A smarter question: what does hyper-personalisation mean in B2B?

Consumer personalisation works on behaviour; something we’re all used to already. Spotify infers your mood from what you played last Tuesday. Netflix surfaces something you didn’t know you wanted to watch. Airbnb suggests destinations you might want to visit on your next holiday. The trick is doing this without asking and without simply employing a recomendation engine.

B2B purchasing doesn’t work that way, and more importantly, shouldn’t. High-consideration decisions (commercial real estate, enterprise software, professional services) are infrequent, high-stakes, and contextual in ways behavioural patterns can’t capture alone. The system needs to ask. The question is which questions, for which buyer, and what happens with the answers beyond matching on stated criteria.

Hyper-personalisation in B2B software means systems that learn which questions are predictive for different buyer types, route qualification accordingly, and cross-reference answers with what similar buyers have actually needed. I’m talking about a totally different capability from personalisation than the “hi there <first name>” email: a system that learns, rather than one that just benefits from configuration.

Most B2B personalisation is configuration

The standard approach gives buyers more ways to specify what they want. More filter fields, richer facets, better search. The platform matches (directly, or fuzzily) on stated criteria. Resulting in a burden on the buyer: know what you need, describe it accurately and you’ll be given some options.

This works for buyers who know exactly what they’re looking for. The problem is that “knowing exactly what you need” is a significant assumption in any complex purchase. The requirements that determine fit (the ones that separate a good outcome from a regretted one) often surface through conversation, not through a form. Gartner’s B2B buying journey research describes why: the typical buying group involves six to ten decision makers, each arriving with information they’ve gathered independently, and the journey loops back through earlier stages rather than moving in a straight line. Requirements shift as the evaluation unfolds.

In practice, the platform doesn’t learn or adapt. Every buyer starts from the same blank state. It gets better at giving people more ways to describe themselves, which is useful, but it’s not what most people really mean by personalisation. It’s self-service configuration, which is a different thing.

A platform that only matches on stated criteria isn’t personalising. It’s making the buyer do the work and calling it a feature.

Two mechanisms need to work together

The first is guided asking. Different buyer types need different qualification paths. Two firms with near-identical initial requirements may need completely different things. A 20-person legal team and a 20-person product studio file the same brief; what they need is not the same. A platform with enough pattern data from real buying decisions learns which follow-up questions reveal the difference: which inputs surface the growth trajectory that changes the recommended term, which details bring out the requirements the buyer hadn’t thought to mention. Qualification becomes shorter and more relevant — not because it asks less, but because it asks better.

The second is pattern inference. Structured answers to the right questions can be cross-referenced with what buyers in comparable situations have needed. Not surface correlation. Substantive inference from buying outcomes: buyers with this profile, this situation, this requirement, have consistently needed X that they didn’t specify upfront. That surfaces before they have to discover it themselves.

Neither of these approaches will work alone. Guided asking without pattern data is just a shorter form, and pattern inference without structured inputs is just noise. But together, they produce something that can understand the buyer’s situation well enough to surface what fits, including what wasn’t stated.

Where AI comes in

AI’s contribution is pattern recognition at a volume no individual can ever match. Given structured briefs and buying outcomes, it can identify which questions are predictive for which buyer types, and which non-stated requirements consistently follow from a given profile. It doesn’t replace the structured data collection; it derives the signal from it.

It’s worth being precise here, because the use of “AI-powered personalisation” does a lot of unexamined work in B2B marketing copy at the moment. A recent benchmark of frontier models, Know Me, Respond to Me, found they recall user facts reasonably well (60–70% accuracy) but struggle to incorporate a user’s latest situation into a response (30–50%). Left to infer from loose conversational context, the model guesses. Given structured inputs and real outcome data, it has something to reason over. The quality of the inference is bounded by the quality of the demand-side data.

There’s also a reason to prefer outcome-grounded inference over the “learn the user’s profile” approach consumer AI is converging on. Researchers at MIT and Penn State found that condensed user profiles in a model’s memory made LLMs measurably more agreeable, more likely to mirror the user’s view than challenge it. In a buying context, that’s the last thing you want: a system that agrees with the brief will never surface the requirement you didn’t state. Inference anchored to what similar buyers went on to need has the opposite bias. It’s grounded in outcomes, not in pleasing you.

What this looks like in practice

Flex workspace makes the gap between stated requirements and actual needs easy to see.

Consider a typical brief: 15 desks, Zone 1, flexible term. It’s close to useless as a matching input beyond basic filtering. It specifies quantity, geography, and tenure. It says nothing about how the team works day to day, whether 15 is a ceiling or a planning number, whether the firm has confidentiality requirements, or whether the environment needs to be client-facing. Those things determine fit in ways desk count and location simply don’t.

Two identical briefs can describe two firms that need spaces that are nothing alike. The brief isn’t the requirement; it’s the opening line of a conversation.

A good broker knows which questions to ask (and, perhaps more importantly, in what order). They’ve had this conversation many times with clients in similar positions; it’s most of what flex workspace brokerage actually involves. They know that a 15-person legal practice and a 15-person product studio may file identical briefs and need spaces that are nothing alike. That knowledge (which questions reveal which information for which client type) normally lives in the broker’s head and travels through relationship. It doesn’t transfer to the platform unless the platform is designed to capture it structurally.

A platform that processes enough structured briefs starts to see the same patterns. It learns which inputs are predictive. It can route qualification differently for a legal team than for a tech team, even when the initial brief looks the same. The brief structure that operators actually need turns out to be the same structure that teaches the platform what it needs to know.

What this means for B2B software

Every high-consideration B2B purchase has a version of this structure. The gap between stated requirements and what buyers actually need gets closed by relationship in most markets: a salesperson, account manager, or consultant who asks the right questions because experience tells them which questions matter (and what the silences between answers mean).

The unfortunate awkward truth is that buyers increasingly prefer not to have that conversation at all. Gartner’s March 2026 sales survey found that 67% of B2B buyers would rather buy without speaking to a rep, up from 61% a year earlier. The appetite for what the relationship produces is undiminished… but the appetite for the process that produces it is falling. Which means software has to carry more of the asking. Assuming of course that software can still elicit a response from potential buyers at all!

Buyers want what the relationship produces without the process that produces it. In my opinion, that gap is where the next phase of B2B software will be built.

If you discuss this topic with any seasoned sales person, there’s a reasonable objection: relationship building isn’t just data collection. The trust, the back-and-forth, the nuance of a real conversation. None of that is, or can be, captured by a form, however well designed. And I’d agree with them on that. What I’m describing isn’t replacing a relationship. It’s what happens before, during and after it: pattern knowledge that previously had to live in one person’s head, made available earlier and throughout in the process (where relevant); improving outcomes for every buyer.

B2B software that gets this right will look less like a search engine with better filters and more like an advisor who has seen your situation before. The questions get smarter. The inference builds on real outcomes. The gap between what you describe and what you need narrows, and narrows further with every transaction the platform processes. I’d hasten to add that this doesn’t just mean a chatbot either! These transactions are not just a form or a chat window - they are multi-channel, based on research and gleaned from multiple sources. A technique which I think Boardy employs really well.

The precondition is demand-side data — not what’s available, but what buyers in specific contexts have needed, structured well enough to derive signal from. Most B2B platforms have spent years building the supply side and have almost nothing on the demand side. And I’m betting that’s where the next phase of personalisation is going to be built.

What we’re building at Great Space

Great Space is built on this hypothesis (I’ve written before about what a workspace deal platform is). The brief structure brokers use isn’t a free-text box and it isn’t a endless number of checkboxes; it captures the inputs that determine fit, including the requirements clients don’t always know to state upfront. As briefs accumulate on the platform, the pattern data builds: which inputs predict which outcomes, which requirements consistently surface for clients in a given position. We’re early on that curve, and I won’t pretend otherwise. But we’ve put in place the structure for the learning to compound.

Matching on Great Space improves over time not because the supply side grows, but because the demand-side understanding deepens. If you’re a CRE broker looking to move more flex and managed workspace alongside your existing deal flow, here’s what that looks like from the broker side — and Great Space is free to try for 30 days.

---

What are you seeing in how B2B buyers describe what they need, and how close that turns out to be to what they actually choose? I’d be curious what other practitioners are observing.