YESWARD

REAL-TIME WEBSITE PERSONALISATION

Real-time personalisation for anonymous visitors, without building a profile.

Yesward personalises from what is happening in the current session — dwell, toggles, comparison loops, the FAQ someone opened — and decides, per visitor, whether personalising is worth doing at all.

DEFINITION

Real-time website personalisation adapts what a visitor sees during the visit, using signals from that visit rather than a stored profile. It works on first-time and anonymous visitors, because it does not need to know who someone is — only what they appear to be struggling with.

Sometimes called intent-based, behavioural or session-based personalisation. The useful distinction is not the label; it is whether the system is allowed to decide against personalising.

TWO MODELS

Personalisation used to start with who you are. Real-time personalisation starts with what you are doing.

SEGMENT-BASED · THE OLDER MODEL

Identify the segment, then choose the experience.

INPUTStored attributes, CRM traits, past sessions, audience membership
NEEDSIdentity, or a durable identifier that survives the visit
FIRST VISITFalls back to a default; the majority of considered-purchase traffic
FAILS ATSomeone whose segment is right but whose current problem is not the one the rule assumes

SESSION-BASED · WHAT YESWARD DOES

Observe, infer the friction, decide whether to act, then act minimally.

INPUTIn-session behaviour: dwell, reversals, toggles, comparison loops, FAQ topics, stalls
NEEDSNothing about the person. No login, no history, no third-party data
FIRST VISITFully supported — it is the normal case, not the fallback
REFUSESTo act when the evidence is weak — the hypothesis stays a hypothesis and nothing renders

SIGNALS → FRICTION

Behaviour is evidence. It is not mind reading.

Everyone else promises to know what your visitor wants. We produce a scored hypothesis about what is in their way, and we say how confident it is.

Choice friction

Too many viable options, no basis to separate them.

SEEN AS: A↔B↔C LOOPS · REPEAT COMPARE VISITS · SCROLL REVERSALS

Price friction

Unclear total cost, not necessarily an unwillingness to pay.

SEEN AS: PRICING DWELL · FEE FAQ OPENED · CALCULATOR RE-RUNS

Commitment friction

The term, the lock-in, the cancellation conditions.

SEEN AS: TERM TOGGLED 4× · CANCELLATION FAQ · CHECKOUT NOT STARTED

Trust friction

Doubt about the company, the promise or the process.

SEEN AS: TERMS PAGE · REVIEWS DETOUR · OFF-SITE RETURN

Coverage friction

Eligibility, availability, dates, address, compatibility.

SEEN AS: REPEAT CHECKS · CHANGED INPUTS · ABANDON AT VALIDATION

Thirteen named frictions sit under these five families, each with the signals that suggest it, the level of response it justifies, and the copy that is forbidden when addressing it. That vocabulary is published, not proprietary marketing language.

The friction vocabulary →

ANONYMOUS VISITORS

Most of your considered-purchase traffic will never tell you who they are.

They arrive from a paid click, compare three options, read two FAQs and leave. No account, no email, no prior session. Profile-based personalisation has nothing to work with; session-based personalisation has everything it needs.

In Europe this is also the commercially safer position. Nothing here depends on third-party cookies, cross-site identifiers or data enrichment — so consent changes reduce what we may act on, not whether the product works.

WHAT A DECISION NEVER CONTAINS

no name or emailno raw IPno cross-site identifierno keystrokesno session replayno free-text captureno profile after 24hno inference stored as fact

And one rule that is a copy rule, not a data rule: behaviour is never referenced back at the visitor. The engine acts on what it sees; it never tells someone it was watching.

“I noticed you've looked at pricing three times…”

FORBIDDEN AT RENDER TIME · NOT A GUIDELINE

THE RESTRAINT ARGUMENT

Should you personalise every visitor? No.

Personalisation coverage became a vanity metric. “96% of sessions personalised” describes effort, not outcome — and on considered purchases, adapting a page for someone who was progressing fine is at best neutral and at worst an interruption they resent.

The honest version of coverage is a distribution, drawn to scale: how many sessions were left alone, how many were held out, and how few needed anything at all.

Read the argument in full →

WHERE 12,480 SESSIONS WENT

Left alone — nothing rendered88.9%
Held out on purpose — the counterfactual2.2%
Passive adaptation — page gains information2.5%
One contextual line, dismissible forever2.1%
Anything more involved than that4.3%

ILLUSTRATIVE DETERMINISTIC DATA · DRAWN TO SCALE

INCREMENTALITY

How do you measure the incremental value of personalisation? Withhold it from someone, permanently.

01 · HOLDOUT

Never zero

A stable-hash share of eligible sessions is withheld for as long as the policy runs. 5% floor, enforced server-side; 20% is the pilot standard.

02 · ELIGIBLE ONLY

Compare like with like

Randomisation happens after the eligibility gates, so the holdout contains sessions that would have been treated — not all traffic.

03 · MARGIN

Net, with an interval

Discounts, model spend and platform fees are netted out, and the 95% interval is printed next to the point estimate.

04 · RETIREMENT

Losers go back to silence

Fourteen days below holdout on margin and the treatment retires itself. Restraint is enforced by the rule, not by good intentions.

The measurement maths → · The arm register and CI bands →

FREQUENTLY ASKED

Real-time personalisation, briefly.

What is real-time website personalisation?

Adapting what a visitor sees during the visit, based on signals from that visit. It does not require identity, history or a logged-in account.

Can you personalise for anonymous visitors?

Yes — and on considered purchases they are the majority. Behaviour in the current session is enough to form a friction hypothesis without knowing anything about the person.

Is this cookieless and GDPR-friendly?

Decisions run on first-party in-session events with no cross-site identifier and no profile persisted beyond 24 hours. Consent state is an input to every decision, so a refusal narrows what the engine may do rather than breaking the page. Your DPO should still read our DPA.

How is this different from intent-based personalisation?

Intent prediction asks how likely someone is to convert. We ask what is in their way, which is a different and more actionable question — you can resolve a blocker, but you cannot resolve a probability.

Will visitors know something adapted?

At level 1 the page simply contains the information they needed. From level 2 upward every surface carries an AI mark, a single dismissal that holds for the session, and the promise in writing.

Does it work alongside my personalisation platform?

Yes, and that is the common case. Merchandising and audience relevance stay where they are; Yesward governs the moment a specific buyer stalls. See the comparisons.

See which sessions your current stack should have left alone.

Two weeks in shadow mode, on one journey. Nothing renders to a single visitor.

Review one leaking journey