Personalisation in Ecommerce

Ecommerce personalisation is the practice of dynamically tailoring online shopping experiences — product recommendations, search rankings, content, pricing, and communications — to individual shoppers or segments based on their observed behaviour, stated preferences, or inferred attributes. McKinsey research (cited by Algolia, 2024 and Bloomreach, 2026-02-10) reports that companies that grow faster drive 40% more of their revenue from personalisation than slower-growing counterparts, and that personalisation most often drives 5–15% revenue lift.

What it covers

Personalisation in ecommerce spans four major application areas:

Recommendation Engine|Recommendation engines — algorithmic surfaces that present "you may also like", "frequently bought together", "recently viewed", and cross-sell/upsell widgets. Amazon's recommendation engine is widely cited as driving approximately 35% of the company's annual sales (cited across multiple sources including Ringly.io, 2026-06-03; original Amazon primary disclosure unconfirmed). Product recommendations account for 7% of ecommerce site traffic but generate 24% of orders and 26% of total ecommerce revenue (Clerk.io, cited by Ringly.io, 2026-06-03) (as-of 2026-06-03).

Search personalisation — re-ranking search results based on individual browsing history, category affinity, and purchase behaviour. Algolia (2024) reports that customer Decathlon Singapore achieved a 36% increase in click-through rate and a 50% increase in conversion rate after implementing personalised omnichannel search (as-of 2024). Personalised search results convert at 1.8× higher rates than generic search for the same visitors (Econsultancy, cited by Bloomreach, 2026-05-14) (as-of 2026-05-14).

Segment-based content — homepage banners, category landing pages, and email campaigns targeted to behavioural or demographic segments. Emarsys (2025) identifies AI hyper-personalisation, cross-channel data unification, and enhanced mobile personalisation as the leading implementation trends for 2025.

Real-time personalisation — adapting experiences within a session based on signals such as referral source, device, geolocation, and click behaviour, rather than relying solely on historical data. See Architecture for a 2026 technical perspective on why batch-based systems structurally underperform.

Benchmarks

All figures stamped as-of their publication date. CVR lift figures vary widely by source credibility — see Contradictions section.

• McKinsey estimates personalisation can lift retail revenues by 5–15% (cited via multiple secondary sources; McKinsey primary report undated) (as-of 2024)
• BCG (2026) research cited by Bloomreach (2026-05-14) states businesses now see 15–30% conversion lift from AI-powered personalisation (as-of 2026-05-14)

• BCG (older estimate, 2017) measured brands running personalisation programs saw revenues increase by 6–10% and grew 2–3× faster than non-personalising brands; BCG also estimated personalisation would shift ~$800 billion in revenue toward the 15% of companies that get it right (Hello Retail, 2026-07-02; BCG primary 2017) (as-of 2017)

• Named vendor case studies:
  • Huckberry: +9.4% website revenue from AI-driven personalisation (Algolia, 2024) (as-of 2024)
  • Decathlon Singapore: +36% CTR, +50% CVR from personalised omnichannel search (Algolia, 2024) (as-of 2024)
  • Staples Canada: double-digit CVR increase from AI personalisation (Algolia, 2024)
  • Your Surprise: +9% CVR increase alongside significant reduction in manual work (Algolia, 2024)
  • Sur La Table: +11.5% AOV (category), +7.6% AOV (search), +6.6% add-to-cart rate (search) from Bloomreach Loomi AI-powered search and recommendations (Bloomreach, 2026-02-10 and 2025-04-11) (as-of 2026)
  • TFG (specialty retail group): +35.2% online CVR, +39.8% revenue per visit, -28.1% exit rates during Black Friday with Bloomreach conversational AI shopping agent (Clarity) (Bloomreach, 2025-04-11) (as-of 2025)
  • HMV: +14% revenue, +34% impressions, +425% landing page views from Bloomreach AutoSegments for Google Ads personalisation (Bloomreach, 2025-04-11) (as-of 2025)
  • boohooMAN: 25× ROI from Bloomreach predictive personalisation for SMS campaigns (Bloomreach, 2026-02-10) (as-of 2026)
  • United Fashion Group: 43.75% conversion rate via contextual personalisation (right incentive per individual based on past purchases) (Bloomreach, 2026-02-10)
  • Terno (grocery): +27% CVR for an "empty fridge" campaign targeting customers based on purchase timing patterns vs non-personalised campaigns (Bloomreach, 2026-02-10)
  • 260 Sample Sale: doubled CVR using Bloomreach Loomi Agent (Bloomreach, undated case study, 2026)
  • My Jewellery (fashion): +20% email open rates from personalised campaigns derived from interactive style quiz zero-party data via Bloomreach Loomi AI (Bloomreach, 2026-05-14)
  • Petco: +13% site conversions from Constructor personalisation (Constructor, 2025-07-28)
  • Grove Collaborative: 20.07× ROI from personalised search and recommendations (Constructor, 2025-07-28)
  • Bonobos: +92% recommendation-driven conversions after switching to Constructor (Constructor, 2025-07-28)
  • Fresh Clean Threads: +50% abandoned cart revenue, +20% welcome-sequence revenue from Attentive AI-driven SMS personalisation vs human-written static copy (CommerceNext, 2024-06-17) (as-of 2024)

• 83% of consumers are more likely to purchase from a brand that suggests products they recently browsed (Wunderkind 2024) (as-of 2024)
• 31% of consumers say they are more likely to remain loyal to a brand due to personalised shopping experiences (Emarsys/SAP 2025) (as-of 2025)
• A Forrester Total Economic Impact study found Bloomreach customers experienced 251% ROI and $2.3 million in cost savings over three years from AI-powered personalisation (Bloomreach, 2026-02-10) (as-of 2026)
• A Forrester Total Economic Impact study (2026) found Algolia delivered $3.1M NPV over three years, helping commerce teams improve relevance, automate merchandising, and grow revenue (Algolia homepage, 2026) (as-of 2026)
• BCG (cited by Bloomreach, 2026-02-10): personalisation lifts sales by 10% or more and delivers 5–8× ROI on marketing spend (as-of 2026)

Adoption and maturity

• In a 2024 B2C ecommerce survey of 1,100 respondents (Algolia), only 56% of retailers provide personalised shopping profiles and just 46% offer recommendations based on browsed or purchased items (as-of 2024)
• Algolia 6th Annual B2C Site Search Trends report (November 2025, n=1,100): 78% cite AI-powered personalisation as a top search feature; 68% say hyper-personalisation is the consumer capability they most want; 61% plan to implement agentic AI within 12 months; search ranked as #1 digital investment ahead of payments and personalisation (Algolia/Coleman Parkes, 2025-11-24) (as-of 2025-11-24)
• 71% of consumers expect companies to deliver personalised interactions and 76% get frustrated when it does not happen (McKinsey, cited by Bloomreach 2026-02-10 and Hello Retail 2026-07-02) (as-of 2026)
• Salesforce State of the AI Connected Customer (2026-07-29, n=16,585 globally): 73% of consumers now feel treated as individuals, up from 39% in 2023 — the largest two-year jump recorded in the series; simultaneously, 71% feel increasingly protective of personal data and 64% believe companies are reckless with customer data (Salesforce, 2026-07-29) (as-of 2026-07-29)
• Twilio Segment 2024 State of Personalisation: 73% of brands agree AI adoption will fundamentally change personalisation, 88% are budgeting for or planning to adopt AI/ML within the next year, and 86% expect a significant shift from reactive to predictive personalisation (Twilio Segment, 2024) (as-of 2024)
• Twilio Segment 2024: 61% of business leaders are concerned that inaccurate data will compromise the effectiveness of AI-driven personalisation (Twilio Segment, 2024) (as-of 2024)
• Gartner predicts 40% of enterprise apps will feature task-specific AI agents by 2026, up from less than 5% in 2025 — signalling a rapid shift toward agentic personalisation approaches (Gartner, cited by Bloomreach 2026-02-10) (as-of 2026)
• 67% of retailers believe they excel at personalising their website, but only 46% of consumers agree (Sailthru/Marigold 2022) (as-of 2022)

• Deloitte Retail Distribution Outlook (2025): 44% of retail executives want to enhance omnichannel experiences in 2025 (as-of 2025)
• 82% of retailers identify maintaining real-time customer data as their biggest personalisation challenge (Mastercard 2023 Retail Touchpoints Report) (as-of 2023)

• 89% of business leaders call personalisation critical to success in the next three years, but only 35% feel they are successfully achieving omnichannel personalisation (Twilio Segment 2022) (as-of 2022)

Agentic personalisation and AI shopping agents

The 2026 frontier for personalisation is agentic commerce — delegated shopping in which a customer sets intent and constraints, then an AI agent handles discovery, comparison, and purchase on their behalf (nShift, 2026-02-25). This fundamentally changes what "personalisation" means: the system is no longer adapting a human shopper's experience — it is satisfying the requirements of a machine agent acting on the shopper's behalf.

Key 2026 developments (as-of 2026):

• Consumer adoption: Bloomreach/Propeller Insights survey (4,040 US/UK adults, May 2026): 75.4% now use AI tools (Claude, ChatGPT, Gemini) to help with shopping — up from ~61% in 2025; for the first time, more respondents (41.4%) said they would choose to shop via AI over a brand's own website (38%); 60.8% say AI increases their confidence in purchase decisions (Bloomreach, 2026-07-22) (as-of 2026-07-22)
• Consumer intent across channels: 73% of consumers are already using AI in their shopping journey — 45% for product ideas, 37% for review summarisation, 32% for price comparison; only 13% report completing a purchase via AI referral, but 70% are at least somewhat comfortable with an AI agent making purchases on their behalf (BusinessWire global study, October 2025, via commercetools 2026-01-08) (as-of 2025-10)
• 58% of consumers have replaced traditional search with generative AI tools for product recommendations (nShift 2026 Retail Delivery Report, 2026-02-25) (as-of 2026)
• Google: launched agentic checkout across Google Search (AI Mode) and Gemini in 2026; the "Buy for me" button is live with selected US retailers (commercetools, 2026-01-08) (as-of 2026)
• Perplexity: launched free shopping with conversational product discovery, personalised product cards, and instant checkout via PayPal (commercetools, 2026-01-08)
• OpenAI: introduced shopping research in ChatGPT using GPT-5 mini with RL for comparative product guides and real-time feedback loops (commercetools, 2026-01-08)
• Universal Commerce Protocol (UCP): an emerging open standard co-invented by Google, Shopify, Stripe, and Walmart enabling AI agents and external LLMs (ChatGPT, Gemini) to return products directly within conversational shopping experiences; retailers retain control over product ranking and merchandising rules within these AI channels (Constructor, 2026-06-09)
• Morgan Stanley predicts nearly half of online shoppers will use AI shopping agents by 2030, accounting for approximately 25% of their spending (Morgan Stanley, cited by commercetools 2026-01-08) (as-of 2026)
• McKinsey projects agentic commerce could drive as much as $1 trillion in US retail revenue and $3–5 trillion globally by 2030 (McKinsey "The Agentic Commerce Opportunity", October 2025) (as-of 2025-10)

Platform responses: Bloomreach launched a Shopping Agent 2.0 and Marketing Agent described as autonomous agents that can convert a single prompt into a fully built multi-channel campaign workflow (Bloomreach, 2026-05-14). Constructor extended its Recommendation Engine|Commerce Reasoning Engine (CRE) to serve UCP-powered AI agent channels using the same behavioural signals that power on-site search (Constructor, 2026-06-09). Commercetools integrated Stripe's Agentic Commerce Suite via its AI Hub (commercetools, 2026-01-08).

Implication for fashion: agentic commerce changes the fashion discovery dynamic — AI fashion agents autonomously browse, compare fit, price, and delivery terms, making machine-readable product metadata (size guides, fit accuracy, return windows) a prerequisite for selection by AI agents, not just human shoppers (nShift, 2026-02-25).

The autonomy shift: Bloomreach (2025-04-11) describes a transition from "marketing automation" (predefined workflows) to "autonomous marketing" in which AI agents proactively shape and adapt customer journeys in real time without manual configuration of every scenario. Bloomreach's Loomi Marketing Agent (launched 2026-08-05) builds multi-channel campaigns spanning SMS and email in a single conversational flow, applying consent-based channel routing automatically, assigning unique voucher codes, and supporting immediate, scheduled, or recurring send timing — all from a single plain-language request (Bloomreach, 2026-08-05).

AI in advertising personalisation: Meta has shifted its internal messaging to "creative is the new targeting," with its Advantage Plus system moving toward a fully autonomous model where marketers only set goals and budgets (AAAI 2026 Keynote via Marketing Mondays, 2026-01-27). Google DeepMind's published approach to AI-personalised content generation frames it as human–AI co-creation using a "super prompt" combining branding, audience goals, and inspiration board (AAAI 2026 Keynote, 2026-01-27).

Architecture: real-time vs batch

Most current ecommerce personalisation implementations have three structural failures (Vespa AI EMEA webinar, 2026-04-27):

1. Stale data — 24-hour batch cycles mean the data driving personalisation is always yesterday's behaviour, not the current session
2. Group-level personalisation — recommendations are by user group (segment), not by individual
3. Fragmented systems — search, ranking, and personalisation live in separate systems that cannot share signals

The architectural alternative is a unified real-time system that (Vespa AI, 2026-04-27):

• Models all data (catalogue, behaviour, content, context) as a unified vector/tensor representation
• Captures every user signal (click, dwell, add-to-basket, filter, return) in real time (milliseconds)
• Ranks results per individual shopper at query time against fresh data, not batch data

Sparse feature tensors — human-readable named dimensions such as "color: black 5, style: minimal 3" — are directly updatable without model retraining, making them preferable to dense embeddings for explainable, real-time personalisation (Vespa AI, 2026-04-27). In the Vespa architecture, every user interaction transfers a weighted portion of that product's feature vector into the user's "taste tensor" profile — meaning the user model is continuously updated per interaction.

The cold-start problem is described as "very minimal" in a sparse-tensor personalisation system because meaningful personalisation is visible in product ranking after only a few clicks (Vespa AI, 2026-04-27).

Constructor trains recommendation models on 100% of onsite clickstream data (every search query, facet click, add-to-cart, bounce, filter applied) and predicts a product's likelihood to convert for each shopper via a metric called "Attractiveness," which replaces simple keyword relevance matching (Constructor, 2025-07-28). A platform capable of 100,000 partial-index writes per second per node can reflect all session signals in the next ranking decision in milliseconds (Vespa AI, 2026-04-27).

In fashion ecommerce, semantic Vector Search using Vision-Language Model (VLM) feature extraction outperforms CLIP models for queries involving images with people or complex backgrounds (Vespa AI, 2026-04-27). The Vespa demo correctly retrieved "trousers with hearts" and "earrings with parrot" from a fashion catalogue including images with human models.

Personalisation logic and business rules (destocking high-inventory items, boosting revenue-maximising products) must be merged within the same ranking expression rather than run as separate services — otherwise relevance optimisation and business goal optimisation degrade each other (Vespa AI, 2026-04-27).

Prediction ≠ decisioning: A CommerceNext 2024 keynote distinguished between affinity prediction (modelling what a user likes) and decision-making (whether marketing that product will change their behaviour). High category affinity does not imply marketing that category will incrementally move purchases (CommerceNext, 2024-06-12) (as-of 2024).

EU privacy and GDPR compliance

The 2026 environment for personalisation in the EU has three simultaneous pressures (Virtual Marketer, 2026-08-04):

1. Browser technical restrictions — progressive removal of third-party cookie access
2. Increased regulatory scrutiny — the EDPB's 2026 Coordinated Enforcement Framework specifically targets transparency and information obligations, placing ecommerce GDPR compliance under direct scrutiny from 25 Data Protection Authorities (Virtual Marketer, 2026-08-04) (as-of 2026-08-04)
3. Consumer sensitivity — 64% of consumers believe companies are reckless with customer data; 71% feel increasingly protective of personal data (Salesforce, 2026-07-29) (as-of 2026-07-29)

Under GDPR, non-essential cookies (advertising, analytics, personalisation) cannot be set until the visitor has given explicit, informed consent via active agreement — pre-checked boxes or continued browsing do not satisfy the requirement (Virtual Marketer, 2026-08-04).

The EU AI Act (entering practical effect in 2026) means most marketing personalisation use cases are expected to fall into the "minimal risk" category, but systems that specifically exploit vulnerabilities of certain groups to influence purchasing behaviour can slide into a higher-risk category requiring stricter obligations (Virtual Marketer, 2026-08-04).

Privacy-compliant personalisation techniques identified for 2026 (Virtual Marketer, 2026-08-04):

• First-party and Zero-Party Data strategies (consented preference collection — see data strategy section below)
• Granular Consent Management Platform (CMP)|consent management — consent passed consistently to all downstream systems (CDP, marketing automation, AI models)
• On-device and edge processing to prevent raw data leaving the device
• Pseudonymisation of user profiles
• Synthetic data for model training

Data platform maturity: 72% of companies now use a Customer Data Platform (CDP) for personalisation and 48% use a data warehouse (Twilio Segment 2024) (as-of 2024). Only 40% of consumers trust brands to keep their personal data secure (Twilio Segment 2022) (as-of 2022).

Fashion ecommerce

• 45% of fashion executives identify AI-driven marketing personalisation as a major value driver for 2025 (McKinsey State of Fashion 2025 webinar) (as-of 2025)
• My Jewellery (Dutch retailer) gamified Zero-Party Data collection via an interactive style quiz powered by Bloomreach Loomi AI, using heart-and-X style profiling to build consented preference data; personalised email campaigns derived from this data achieved +20% email open rates (Bloomreach, 2026-05-14) (as-of 2026)
• Algolia (2024) notes fashion retailers can deliver style-specific personalisation — surfacing products matching a customer's known preferences for shoe colour and size — using purchase and browse history (Algolia, 2024)
• In fashion ecommerce, AI hyper-personalisation means models analysing browsing behaviour, purchase history, geographic location, and social media sentiment to curate a unique homepage for every visitor; leading brands are moving beyond A/B testing into real-time personalisation where every product image is algorithmically selected for an individual's context (InventorySource, undated)

• McKinsey's "Rewiring Retail in Europe: The AI Imperative" (2026-06-10) includes a case study of Zara's AI platform identifying emerging trends 3–4 weeks faster than traditional methods — a fashion-specific application of AI that directly feeds downstream personalisation (McKinsey, 2026-06-10) (as-of 2026-06-10)
• For agentic commerce, AI fashion agents autonomously browse, compare fit, price, and delivery terms, making machine-readable product metadata (size guides, fit accuracy, return windows) a prerequisite for selection — not just for human UX (nShift, 2026-02-25)

Data strategy and implementation

The shift toward privacy-compliant personalisation has elevated two data types (Algolia, 2024):

• Zero-Party Data — information voluntarily supplied by shoppers (quiz responses, preference toggles, style profiles). Consent is explicit.
• First-party data — observed browse, cart, and purchase behaviour collected on an opt-in basis. No third-party dependencies.

Algolia's 2023 Ecommerce Site Search Trends report (via their 2024 blog post) found that more than half of retailers developing personalisation in-house recognised they could not evolve it fast enough to keep pace with market expectations.

Algolia AI Personalisation (launched public beta June 2024) generates user profiles consisting of affinities — attributes of products each individual user has engaged with — and applies them at query time to re-rank results, covering personalised search/browse, autocomplete, facet reordering, recommendations, inline segmentation, and promotions by segment (Algolia, 2024-06-04) (as-of 2024-06-04).

Shopify (September 2025) introduced customer segmentation by product category taxonomy, allowing merchants to group customers by the types of products they've viewed or purchased using over 16,000 defined product categories from Shopify's standard taxonomy; segments auto-include parent categories (e.g., a buyer of "Facial Cleansers" appears in "Beauty" and "Health & Beauty" segments) and work with Shopify Email, targeted discounts, and any marketing tool supporting customer segments (Shopify Changelog, 2025-09-19) (as-of 2025-09-19).

Shopify Spring '26 Edition (June 2026) announced 150+ platform updates positioning "AI chats" as a first-class sales surface alongside online, in-store, and social channels (Shopify Changelog, 2026-06-17) (as-of 2026-06-17).

Constructor positions personalised recommendations as part of a unified "search to sales" strategy — behavioural signals from search queries refine recommendation accuracy, and a single recommendation engine also powers AI shopping agents, retail media, and cross-channel email/SMS content (Constructor, 2025-07-28).

Gartner 2026 must-have capabilities for personalisation engines (Gartner, cited by Bloomreach 2026-02-10): embedded generative AI for content creation and optimisation; real-time digital behaviour tracking and activation; automated ML that improves outcomes over time; extensive testing capabilities (A/B, multivariate, multi-armed bandit); customer experience data profile management. Both Algolia and Constructor were named Leaders in the Gartner Magic Quadrant for Search and Product Discovery 2026 (Algolia, 2026; Constructor, 2026-06-09) (as-of 2026).

Market size

Additional market estimates (all from secondary/aggregator sources — use with caution):

• Customer experience personalisation software: projected to reach $11.6 billion by 2026, up from $7.6 billion in 2021 (Statista, cited by Ringly.io, 2026-06-03) (as-of 2026-06-03)
• Global Recommendation Engine market: projected to reach $15.13 billion by 2026 from $2.12 billion in 2020 (Clerk.io, cited by Ringly.io, 2026-06-03) (as-of 2026-06-03)
• Hyper-personalisation market: projected to reach $80.2 billion by 2032 at 18.1% CAGR (Precision Business Insights, cited by Ringly.io, 2026-06-03) (as-of 2026-06-03)
• Broader AI-in-ecommerce market: $9.01 billion in 2025, projected to exceed $64.03 billion by 2034 at 24.34% CAGR (Precedence Research, cited by Emarsys, 2025) (as-of 2025)
• AI agent-mediated commerce: McKinsey projects $3–5 trillion globally by 2030 (McKinsey, 2025-10-17)

Contradictions

Key terms

Term	Meaning
Recommendation engine	Algorithm surfacing products to individual users based on behaviour, similarity, or purchase history
Zero-party data	Data explicitly volunteered by the customer (e.g. quiz results, style preferences)
First-party data	Behavioural data collected by the retailer on opt-in (browse, cart, purchase events)
Hyper-personalisation	Real-time, individual-level personalisation using AI and streaming data, beyond segment-level targeting
Personalisation maturity	A framework for assessing how advanced a retailer's personalisation capabilities are, from manual segmentation (L1) to AI-driven real-time (L3+)
Cold-start problem	The challenge of personalising for new or anonymous shoppers with no historical data; described as "very minimal" in sparse-tensor architectures (Vespa AI, 2026)
Agentic personalisation	Use of autonomous AI agents to deliver tailored experiences at scale, with the agent independently optimising touchpoints rather than following predefined rules (Bloomreach, 2025-04-11)
Taste tensor	A user profile representation built from weighted feature vectors of products the shopper has interacted with, updated in real time with each interaction (Vespa AI, 2026)
Attractiveness	Constructor's metric for a product's likelihood to convert for a specific shopper, replacing keyword relevance matching (Constructor, 2025-07-28)
Universal Commerce Protocol (UCP)	Open standard co-invented by Google, Shopify, Stripe, and Walmart enabling AI agents to return and purchase products from merchant catalogues within conversational interfaces (Constructor, 2026-06-09)
Autonomous marketing	The transition from predefined marketing automation workflows to AI agents that proactively shape and adapt customer journeys in real time (Bloomreach, 2025-04-11)
Next best action (NBA)	Traditional personalisation approach that collapses first-party data into segments and applies rules per segment — criticised for losing individual-level signal and relying on "next best guess" rather than decisioning ML (CommerceNext, 2024-06-12)