow and Hess (LMU Munich and University of Münster), The DATA BASE for Advances in Information Systems, 2023. Study 1: 30 interviews with voice-assistant users to build a model of benefits and risks. Study 2: a survey of 176 people to test it. Convenience is by far the strongest reason people value voice shopping (β = .588), with enjoyment second (β = .207). Efficiency on its own did not matter.

1. What we took from it

  • Convenience is by far the strongest reason people value voice shopping (β = .588), with enjoyment second (β = .207). Efficiency on its own did not matter.
  • Three risks hold people back: the assistant being unreliable (β = .430), the shopper losing control (β = .269) and not understanding what it is doing (opacity, β = .171).
  • Benefits raise the intention to use voice commerce and risks lower it; together they explained 61.8% of that intention.

2. Where we used it

The Voice Shopping page, research card 1 (Impact 9/10), and steps 1, 2 and 5 of its method.

3. How we used it

Convenience comes first: one spoken sentence carries several constraints (“light, under $300, for stairs”) and returns three strong matches, not 48. The three risks are designed out. Reliability: product names are mapped to the store’s SKUs, mishearings included, and the assistant asks a clarifying question instead of returning an error. Control: follow-ups such as “anything cheaper?” refine the results and the shopper stays in charge of every step. Opacity: results appear on screen as cards, so the shopper can see what the assistant found and why.

Convenience is by far the strongest reason people value voice shopping.

β = .588convenience, the strongest driver
β = .430unreliability, the strongest risk
61.8%of the intention to use, explained
Figure. Key numbers from the paper, and from how the tool applies it.
Paper 5 of 14Voice Shopping · Stage 4: Conversational shopping

Media Equation Theory

University of Minnesota and California State Polytechnic University, Pomona · Journal of Global Fashion Marketing, 2023, peer-reviewed · Two controlled experiments, 86 and 112 participants, in a real voice-shopping task.

What the paper is. “Building trust with voice assistants for apparel shopping: the effects of social role and user autonomy” by Huh, Whang and Kim (University of Minnesota and Cal Poly Pomona), Journal of Global Fashion Marketing, 2023. Two experiments on voice shopping for clothes, with 86 and 112 participants, built on media equation theory: people respond to machines with human social rules when the machine communicates like a person. Giving the shopper more autonomy (control over the interaction) raised perceived human-likeness, then trust, then purchase intention. That chain (human-likeness, then trust) is what carried the effect.

1. What we took from it

  • Giving the shopper more autonomy (control over the interaction) raised perceived human-likeness, then trust, then purchase intention. That chain (human-likeness, then trust) is what carried the effect.
  • Whether the assistant was framed as a “partner” or a “servant” made little difference on its own; control did.
  • A basic human-like cue, such as natural speech, is enough. Retailers do not need a richer human persona, they need error-free understanding and responsiveness.

2. Where we used it

The Voice Shopping page, research card 2 (Impact 8/10), and steps 4 and 5 of its method (“Sound human”, “Follow-ups”).

3. How we used it

The assistant speaks in full sentences (“Got it, something light for the stairs”) with the same tone every time, so it feels like a person rather than a voice search box. Shopper autonomy is built in: the shopper leads with follow-up questions, refines by price, weight or feature, and adds to cart only when they choose to. Nothing is read aloud that can be shown, so the shopper decides with their eyes.

Autonomy raised human-likeness, then trust, then purchase intention.

86 + 112participants across two experiments
Controlnot persona, is what carried the effect
Figure. Key numbers from the paper, and from how the tool applies it.
Paper 6 of 14Voice Shopping · Stage 4: Conversational shopping

Social Presence and Trust

University of Portsmouth and University of Winchester · Psychology and Marketing, 2021, a long-established peer-reviewed journal · Structural equation model on 466 voice-assistant users, confirmed by a qualitative study.

What the paper is. “Alexa, she’s not human but… Unveiling the drivers of consumers’ trust in voice-based artificial intelligence” by Pitardi and Marriott (University of Portsmouth and University of Winchester), Psychology and Marketing, 2021. A structural model tested on 466 voice-assistant users, followed by interviews. Usefulness and ease of use shape the attitude toward using a voice assistant, but they do not create trust.

1. What we took from it

  • Usefulness and ease of use shape the attitude toward using a voice assistant, but they do not create trust.
  • Trust comes from the social side of the interaction: social presence (the feeling that someone is there) and perceived competence.
  • People apply human social rules to assistants that behave consistently, so a natural, consistent conversational flow is what earns trust and, through it, acceptance of what the assistant suggests.

2. Where we used it

The Voice Shopping page, research card 3 (Impact 8/10), and step 4 of its method (“Sound human”).

3. How we used it

The assistant keeps a consistent speaking and listening behaviour: a visible listening state while the microphone is on, the same tone in every reply, and context carried across turns so it never “forgets” mid-conversation. Competence is shown, not claimed: it searches the same catalogue as the rest of the store, puts in-stock products first and shows image, price and the key spec for each result.

Trust comes from the social side of the interaction.

466voice-assistant users in the structural model
2sources of trust: social presence and competence
Figure. Key numbers from the paper, and from how the tool applies it.
Stage 6: Frictionless purchase

Mobile Experience

Three papers give the mobile layer its permission, its target and its test: a site may change its interface per context (Morphing), each extra step costs a sale (Search Cost), and people keep using what is useful and easy (TAM).

Paper 7 of 14Mobile Experience · Stage 6: Frictionless purchase

Website Morphing

MIT Sloan School of Management · Marketing Science, 2009, the INFORMS journal and one of the most rigorous in the field · Field data from a live BT Group (British Telecom) website, 835 respondents.

What the paper is. John Hauser, Glen Urban, Guilherme Liberali and Michael Braun (MIT Sloan), Marketing Science, 2009. They built a system that infers a visitor’s cognitive style from clickstream data and automatically changes the look, navigation and detail level of a website to match it, tested on an experimental BT Group broadband website with data from 835 respondents. A website does not have to show every visitor the same interface. Its structure, navigation, tools and presentation can adapt to the visitor.

1. What we took from it

  • A website does not have to show every visitor the same interface. Its structure, navigation, tools and presentation can adapt to the visitor.
  • Matching the interface to the visitor raised purchase intentions by 21% with perfect information, and by almost 20% when the system had to learn the style from clicks.
  • The gain came from changing the “look and feel”, not only the content.

2. Where we used it

The Mobile Experience page, research card 1 (Impact 9/10), and step 2 of its method (“Keep the store”).

3. How we used it

The paper morphs the site to cognitive style; we apply the same adaptive-interface principle to context: the phone. The existing store, domain, products, cart and checkout stay exactly as they are, and a layer switches on only for phone visitors, changing how the storefront is presented and operated (bottom navigation, one-screen categories, focused product pages). The desktop site is untouched, so one store serves two interfaces.

The site changes; the visitor does not have to.

+21%purchase intention with perfect information
≈ +20%when the style is learned from clicks
835respondents on a live BT website
Figure. Key numbers from the paper, and from how the tool applies it.
Paper 8 of 14Mobile Experience · Stage 6: Frictionless purchase

Search Cost Theory (WeStore vs. AppStore)

Owen Graduate School of Management, Vanderbilt University, and Fisher College of Business, The Ohio State University · Working paper on SSRN · Customer-level clickstream data from JD.com, one of the largest online retailers in the world.

What the paper is. “WeStore or AppStore: How Customers Shop Differently in Mobile Apps vs. Social Commerce” by Kejia Hu (Vanderbilt) and Nil Karacaoglu (Ohio State), a working paper on SSRN. They modelled the complete shopping journey of a large online retailer’s customers across two channels, its native app and its lightweight WeChat mini-program, using a sequential search model that estimates the cost of starting a search and the cost of each extra search step. Shopping gets harder with every extra search, page change or navigation step; each step is a cost the shopper weighs against giving up.

1. What we took from it

  • Shopping gets harder with every extra search, page change or navigation step; each step is a cost the shopper weighs against giving up.
  • In the lightweight channel the cost of each additional step was lower, and customers explored more products (3.11 versus 2.64 per session) and converted more often: 41% versus 26.8% in the studied dataset.
  • Lowering the cost of each step is a lever a retailer controls.

2. Where we used it

The Mobile Experience page, research card 2 (Impact 9/10), and steps 1, 3, 4, 6 and 7 of its method.

3. How we used it

Step 1 counts the taps, screens and page loads between opening the store and buying, which is the paper’s marginal search cost made visible. The layer then removes steps: bottom navigation puts Home, Search, Categories, Cart and Account one tap away; categories open on one screen with products beside them and filters as chips, with no page loads in between; the cart badge and Checkout button are always in view. Step 7 counts taps to product before and after, so the reduction is measured.

Each extra step is a cost the shopper weighs against giving up.

41% vs 26.8%conversion, lightweight channel vs. native app
3.11 vs 2.64products explored per session
Figure. Key numbers from the paper, and from how the tool applies it.
Paper 9 of 14Mobile Experience · Stage 6: Frictionless purchase

Technology Acceptance Model

MIT Sloan School of Management · Fred D. Davis’s PhD dissertation, 1985, the origin of the Technology Acceptance Model, one of the most widely used and cited models in technology-adoption research · A field survey of 100 users and a lab experiment with 40 MBA students.

What the paper is. Fred D. Davis’s PhD dissertation at MIT Sloan, 1985, “A Technology Acceptance Model for Empirically Testing New End-User Information Systems”. It proposed and tested the model behind decades of adoption research, with a field survey of 100 organisational users and a lab experiment with 40 MBA students on two graphics systems. Two beliefs decide whether people accept and keep using a technology: perceived usefulness (it helps me get something done) and perceived ease of use (it takes little effort).

1. What we took from it

  • Two beliefs decide whether people accept and keep using a technology: perceived usefulness (it helps me get something done) and perceived ease of use (it takes little effort).
  • Those beliefs, not novelty or appearance, drive the attitude that leads to actual use.
  • Ease of use also feeds usefulness: something easier to use is judged more useful.

2. Where we used it

The Mobile Experience page, research card 3 (Impact 8/10), and steps 3 to 6 of its method.

3. How we used it

We do not make the site look like an app for appearance’s sake. Every change has to score on the two TAM beliefs: familiar mobile patterns (bottom tabs, large tap targets, sticky Add to cart, collapsed details) for ease of use, and a faster route from opening the store to the product or the cart for usefulness. Because nothing has to be downloaded, the first barrier to use is gone entirely.

People adopt what is useful and easy to use, not what merely looks new.

2beliefs that decide adoption: usefulness and ease of use
100 + 40users in the field survey, MBA students in the lab
Figure. Key numbers from the paper, and from how the tool applies it.
Stage 5: Purchase confidence

Smart Reviews

Three papers answer why reviews must be surfaced (the meta-analysis), which review to surface (review structure) and what must never happen (AI content posing as a customer).

Paper 10 of 14Smart Reviews · Stage 5: Purchase confidence

How Online Reviews Affect Purchase Intention: A Meta-Analysis

Wuhan University, School of Information Management · Data and Information Management, 2024, a peer-reviewed Elsevier journal · A meta-analysis of 156 studies, 214 effect sizes and 69,006 observations, the broadest evidence base in this deck.

What the paper is. Keda Qiu and Liyi Zhang (Wuhan University), Data and Information Management, 2024. A meta-analysis that pools 156 studies, 214 effect sizes and 69,006 observations to settle what actually moves purchase intention in online reviews, and what moderates it (culture, product type, study design). Every review factor studied has a significant effect on purchase intention. Review valence (how positive the review is) is the strongest, r = 0.563.

1. What we took from it

  • Every review factor studied has a significant effect on purchase intention. Review valence (how positive the review is) is the strongest, r = 0.563.
  • Next come review usefulness (r = 0.481), review credibility (r = 0.460) and how similar the reviewer feels to the reader (r = 0.451). Review volume matters less (r = 0.317).
  • Reviews that are useful, credible and relevant to the reader do more than many reviews.

2. Where we used it

The Smart Reviews page, research card 1 (Impact 9/10), and steps 4 and 6 of its method (“Right moment”, “Rotate and learn”).

3. How we used it

If reviews move the purchase decision this much, they cannot stay hidden below the fold. The popup brings a relevant review into the product page after a suitable interval, near Add to cart, while the shopper is deciding, and rotates a different useful review as browsing continues. It is dismissed with one tap and does not return that session, and helpfulness votes feed the ranking, so the reviews shown are the credible, useful ones the paper found matter most.

Reviews left at the bottom of the page are evidence the shopper never sees.

r = 0.563review valence, the strongest combined effect
156studies, 214 effect sizes
69,006observations
Figure. Key numbers from the paper, and from how the tool applies it.
Paper 11 of 14Smart Reviews · Stage 5: Purchase confidence

The Role of Review Structure in Perceived Helpfulness

UQ Business School, University of Queensland, and Judge Business School, University of Cambridge · Scientific Reports, 2026, a Nature Portfolio peer-reviewed journal · 195,675 real Amazon reviews across 5,487 products.

What the paper is. Yingyue Luna Luan (University of Queensland) and Yeun Joon Kim (University of Cambridge), Scientific Reports, 2026. They analysed 195,675 Amazon reviews across 5,487 products, modelling each review’s opening tone and how its sentiment moved from start to finish, against the helpfulness votes it received. Helpfulness depends on how a review is structured, not only on whether it is positive or negative.

1. What we took from it

  • Helpfulness depends on how a review is structured, not only on whether it is positive or negative.
  • The best structure depends on the product’s rating: for highly rated products, reviews that grow more positive are most helpful; for average-rated products, reviews that move towards criticism help most; for low-rated products, reviews that open constructively before criticising win.
  • How the information is organised matters as much as what it says.

2. Where we used it

The Smart Reviews page, research card 2 (Impact 9/10), and steps 1 and 3 of its method (“Read the reviews”, “Pick the review”).

3. How we used it

Every review is read once and tagged by what it answers (use case, fit, durability, comfort), whether it raises a question and answers it, and the product’s rating context (highly rated or mixed). When a shopper’s buying question is known, the system picks the review that raises and answers that question, choosing a balanced review for mixed-rated products. It never rotates a random five-star review.

How the information is organised matters as much as what it says.

195,675Amazon reviews analysed
5,487products
Figure. Key numbers from the paper, and from how the tool applies it.
Paper 12 of 14Smart Reviews · Stage 5: Purchase confidence

Consumer Reactions to Perceived Undisclosed ChatGPT Usage in an Online Review Context

Goddard School of Business and Economics, Weber State University · Telematics and Informatics, 2024, a peer-reviewed Elsevier journal (the PDF is the SSRN preprint) · Two controlled experiments with about 400 participants in total.

What the paper is. Clinton Amos and Lixuan Zhang (Weber State University), published in Telematics and Informatics, 2024 (the PDF is the SSRN preprint). Two between-subjects experiments on Prolific, with 194 and about 200 participants, in which people read TripAdvisor hotel reviews and were then told the review was written by a person or by ChatGPT. Reviews people believe were generated by ChatGPT are rated less useful, less trustworthy and less authentic than the same reviews attributed to a human.

1. What we took from it

  • Reviews people believe were generated by ChatGPT are rated less useful, less trustworthy and less authentic than the same reviews attributed to a human.
  • The effect held for positive and negative reviews alike, and even when the reviewer carried a TripAdvisor Expert badge.
  • Perceived authenticity explains the drop: once a review does not feel like a real experience, its value collapses.

2. Where we used it

The Smart Reviews page, research card 3 (Impact 9/10), and step 5 of its method (“Keep it honest”).

3. How we used it

AI does real work in the tool: it reads and tags reviews, matches them to the shopper’s question and can summarise product images, descriptions and specifications. But it never wears a customer’s voice. Real reviews are labelled “Customer Review”, generated summaries are labelled “AI Product Insight”, and the two are visually distinct, always. That separation protects the authenticity that makes reviews worth showing in the first place.

AI does real work in the tool, but it never wears a customer’s voice.

≈ 400participants across two experiments
2experiments: the effect held for positive and negative reviews
Figure. Key numbers from the paper, and from how the tool applies it.
Stage 2: Product recommendation

Recommendations

Two papers, one chain: shoppers choose from a small consideration set (Hauser and Wernerfelt), and recommendations sell because they change that set (Li, Grahl and Hinz). The tool rebuilds the shopper’s set on the next visit.

Paper 13 of 14Recommendations · Stage 2: Product recommendation

Consideration Set Effect

MIT Sloan School of Management · Journal of Consumer Research, 1990, one of the leading peer-reviewed journals in marketing · A foundational paper on consideration sets, still cited in choice research today.

What the paper is. “An Evaluation Cost Model of Consideration Sets” by John Hauser and Birger Wernerfelt (MIT Sloan), Journal of Consumer Research, 1990. A theory paper, checked against published data, on why a rational shopper considers only a few of the options available: evaluating every option costs time and effort, so people screen first and choose second. Buying is a two-stage process. Faced with many products, shoppers use a quick screen to form a small consideration set, then decide only within that set.

1. What we took from it

  • Buying is a two-stage process. Faced with many products, shoppers use a quick screen to form a small consideration set, then decide only within that set.
  • The set is small: a median of four shampoos out of more than 30, two to five cars out of more than 160. Across the studies reviewed, considered sets ran from two to eight items while the available range was 6 to 47.
  • The size of the set is a trade-off between the cost of evaluating one more option and the benefit of having it.

2. Where we used it

The Recommendations page, research card 1, and step 2 of its method (“Build the set”). The deck calls it “the research principle we use”.

3. How we used it

Instead of asking the shopper to rebuild their set from scratch on every visit, the store keeps it. Step 1 records what was viewed, explored, saved and added to cart; step 2 ranks those signals by intent (cart, then saved, then viewed), keeps only in-stock items in the right variant, and caps the set at four to six products, the size the paper says shoppers actually hold in mind. The set refreshes as behaviour changes.

People screen first and choose second.

2–8items in a considered set, out of 6 to 47 available
4 of 30+shampoos: the median considered set
Figure. Key numbers from the paper, and from how the tool applies it.
Paper 14 of 14Recommendations · Stage 2: Product recommendation

How Do Recommender Systems Lead to Consumer Purchases?

HEC Paris, University of Cologne and Goethe University Frankfurt · Information Systems Research, the INFORMS journal and one of the top peer-reviewed journals in information systems · A randomised controlled field experiment on a large European online retailer’s live website.

What the paper is. Xitong Li (HEC Paris), Jörn Grahl (University of Cologne) and Oliver Hinz (Goethe University Frankfurt), Information Systems Research. A randomised controlled field experiment on the website of a large European online book retailer, with two pilot lab experiments first, that switched personalised recommendations on for some visitors and off for others, then traced how the effect reached the purchase. Personalised recommendations raised the odds of placing an order by 12.4% and revenue per session (basket value) by 1.7%, measured against shoppers without recommendations.

1. What we took from it

  • Personalised recommendations raised the odds of placing an order by 12.4% and revenue per session (basket value) by 1.7%, measured against shoppers without recommendations.
  • The whole effect ran through the consideration set: recommendations made the set 3.2% larger (breadth) and got shoppers more involved with each product in it (depth), and those two changes produced the purchases. The direct effect was not significant.
  • Breadth mattered far more than depth: showing the right extra options is what moves the sale.

2. Where we used it

The Recommendations page, research card 2 (Impact 9/10), steps 3 to 6 of its method, and the two headline numbers on the page.

3. How we used it

Steps 3 and 4 do the paper’s two jobs. “Welcome back” restores the set the shopper already had, cart item first, above any generic promotion, so the next visit does not start from zero. “You may also like” widens the set outward from what they viewed, same category, adjacent attributes, each item explained (“Because you viewed…”), which is the breadth effect the paper found does most of the work. Step 5 keeps the cart and the set for 30 days across phone and desktop, and step 6 measures recommendations separately: clicks and purchases from recommendations, purchase propensity and basket value, the same two outcomes the experiment measured.

Showing the right extra options is what moves the sale.

+12.4%odds of placing an order
+1.7%revenue per session
+3.2%larger consideration set
Figure. Key numbers from the paper, and from how the tool applies it.