There is an extraordinary amount of intelligence hidden in the ordinary behaviour of an ecommerce customer. The problem is not that retailers cannot see it. They have been collecting it for years. The problem is that most of the technology they use has remarkably little idea what it means.

Somewhere on your website today, a customer will look at a product, leave it, consider something else and return. Perhaps they first noticed it yesterday and have since examined three alternatives. Perhaps they will check the price again, enlarge a photograph they previously ignored, read the specification for a second time and disappear without buying anything. To most ecommerce systems, this is an unremarkable collection of events: product views, sessions, clicks and elapsed time. To anybody who has ever watched a real person wrestle with a purchase, it may be something rather more interesting. The customer could be making up their mind. The signals are already there; the difficulty is whether the retailer is listening closely enough to understand what they mean.

That distinction matters because ecommerce has become exceptionally good at recording behaviour without necessarily becoming much better at understanding it. We know what somebody viewed, what they searched for, which email they opened, what they bought six months ago and whether they abandoned a basket last Tuesday. Yet much of the machinery built around that information still waits for an explicit event before deciding that something commercially significant has happened. The customer must search, click, add, abandon, purchase or otherwise perform a sufficiently convenient action to trigger a response.

Human beings are considerably less obliging. Desire does not arrive as an event in an analytics platform. It develops.

The purchase begins long before the basket

Consider two customers who each view the same £180 pair of shoes three times. On a conventional dashboard their behaviour may appear virtually identical, yet one may simply be wandering through a dozen pairs while the other first saw those particular shoes yesterday, returned to them after dinner, considered an alternative this morning and came back to the originals that afternoon. Three views apiece, but potentially two completely different states of mind. The interesting question is therefore not how many times somebody viewed a product, but whether their behaviour suggests that the product is becoming more important to them.

This is particularly obvious when we think about purchases that involve pleasure, taste or discernment rather than necessity. A bottle of wine for an important dinner, a jacket one does not strictly need, a new bicycle, a watch, a piece of furniture or an unnecessarily expensive box of chocolates rarely begins with a perfectly articulated intention. Something catches the eye. An alternative is considered. The price suddenly seems excessive. The original item is revisited. Owning it is imagined. A perfectly sensible reason for not buying it is found, followed some hours later by an equally persuasive reason why buying it would be entirely reasonable.

This apparent indecision is not noise surrounding the purchase. Very often, it is the purchase taking shape.

Retail technology has traditionally been more comfortable with what happens afterwards because explicit intent is easier to classify. A basket addition is wonderfully unambiguous. An abandoned checkout provides a convenient reason to send something. A completed order is better still. Yet by the time any of these things happens, much of the interesting work inside the customer’s head has already taken place. Alternatives have been weighed, preference has strengthened, doubts have appeared and perhaps disappeared, and one product has begun to acquire significance over another.

The commercial opportunity lies in recognising that movement before the customer has formally declared it.

Your catalogue is different for every person who enters it

Retailers inevitably think of themselves as having a catalogue. It might contain 5,000 products or 50,000, organised into categories and ranked according to a mixture of popularity, margin, seasonality, availability and merchandising judgement. There will be bestsellers, new arrivals, products that sell themselves and others stubbornly occupying warehouse space. All perfectly sensible from the retailer’s point of view, but almost entirely irrelevant to the way an individual customer experiences those products.

A bestseller can be utterly uninteresting to one person while an obscure item buried several pages into a category may be precisely what another has been hoping to find. Its lack of popularity does not diminish its relevance to that individual. Indeed, viewed from the customer’s perspective, the retailer does not really possess one catalogue at all. It possesses thousands of possible versions of the same catalogue, because every customer brings their own tastes, circumstances, previous purchases, aspirations, prejudices and peculiarities to it.

This is where conventional personalisation begins to show its age. Its natural instinct is to find similarities between people. Customers who bought this also bought that; people interested in this category tend to prefer those products; customers displaying certain characteristics are placed into a particular audience. The groups can become extraordinarily sophisticated, but the underlying compromise remains the same: the individual is being understood by reference to other people.

That was perfectly reasonable when there was no practical alternative. A knowledgeable shopkeeper might remember that Mrs Jones preferred a particular cut, that Mr Smith would happily spend money on wine but hated ostentation, or that another regular customer needed considerable reassurance before indulging himself. An ecommerce retailer with 100,000 customers could hardly employ 100,000 attentive shopkeepers, so digital marketing learned to approximate. The approximation became so familiar that the industry eventually gave it an impressive vocabulary and stopped thinking of it as a compromise.

The difficulty is that the customer has never agreed to remain conveniently similar to the people with whom a system has associated them. Nor, for that matter, have they agreed to remain similar to themselves.

Preferences change. Circumstances change. A purchase alters what becomes relevant next. Curiosity develops where none existed before, while an enthusiasm that seemed dependable quietly disappears. Something ignored repeatedly for six months can suddenly become interesting because another part of the person’s life, behaviour or taste has changed. If the individual is constantly moving, relevance cannot sensibly be calculated once and treated as a permanent characteristic. The interpretation has to move with them.

Hesitation is not the same as rejection

There is another reason this matters. Ecommerce has acquired an understandable impatience with hesitation. We spend considerable sums attracting people to websites and naturally prefer them to convert, preferably without wandering off to think about it. When they do hesitate, the industry’s instinct is often to regard the delay as a problem requiring intervention. Hence the extraordinary enthusiasm for urgency, reminders, incentives and discounts.

But people often hesitate precisely because something matters to them. Nobody spends three evenings agonising over which washing-up sponge to buy. A discretionary purchase that involves taste, aspiration, indulgence or a meaningful amount of money is different. The customer may return repeatedly because desire is strengthening while justification struggles to keep pace. What looks like indecision in aggregate can be perfectly coherent when interpreted at the level of one person.

This is commercially important because the wrong interpretation can be expensive. If a customer already has a strong preference for a product, immediately offering 15% off may not create a sale at all; it may simply reduce the margin on a sale that was going to happen anyway. Showing a collection of alternatives at precisely the moment somebody is approaching certainty may be equally unhelpful, introducing doubt where none was required. Conversely, waiting until enthusiasm has faded and then delivering the supposedly perfect recommendation is little better. Relevance is not simply a question of selecting the correct product. Timing is part of relevance.

There is therefore a valuable interval between casual interest and declared purchase intent. Too early, and the retailer is guessing. Too late, and the customer has already made the decision, bought elsewhere or lost interest. Between the two sits the moment at which accumulated behaviour begins to suggest that one product, or one type of product, is acquiring unusual significance for one particular person. Recognising that moment is considerably more useful than merely recognising that a basket has subsequently been abandoned.

From personalisation to individual judgement

This is the distinction at the heart of autonomous individualisation. The purpose is not to make a communication appear more personal, nor to construct ever smaller groups until a segment feels sufficiently precise. It is to interpret each consumer independently and continuously, calculating what that individual is most likely to buy next as their behaviour and preferences evolve.

SwiftERM does this autonomously because the scale of the problem makes manual intervention absurd. Every customer is changing, every product competes for relevance differently for every individual, and every new piece of behaviour can alter the relationship between the two. The calculation therefore cannot sensibly be a periodic marketing exercise. It has to be continuous.

The consequence is more profound than better product recommendations. If a retailer becomes substantially better at recognising genuine individual preference, it needs less of the marketing traditionally used to compensate for uncertainty. Less indiscriminate frequency, less relentless promotion and, importantly, less dependence upon discounting as the universal answer to weak conversion. Put something somebody genuinely wants in front of them when that desire has become commercially meaningful and the product itself can do rather more of the selling.

There are benefits elsewhere too. Products outside the bestseller list have a greater opportunity to find the particular customers for whom they are highly relevant, improving inventory liquidity rather than continually concentrating demand on what already sells easily. Existing traffic becomes more productive. Customer acquisition expenditure has the opportunity to produce greater lifetime value. Margin is better protected because relevance, rather than price reduction, is doing more of the work.

For years the ecommerce industry has asked how it can persuade customers to buy more. Perhaps that is increasingly the wrong question. A more interesting one is what is this particular customer already becoming likely to want?

Listening to what they have been telling you all along

None of the behavioural signals involved is new. Customers have always returned, compared, hesitated, reconsidered, rejected alternatives and gradually developed preferences. Retailers have been recording much of this behaviour for years. What has changed is the ability to interpret it autonomously at the level at which buying decisions actually occur: the individual.

That is why the next significant advance in ecommerce is unlikely to come from collecting yet more customer data. Most established retailers already possess far more than they use intelligently. The greater opportunity is listening to what the existing behaviour is telling you, understanding how its significance changes over time and what it reveals about the emerging preferences of one person rather than the statistical habits of a convenient group.

Somewhere on your site today, somebody is looking at something for the third or fourth time. They have not put it into their basket. They have not asked for help, begun checkout or triggered the recovery campaign somebody carefully configured months ago. They may not even have admitted to themselves that they particularly want it.

But their behaviour is already telling you something.

The question is whether you’re listening.




Accreditations / References

Murray, K. B. & Häubl, G. — Journal of Interactive Marketing
Personalization without Interrogation: Towards more Effective Interactions between Consumers and Feature-Based Recommendation Agents. Research into individual-level consumer preference models and personalised product recommendation. 

Journal of Retailing and Consumer Services — Elsevier
To hesitate or not to hesitate: Can popularity cues minimize the hesitation to checkout in e-commerce? Research examining purchase intention, information search and consumer hesitation in ecommerce. 

Baymard Institute — Ecommerce Checkout Research
Long-running quantitative and qualitative research into browsing, purchase readiness, cart abandonment and ecommerce consumer behaviour. 

Jansen, L. Z. H., Bennin, K. E., van Kleef, E. & Van Loo, E. J. — Computers in Human Behavior
Online grocery shopping recommender systems: Common approaches and practices. A systematic review examining how consumer data are used to infer preferences and needs in online recommendation systems. 

Alptekinoğlu, A. — Production and Operations Management
Flexible Products for Dynamic Preferences. Research addressing the fact that individual consumer needs and preferences change dynamically over time. 

Those five collectively underpin hesitation, developing intent, individual preference, recommendation systems and changing preference over time — essentially the intellectual foundations of the article without pretending that any one source supplied the SwiftERM argument.