Is your strategy keeping up with the evolution of retail?
A century of interaction pattern shifts, the generations that drove them, and the companies that owned a category and lost it — this is a working brief for people who build, ship, and market digital products and services in retail domain.
The chart below places each commerce format at the point it reached prominence, on the row of the generation that adopted it first. The ring around each point represents the time and convenience it unlocked. The rings grow. The gaps between them shrink.
Section 01
The map: a century of commerce, plotted against the people who adopted it
Every commerce format ever invented has sold the same thing: SAVE TIME. The self-service store sold the time you used to spend waiting at a counter. The hypermarket sold the trips you no longer had to make. E-commerce sold the journey. Quick commerce sells the twenty minutes between wanting something and having it. And the formats now emerging sell something more valuable still — the effort of deciding.
If you build or market software, this is not a retail story. It is the longest continuous dataset we have on what happens when the interface between a company and its customer changes — and every one of those changes rhymes with a platform shift you have already lived through, or are about to.
The self-service store is the moment packaging had to sell without a salesperson, which is the same problem as a product page carrying a self-serve signup with no SDR attached. A&P's price war is every growth team discounting into a cost base that cannot support it. Sears sitting on stores, a catalogue and a credit book while failing to invent e-commerce is the canonical incumbent-with-all-the-pieces story. And the current agentic shift is the first time in this century that the thing evaluating your product is not a person — which changes what "positioning" and "conversion" even mean.
Most roadmaps are still built around the wrong question. They ask which channel or touch-point should we add? The more useful question, and the one this paper is organised around, is: which piece of the user's effort are we removing or shrinking, and is anyone removing a bigger chunk of it?
What follows: a map of the century against the generations who drove it, the model / product / marketing pattern that worked in each era, the structural reasons category owners lost, why behavioural science and human-centred design became core product disciplines rather than marketing accessories, and a Wardley mapping framework for keeping pace deliberately rather than reactively.
The evolution of commerce, 1900–2050
Each commerce format plotted at the point it reached prominence, on the row of the generation that adopted it first. Ring size represents the time and convenience it unlocked.
Ring diameter is indicative, not measured — it represents the relative gain in time and convenience each format delivered over the one before it. Subscription commerce is drawn off the generation lines with a dashed ring because it is not a place to shop; it is a pricing and fulfilment layer that attaches to whichever channel is already running.
Three things the shape tells you
The formats are compressing. Forty-seven years separate the self-service store from the supermarket's dominance. Fifteen separate e-commerce from mobile commerce. Two separate social commerce from quick commerce. A shift that once took a generation now takes less time than an enterprise replatforming project. If your roadmap horizon is three years and your architecture commits you for five, the compression is already working against you.
Adoption is generational, but obsolescence is not. Every format on this chart is still trading. Supermarkets did not die when hypermarkets arrived. What changes is not survival but relevance at the margin — the format stops winning new demand and starts defending old demand, usually on price. In software the same curve shows up as healthy revenue with deteriorating new-logo acquisition and rising discount depth: the retention metrics look fine right up until the cohort that never adopted you becomes the whole market.
The effort being removed keeps moving up the stack. Early formats removed physical effort: walking, carrying, travelling. Middle formats removed transactional effort: comparing, queueing, going. Current formats remove cognitive effort: choosing, remembering, deciding. This is the same climb your own product category is making — from doing the task, to organising the task, to deciding whether the task is needed. Each step up is harder to defend, because each step requires understanding the user more intimately than the last, and intimacy is the one input a competitor cannot buy off the shelf.
Section 04
How brand marketing changed, era by era
Prominence 1995 — 2010
E-Commerce
Business model
Negative working capital — collect from the customer before paying the supplier — combined with an infinite shelf and gross profit recycled straight back into selection, price and logistics. The model is not "sell online"; it is "compound faster than the incumbent can react".
Product strategy
Discovery tools no physical store could offer: search, filters, reviews, recommendations, purchase history. The long tail becomes economically viable for the first time.
Marketing strategy
SEO, affiliates and word of mouth, all resting on delivery reliability. The marketing was the operations; a parcel arriving when promised did more for acquisition than any campaign.
Failure learning
Webvan built automated distribution centres across multiple cities before proving repeat demand in one, and burned roughly $800 million in under three years. Pets.com spent more to acquire a customer than that customer would ever return. Boo.com launched a site most consumers' connections could not load. Build demand proof before fixed cost, and never let acquisition cost outrun contribution margin. Note also that Webvan's idea was correct — grocery did move online. Being right about the destination is worthless if you arrive with an empty tank.
For each format we have touched upon the business model that made the money, the product strategy that made it distinctive, the marketing strategy that made it known — and the companies that had every advantage and lost anyway.
Read the failures first. Success in this century is frequently a matter of timing, which does not transfer. Failure is almost always a matter of structure, which does. Every collapse below has a direct analogue in a software business, and the analogue is usually one your own roadmap is currently exposed to.
Prominence 1916 — 1930
Self-service
Business model
Move the labour of selection from the clerk to the customer. The store gives up service and buys back throughput: more baskets per hour, fewer staff per basket, margin earned on volume rather than on advice.
Product strategy
A fixed circulation route, goods pre-packaged and pre-priced, and — critically — every item legible without help. Packaging stops being a container and becomes the sales pitch.
Marketing strategy
The store itself is the advertisement. Turnstiles, one-way aisles and the sheer novelty of touching the merchandise drew crowds. Manufacturer brands, freed from the clerk's recommendation, began advertising directly to the shopper for the first time.
Failure learning
Clarence Saunders invented the format and lost Piggly Wiggly in a 1923 stock-market fight, then spent the next two decades on Keedoozle — a fully automated, coin-and-conveyor grocery store that failed repeatedly because the machinery could not keep up with a busy Saturday. He automated ahead of the technology and ahead of the demand. Inventing the category confers no protection against losing it, and the second act is usually where the capital burns.
Prominence 1930 — 1950
SUPERMARKET
Business model
Cash-and-carry, high volume, thin margin, cheap peripheral real estate. Michael Cullen's original arithmetic was explicit: take rent out of the equation by moving to the edge of town, and pass the saving to the shopper in return for footfall.
Product strategy
Every grocery category under one roof, so the shopper stops making four trips. Own-label enters as the margin defence against national brands that now advertise over the retailer's head.
Marketing strategy
Weekly price advertising in the local press, loss leaders on known-value items, and price image built through a handful of products the shopper could actually compare.
Failure learning
A&P was, for a period, the largest retailer on earth, and it disappeared. Its 1972 "WEO" price war cut prices across the estate without cutting the cost base that supported them, destroying margin without shifting share durably. It also stayed in small urban stores while its customers moved to the suburbs. Price is only a weapon if your costs are structurally lower than the competitor you are aiming it at. Otherwise it is a countdown.
Prominence 1950 — 1963
CHAIN RETAIL
Business model
Design one store, then replicate it. Central buying, standardised fit-out and shared systems amortise across hundreds of sites; the unit economics improve with every opening rather than with every sale.
Product strategy
Consistency as the offer. A disciplined range, reliably in stock, identical in every town — which is exactly what a newly mobile, newly suburban population wanted.
Marketing strategy
National brand building on television, executed locally through flyers and catalogues. The chain's name becomes a guarantee, which is the first time in this story that the retailer is the brand.
Failure learning
Sears held stores, a mail-order catalogue and a consumer credit book — every component of e-commerce, assembled decades early. It defended the existing P&L instead of cannibalising it. Woolworth owned the high street and could not answer the discounters. Scale advantage decays the moment a competitor arrives with a lower-cost operating model, and incumbency makes that competitor invisible for longer than it should.
Prominence 1963 — 1980
HYPERMARKET
Business model
Non-food margin subsidises food price perception. The shopper comes for cheap groceries and leaves with a television. Large basket, low frequency, and a catchment wide enough to make the format defensible against anyone who cannot match the land.
Product strategy
Everything, in one trip. Deep own-label ranges across tiers, destination categories to justify the drive, and enough non-food to make the visit feel like an outing rather than an errand.
Marketing strategy
Price leadership claimed loudly and demonstrated on a comparison basket, plus catchment dominance — be the only store worth driving to within thirty minutes.
Failure learning
Walmart withdrew from Germany and South Korea in 2006; Carrefour exited the United States; Tesco spent five years and a great deal of money on Fresh & Easy in California before closing it in 2013. In each case a proven format was exported without re-reading local shopping frequency, basket size, store-opening law and labour cost. Formats travel badly. Operating principles travel well. The discipline is to export the second and rebuild the first.
Prominence 1980 — 1995
big-box retail
Business model
Dominate one category so completely that suppliers negotiate on your terms. The real product is inventory turn: buy well, sell fast, finance the next order out of the last one.
Product strategy
Unmatched depth in a single category, plus the services that make a complex purchase safe — delivery, installation, warranty, knowledgeable staff.
Marketing strategy
Price authority plus assortment authority: if we don't have it, nobody does. Advertising shifts from persuasion to reassurance.
Failure learning
Circuit City removed thousands of experienced commissioned salespeople in 2007 to reduce payroll, and with them the only reason to visit a shop selling identical televisions to everyone else; it filed for bankruptcy within two years. Toys R Us emerged from a 2005 leveraged buyout with debt so heavy it could not fund a credible online business, and liquidated in 2018. Cutting the differentiator to protect the margin is the fastest exit from a category, and a balance sheet that cannot fund reinvention is a strategy decision disguised as a finance decision.
Prominence 2005 — 2015 · layer
SUBSCRIPTION
Business model
Recurring revenue in exchange for a habit. Higher lifetime value, far more predictable demand planning, and a customer who stops comparison-shopping. Note that this is not a channel — Costco was running the mechanic decades before the internet, and Prime now runs across web, app and agent alike.
Product strategy
Attach the subscription to something genuinely habitual — delivery, replenishment, content — and make the membership improve everything else the business sells.
Marketing strategy
Free trials feeding an ecosystem rather than a standalone product. The offer is not the discount; it is the removal of a recurring decision.
Failure learning
Blue Apron and Birchbox both faced churn that outran acquisition spend. MoviePass sold an offer that lost money on every active user and scaled the loss. Amazon Prime worked because it made everything else Amazon sold stickier — the subscription was the flywheel's axle, not a product. A subscription unattached to a flywheel is a discount with a billing cycle. And the regulatory floor has risen sharply: Amazon settled an FTC action over Prime enrolment and cancellation for $2.5 billion in September 2025, and although the Eighth Circuit vacated the FTC's click-to-cancel rule in July 2025 on procedural grounds, the Commission restarted rulemaking in January 2026 while state statutes and class actions continued regardless. Designing friction into cancellation is now a litigation exposure, not a retention tactic.
Prominence 2010 — 2016
MOBILE COMMERCE
Business model
Session frequency replaces store visits. The phone converts shopping from an occasion into an interstitial activity, and the push notification becomes owned media the retailer does not pay for by impression.
Product strategy
One-tap checkout, stored payment, saved baskets, reorder lists, location awareness. Every removed tap is measurable revenue, which makes design an engineering discipline rather than a taste question.
Marketing strategy
App-install campaigns — but only defensible when the app makes repeat purchase genuinely faster. Otherwise the retailer is paying to acquire an icon nobody opens.
Failure learning
The great majority of retailer apps in this period were the website in a shell, and were deleted within a week. An app has to earn its place on the home screen with something the mobile web cannot do — stored preference, offline access, scan-and-go, a loyalty identity. Install counts were the vanity metric of the decade; the honest metric was the ratio of app orders to app installs.
Prominence 2020 — 2022
SOCIAL COMMERCE
Business model
Content becomes both the acquisition channel and the merchandising. Creators carry the customer acquisition cost in exchange for commission, which converts a fixed marketing budget into a variable one.
Product strategy
Short-form video, live selling, group buying and in-feed checkout — formats where discovery is passive and purchase is impulsive, which is the opposite of search-driven e-commerce.
Marketing strategy
Algorithmic reach plus affiliate economics. The brand's own channel matters less than its ability to be talked about credibly by someone else.
Failure learning
Facebook ended live shopping in the US in 2022 and Instagram followed in 2023, both after adoption stalled well short of the Chinese comparison that justified the investment. Wish scaled rapidly on very cheap goods and lost customers to delivery times and return experiences it never fixed. The commerce layer only holds if the fulfilment behind it is as good as the content in front of it — and a behaviour that works in one market's ecosystem does not automatically port to another's.
Prominence 2016 — 2020
OMNICHANNEL
Business model
The store becomes a fulfilment node. Existing real estate — previously a fixed cost under pressure — turns into a last-mile advantage that pure-play competitors have to buy from scratch.
Product strategy
Click-and-collect, curbside, ship-from-store, return anywhere, and a single customer identity across all of it. The technical requirement underneath is unglamorous and non-negotiable: accurate, real-time, store-level inventory.
Marketing strategy
Loyalty data unified across channels, allowing personalisation at household rather than segment level — and, increasingly, retail media sold against that data.
Failure learning
Sears ran Shop Your Way, a genuinely sophisticated loyalty programme, on top of stores customers no longer wanted to enter. Many European grocers launched collection points that failed on substitution accuracy and pick quality rather than on technology. Omnichannel is an operations and inventory-accuracy problem first and a CRM problem second. Data layered over a broken store experience changes nothing except the precision with which you observe the decline.
Prominence 2022 — 2025
QUICK COMMERCE
Business model
Dark-store density. Profitability arrives per store rather than per company: each new store loses money until it matures, so growth and losses rise together until the network reaches critical density. Advertising and retail media revenue then close the remaining gap.
Product strategy
Small, urgent baskets delivered in ten to thirty minutes, on a ruthlessly curated range. Expansion into higher-margin categories follows once the delivery promise is trusted.
Marketing strategy
Reliability of the promise is the marketing. After that, basket-building — because the economics only work when the average order value rises.
Failure learning
Getir, Gorillas, Jokr and Flink all collapsed in Europe and the United States by 2024. The identical ten-minute model became one of the largest stories in Indian retail: Blinkit held roughly 46–48% share in early 2026, with Swiggy Instamart near 24% and Zepto near 22% of a sector valued around $11.5 billion at the end of 2025, running at roughly 7.8 million orders a day. Bernstein's analysis found around 3,600 of the top 3,800 stores in the largest eight cities were profitable while tier-two stores continued to lose money. Unit economics here are geographic, not strategic. Delivery labour cost and urban density decide the outcome — the model does not travel, the density does. Before importing a format, calculate the households within a fifteen-minute radius and the cost of the rider, and let that arithmetic overrule the deck.
Prominence 2025 — 2035
AI AGENTIC COMMERCE
Business model
Still forming, and worth watching rather than betting the estate on. The shape that survived 2026's first contact with reality is: discovery moves to the assistant, the transaction stays on the retailer's own surface. Amazon's Rufus, by contrast, keeps both inside the walled garden — it served around 300 million users and was credited with roughly $12 billion in incremental sales in 2025.
Product strategy
A structured, machine-readable catalogue with real-time inventory and pricing, plus agent-authentication and payment protocols. Google's Universal Commerce Protocol launched at NRF 2026 with Shopify, Etsy, Wayfair and Target; OpenAI's Agentic Commerce Protocol and Stripe's Agentic Commerce Suite arrived in late 2025. Multi-protocol implementations are already outperforming single-protocol ones.
Marketing strategy
Citation replaces ranking. Being named by the assistant is the new shelf position, and AI-referred traffic converts better precisely because the shopper arrives having already compared and decided. Walmart captures roughly 36% of ChatGPT referral traffic while Amazon, having blocked the crawlers, captures almost none.
Failure learning
OpenAI launched Instant Checkout with Etsy in September 2025 and effectively abandoned it by March 2026. Forrester identified three compounding causes: no real-time inventory infrastructure, no system for collecting and remitting state sales tax, and a conversion wall — Walmart measured in-chat checkout converting roughly three times worse than a click-through to walmart.com, even as ChatGPT delivered around twice the new-customer rate of search. Amazon meanwhile blocked dozens of agents and sued Perplexity in November 2025. Checkout is a hard, boring, regulated system, and a conversational interface does not make the hard parts disappear. Optimise for being discovered by agents; do not surrender the transaction, the tax obligation or the customer record to one.
Prominence 2035 — 2050
AMBIENT COMMERCE
Business model
Speculative. Replenishment triggered by context — the fridge, the car, the meter — with the retailer paid for reliability rather than for discovery. Whoever owns the default reorder owns the category, and owns it without an advertising budget.
Product strategy
Sensors, connected appliances, and standing household rules with a visible, easy override. The override is the product, not an afterthought.
Marketing strategy
Trust maintenance rather than persuasion, because there is no longer a moment of consideration in which to persuade.
Failure learning
The precursors already failed instructively. Amazon retired Dash Buttons in 2019, and in 2024 pulled Just Walk Out from its larger Fresh stores in favour of smart carts that keep the shopper in the loop. Shoppers will surrender the repeat purchase long before they surrender the decision. Automate the boring reorder; leave the choice — and visible, reversible control over it — with the person.
Section 03
What worked, and what the failures taught
Brand marketing did not evolve gradually. It changed shape each time the retail format changed who was standing between the product and the person.
When the clerk disappeared, the brand was invented. Self-service removed the only person in the store who could recommend anything, and the package had to do that job instead: name, colour, mark, claim. Nearly everything we now call packaging design descends from that vacancy.
When mass media arrived, the decision moved upstream. Radio and then television let a manufacturer build preference before the shopper reached the aisle, so the shelf only had to confirm a choice already made. This is also when the first formal claim appeared that people do not buy for the reasons they give — Ernest Dichter's motivational research applied depth interviewing to consumer goods and made "why" a research question rather than an assumption.
When chains standardised, the retailer became a brand. Sameness across towns was itself the proposition, and the retailer's name started carrying a guarantee that individual products did not need to.
When the hypermarket arrived, brands had to defend a position. Private label competed on the retailer's own shelf using the retailer's own data, which forced national brands out of description and into positioning — named as a discipline by Ries and Trout in 1972. The same decade gave marketing its theoretical spine: Kahneman and Tversky's work on heuristics and biases in 1974, and prospect theory in 1979, made loss aversion, anchoring and framing nameable and testable.
When e-commerce arrived, the brand stopped being what you broadcast. This is the deepest break in the century. Reviews turned social proof from a persuasion technique into a structural feature of the storefront, and a brand became, functionally, what strangers said about it. Simultaneously, Nielsen's usability heuristics in 1994 and the arrival of cheap A/B testing made behavioural claims measurable for the first time. You no longer argued about the button. You tested it.
When mobile arrived, the brand shrank to an icon. Attention rather than shelf space became the scarce resource, and design quality moved directly onto the revenue line — on a small screen operated by a thumb, poor design is not ugly, it is a conversion loss.
When social commerce arrived, the brand was borrowed. Parasocial trust in a creator transfers to the product, which means the retailer depends on a relationship it does not own and cannot fully control.
When agents arrived, the audience stopped being exclusively human. A brand now has to be legible to a machine intermediary as well as attractive to a person: structured data, clean pricing, live availability, protocol compliance. And the choice architecture has moved inside the agent, which means for the first time the company does not control the framing of its own offer.
What this means for a marketing team in tech
Each of those shifts moved budget from one function to another, and the shift always looked optional for about eighteen months before it looked mandatory. The current one is doing the same. Three practical consequences:
Your documentation is now marketing collateral. When an assistant answers "what should I use for X", it synthesises from docs, changelogs, comparison pages and forum threads — not from your homepage. The pages your team deprioritises are the ones the model reads. Structured, accurate, machine-legible product information is a demand-generation asset with a line item, not a developer-experience nicety.
Citation is the new ranking, and it is measured differently. Position in a results page was observable and gameable. Being named inside a generated answer is neither. The tractable metrics are share of answer across a defined question set, the accuracy of what the model says about you, and the conversion rate of assistant-referred traffic — which typically runs higher than search, because the visitor arrives having already compared.
Positioning has to survive compression. An agent will reduce your product to a sentence and a handful of attributes. If your differentiation only exists in a narrative that takes three paragraphs to land, it will not survive the summarisation. Test your positioning by asking whether it holds up in one line, stated by someone who does not work for you.
Section 05
Why behavioural science and human-centred design kept gaining ground
There is a single sentence that explains the entire trajectory:
Behavioural science and human-centred design grew in importance in exact proportion to how much control the retailer lost.
When you owned the store, you could arrange the shelf. Placement, adjacency, lighting and route were levers you controlled outright, and a reasonable amount of commercial success came from operating them competently. When the shopper moved to a website, then a phone, then a feed, then an assistant, those levers were handed to someone else one at a time. What remained was the ability to understand the person well enough to be chosen in an environment you do not own — and that is precisely what these two disciplines provide.
Three phases, and the line between them
Phase one: intuition (1916–1970) Practitioners applied behavioural principles without a vocabulary for them. One-way aisles, eye-level placement, the bakery at the entrance. Post-war human factors research established that environments should adapt to people rather than the reverse, but this was aimed at cockpits and control rooms, not shops.
Phase two: formalisation (1970–2005) Kahneman and Tversky provided the theory. Cialdini catalogued the mechanisms of influence in 1984. Paco Underhill put cameras in stores and demonstrated that real shoppers behaved almost nothing like the planogram assumed. Don Norman published The Design of Everyday Things in 1988 and IDEO was founded in 1991, establishing human-centred design as a named practice — though still oriented towards products rather than purchasing.
Phase three: instrumentation (2005–present) Digital commerce made every behavioural hypothesis testable at scale and in production. Nudge arrived in 2008 and gave choice architecture respectability in the same period that Harry Brignull began cataloguing dark patterns. Fogg's behaviour model and Eyal's Hooked turned trigger design and variable reward into product requirements. Service design and journey mapping moved out of the agency deck and into operations, because in an omnichannel business the failure points sit in the seams — the collection slot, the substitution, the return.
The ethical inflection, which is now also a commercial one
The same knowledge that lets you design for someone lets you design against them. The default effect that makes a subscription convenient is the identical mechanism that makes it hard to leave; the scarcity cue that helps a shopper decide is the identical mechanism that manufactures regret.
For two decades this was treated as a matter of taste. It is now a matter of law and of balance sheet. Regulators have moved to treat deceptive interface design as a substantive violation rather than a stylistic one, cancellation flows are being assessed against the ease of the sign-up they mirror, and the Amazon Prime settlement in 2025 put a nine-figure number against the practice. The strategic reading is straightforward: behavioural technique applied against the customer's interest is now a liability that accrues quietly and settles expensively.
Which brings the discipline back to where human-centred design started. In an agentic and ambient future, the relationship with the customer is mediated by systems you do not own, acting on standing permissions. In that world, the only durable asset is a user who trusts the default you set for them — and trust is not a campaign. It is an accumulated record of decisions made in their favour when nobody was checking.
Where this lands on a product team's backlog
The practical translation is that three things stop being design polish and become product requirements with owners and acceptance criteria. Cancellation and downgrade flows must be no harder than the flows that created the commitment — this is now a compliance surface, not a retention lever. Consent and permission scopes need to be legible in the moment they are granted, particularly for anything an agent will later act on. And every automated action needs a visible override, because the failure mode of ambient and agentic products is not inaccuracy; it is the user discovering something happened that they would not have chosen.
The team-level test is simple and worth adopting verbatim: would we be comfortable explaining this mechanism to the user it is aimed at? If the honest answer is no, you have not found a growth lever. You have found a liability with a delay fuse.
Section 06
The three failure patterns
Read the whole century of failures together and they reduce to three recurring structures. Almost every collapse in Section 03 is one of these.
Right idea, wrong cost base. A&P's price war, Circuit City's staff cut, Webvan's warehouses, MoviePass's pricing. The strategy was directionally correct; the economics underneath could not carry it. The diagnostic question is not is this what customers want? but can we still make money when they get it?
Right format, wrong geography. Walmart in Germany, Tesco in California, Getir in Europe, live shopping outside China. A format encodes a set of local assumptions — density, labour cost, trip frequency, planning law, payment habit — that are invisible until they are absent. Export the operating principle, rebuild the format.
Automating a decision the customer wanted to keep. Keedoozle, Dash Buttons, Just Walk Out in large stores, in-chat checkout. Convenience is welcome right up to the point where it removes agency, and the boundary is consistently earlier than technologists expect. People will delegate the errand long before they delegate the choice.
Section 07
A framework for keeping pace: mapping the commerce stack
The commerce stack as of 2026. Red arrows show direction of travel, not certainty of arrival. Note the anomaly at the centre: real-time inventory and price is a component everyone depends on and almost nobody has industrialised, which is why it appears in the failure analysis of omnichannel, quick commerce and agentic commerce alike.
Everything so far is diagnosis. The framework below is the treatment, and it borrows Simon Wardley's mapping method because it is the only strategy tool in common use that puts movement on the page. Roadmaps show sequence. Two-by-twos show position. Neither shows the one thing that decides whether your bet is early, correct or already commoditised: which direction each component is travelling, and how fast.
The two axes, briefly
A Wardley map is a value chain plotted against evolution. The vertical axis is the chain of dependencies, anchored at the top by a user need and descending through progressively less visible components. The horizontal axis is how evolved each component is, moving left to right through four stages: genesis (novel, uncertain, built by hand), custom-built (understood, still bespoke per company), product (a market of comparable offerings), and commodity (a utility you rent and stop thinking about).
The single climatic rule that makes it useful: everything drifts rightward. Competition and supply push every component towards commodity, whether or not you participate. Your strategic choices are what to build where nothing exists, what to buy where a market exists, and — the one most teams miss — what to stop building because it has commoditised underneath you.
What the map makes visible that a roadmap does not
The anchor is a need, not a channel. "Get it with least effort" has been the anchor for a century; the components serving it have been replaced repeatedly. Anchoring a map on a channel — our mobile strategy, our AI strategy — encodes today's implementation as if it were the goal, which is precisely the error Sears made with its catalogue.
Position dictates method. The commonest organisational failure is applying one operating model to the whole stack. Genesis components need small teams, short cycles and tolerance for write-offs. Product components need roadmaps and competitive benchmarking. Commodity components need SLAs, procurement and cost discipline. Running a genesis bet through a quarterly stage-gate kills it; running a commodity dependency as an exploratory project wastes money on something you should be renting.
Inertia is visible as a gap. Where a component has evolved but your organisation still treats it as differentiating, you have inertia — and inertia in a map looks like a component sitting left of where the market has moved. Checkout is the current example: several companies are still building payment logic that a rail now provides.
Anomalies are the opportunity. The real-time inventory node is the clearest one on this map. It is depended upon by three separate formats, it is universally bespoke, and it is the named root cause in the failure of in-chat checkout. Components that should have commoditised and have not are where disproportionate value sits.
Five plays, by position
Running it as a practice
The map is worth little as a one-off artefact and a great deal as a cadence. A workable rhythm for a product and marketing leadership team:
Quarterly, ninety minutes. Redraw the map from the current user need. Do not update last quarter's version — redrawing surfaces the assumptions that quietly changed. Mark every component that moved right, and every component you are still building that a market now serves. Then ask the only question that matters at the end of a mapping session: given where things are heading, is our largest current investment aimed at where the value is moving, or at where it used to be?
Section 08
A diagnostic for your own strategy
Eight questions. They are deliberately uncomfortable.
01
Which specific piece of user effort do we remove, and who removes a bigger piece? If the answer names a surface — our app, our AI feature — rather than a unit of effort, you have a channel plan, not a strategy.
02
Where is our category in its life cycle? Winning new demand, or defending existing accounts on price? Check new-logo acquisition and average discount depth, not net revenue retention. NRR is the metric that hides this transition longest.
03
Would our cost base survive winning? Model three times the volume at current unit economics — including inference cost, support load and the fully loaded cost to serve — before scaling a pilot. Several AI features are Webvan-shaped: correct thesis, uneconomic at volume.
04
Is our own data accurate enough, in real time, for what we are promising? Availability, pricing, entitlements, state. This is the node three separate formats have failed on, and it is unglamorous enough that it rarely gets an owner.
05
Are we discoverable by agents without surrendering the transaction to one? Structured data, live pricing, protocol coverage — while checkout, compliance obligations and the customer record stay with us.
06
What share of our demand passes through an interface we do not own? Quantify it. That number is your platform risk, and it is usually higher than the leadership team assumes.
07
Which of our growth mechanisms would we be comfortable explaining to the user it targets? Anything that fails is a liability accruing quietly.
08
What are we automating, and did the user ask for it? Automate the errand. Leave the decision, and make the override obvious and reversible.
The pattern in this map is not that new formats replace old ones. It is that the winning position keeps moving closer to the user's actual life — from a building, to a website, to a pocket, to a feed, to an assistant, to the environment itself. Each move demands a more intimate understanding of the person and grants less control over the context they are in.
Which means the question for a product and marketing team is not whether you have shipped the newest interface. It is whether you understand your user well enough to be chosen in a room you no longer own.
References
Sources and further reading
Cited sources
Reuters, citing Datum Intelligence, on Indian quick-commerce market shares and sector value, January 2026 — reported via Quick Commerce Market Share by Company and StartupFeed, June 2026.
Redseer and Bernstein data on Indian quick-commerce order volumes and dark-store profitability, and the collapse of Getir, Gorillas, Jokr and Flink — Digital in Asia, June 2026.
Global quick-commerce operators and market structure — Persistence Market Research.
OpenAI's withdrawal from Instant Checkout and the shift to retailer apps inside ChatGPT — CNBC, March 2026.
Forrester's analysis of the Instant Checkout pull-back, including inventory, tax and conversion causes — reported via Stellagent, April 2026.
Walmart's measured conversion gap between in-chat checkout and its own site, and the "discover in AI, buy on site" model — Digital Applied, July 2026.
Agentic commerce protocols (UCP, ACP), Stripe's Agentic Commerce Suite, and Rufus user and sales figures — Paz.ai, 2026.
Amazon's crawler blocks, the Perplexity litigation, and Walmart's share of ChatGPT referral traffic — Opascope, April 2026 and Ekamoira, February 2026.
Forrester on zero-click search reaching checkout and the Agentic Commerce Protocol — Forrester Blogs.
Eighth Circuit vacatur of the FTC click-to-cancel rule, July 2025 — WilmerHale, August 2025.
The FTC's renewed negative-option rulemaking, January 2026 — Goodwin, February 2026 and Crowell & Moring, February 2026.
The $2.5 billion Amazon Prime settlement and the dark-patterns enforcement context — The Regulatory Review, October 2025.
Further reading
Amos Tversky and Daniel Kahneman, "Judgment Under Uncertainty: Heuristics and Biases", Science, 1974; and "Prospect Theory", Econometrica, 1979.
Al Ries and Jack Trout, Positioning: The Battle for Your Mind, 1981 (from articles first published in 1972).
Robert Cialdini, Influence: The Psychology of Persuasion, 1984.
Donald Norman, The Design of Everyday Things, 1988.
Jakob Nielsen, "10 Usability Heuristics for User Interface Design", 1994.
Paco Underhill, Why We Buy: The Science of Shopping, 1999.
Richard Thaler and Cass Sunstein, Nudge, 2008.
BJ Fogg, "A Behavior Model for Persuasive Design", 2009.
Harry Brignull, deceptive.design (originally darkpatterns.org), from 2010.
Nir Eyal, Hooked: How to Build Habit-Forming Products, 2014.
