You can target down to the street, the device, the daypart, and the “most likely to convert” micro-moment, then still lose because you texted someone who hates texts.
That is the basic problem with modern personalisation. It can know an impressive amount about what someone might do next and still ignore how they asked to be contacted. It optimises for precision while leaving permission to sit in the plumbing.
The preference-last trap
Preference-last is what happens when a team treats contact preference as admin and behavioural data as strategy.
It looks sensible because the dashboards reward what is measurable. “Did we contact them the way they asked?” tends to live in a different system, owned by someone else, and defended by nobody.
Why preference breaks personalisation
Preference is a gate, not a segment. If you push someone through the wrong gate, the rest of the journey does not get a chance to be “clever”, because it never gets processed.
The B2B Institute’s 95-5 framing is a useful reminder here: most buyers are not in-market at any one time, which is exactly when small irritations become durable memories.
Research on personalisation backfire suggests that when privacy concern is made salient, more intrusive personalisation can become less effective rather than more persuasive.
Channel mismatch can create a related form of reactance. What the brand calls relevance can feel, to the recipient, like the brand taking liberties. The grim bit is that getting preference right is usually invisible, while getting it wrong is memorable.
Simple brands raise the bar they will be judged by
A simple brand experience is easier to process. Processing fluency research helps explain why that ease can shape trust, credibility, and judgement.
Simplicity raises the standard. Once a brand looks competent, people expect the basics to work. Satisfaction research has long treated expectations and perceived performance as linked; the useful point here is simpler: once a brand has trained people to expect competence, small failures are judged against that expectation.
So yes, the slicker your comms stack gets, the more insulting it feels when you ignore the simple human request of “don’t call me”.
A concrete example: the “perfect” journey that creates a complaint
A mid-sized insurer builds a renewal journey that pulls policy type, renewal date, prior claim status, and predicted churn risk into beautifully timed comms. The team uses email for most customers, SMS for “high risk”, and a call for anyone who clicks but does not complete.
On paper, it looks like grown-up lifecycle marketing. In reality, preference data lives in the billing system, the CRM stores consent, the contact centre has its own flags, and the marketing automation tool is guessing.
A customer who selected “email only” gets an SMS at 07:12 because the churn model likes early mornings. The SMS triggers a complaint, the customer opts out completely, and the brand learns a valuable lesson about how “incremental uplift” works when the uplift is negative.
That is where personalisation turns into a marketing ops problem.
Why this keeps happening
Preference data is operationally awkward. It is multi-channel and time-sensitive, and direct marketing rules can differ by channel, consent status, and recipient type.
Behavioural and transactional data is operationally glamorous. It arrives in neat tables, it feeds models, it moves numbers in a sprint, and it makes everyone feel like the business is “data-led”.
Preference also lacks a natural owner. Marketing wants the upside, compliance wants the risk reduced, support wants fewer angry calls, and product would rather ship something that looks like progress.
So preference becomes a form field, not a system.
Decisions that protect preference
Make preference a first-class product setting, even if it means fewer “personalised” journeys launched per quarter.
Make the source of truth for preference singular, even if it exposes messy legacy realities.
Treat “unknown preference” as a risk state. Default to the least intrusive channel you can justify, even if it reduces short-term response rates.
Re-confirm preferences periodically, especially when the message type, channel, or relationship has changed.
Bloom & Wild’s special-occasion opt-outs are a useful consumer example of this trade-off: fewer assumed occasions, more explicit control.
Keep channel experiments inside the boundaries people agreed to, even if optimisation feels slower and less exciting.
The parts personalisation cannot rescue
Preference has a shelf life. People change roles, move house, take on different responsibilities, become harder to reach, or stop wanting a particular channel used for a particular kind of message. A preference captured once can quietly become a bad instruction.
Preference is also plural in B2B. The buyer, user, approver, and finance contact do not share a brain or an inbox. One contact record can hide several different communication needs, especially when the message moves from sales, to onboarding, to billing, to support.
Personalisation cannot rescue that. It can make the wrong contact, wrong channel, or wrong assumption arrive with more confidence.
Preference has to arrive first
Personalisation is frequently a way to feel sophisticated while dodging the harder work of building systems that respect people.