Luxury service has always been organised around a person. A named associate who knows a client's size, remembers the anniversary, understands which colourway was returned and why, and can judge from a single sentence whether a request is routine or delicate. Clienteling โ the practice of assigning named associates to named clients and recording preferences, purchases and occasions โ was built on the assumption that this knowledge is the house's most valuable inventory. It is not held in a warehouse. It is held in a relationship.
That relationship is now being asked to share space with software. Tools positioned as concierge assistants are appearing across the client journey: scheduling appointments, answering routine questions about care and availability, drafting follow-up messages, triaging repair requests and surfacing product suggestions from a client's recorded history. The capabilities are real, and the commercial logic behind them is not difficult to follow. Associates are expensive, clients expect immediacy, and much of what a boutique handles in a day is genuinely repetitive.
The strategic question is not whether the technology works. It is whether a house can adopt it without surrendering the two things clients are actually paying for: discretion and judgement. Those are harder to specify, harder to measure and much harder to restore once damaged.
Where assistance belongs, and where it does not
The first design decision is a map of the client journey, divided not by what a system can do but by what happens when it gets something wrong.
Low-consequence, reversible tasks sit comfortably on the assisted side. Confirming an appointment time. Explaining how a material should be stored. Checking whether a repair has been logged. Offering a client three available slots rather than one. Sending a reminder before a collection. These are administrative acts, and an imperfect answer costs a small correction.
High-stakes moments belong to people. A bespoke commission, where the conversation is the product. A complaint about a piece that has failed. A discussion of provenance, authenticity or valuation. Anything touching a client's personal circumstances โ a bereavement, a separation, a gift intended to be anonymous. Anything with legal or privacy implications. A negotiation on price or allocation.
The distinction matters because the failure modes are asymmetric. A clumsily worded appointment reminder is an irritation. A clumsily worded message to a client discussing a sensitive family matter is a breach of the discretion that justified the price in the first place. Houses that blur the line in pursuit of efficiency tend to discover the difference only after the fact.
Data governance comes before deployment
Client profiles are unusually sensitive records. They combine identity, spending capacity, taste, gifting habits, travel patterns and, often, information about the people a client buys for. A system that reads those records inherits the obligation to protect them.
Four questions should be settled before any assistant is connected to a client database. What is the lawful basis for holding each field, and has consent been captured in a way that would survive scrutiny under regimes such as the European Union's General Data Protection Regulation? How long does a note about a preference or a personal detail persist, and who decides when it is deleted? Who inside the organisation can see a full profile, and is that access role-based rather than universal? And what happens to the data once it leaves the house's own systems โ what do third-party suppliers retain, and on what terms?
A preference note written by an associate for a colleague's benefit is a very different object from a preference note ingested by a vendor's model. Treating them as the same thing is where most governance failures begin. The safest posture is to assume that anything a client tells a boutique could one day be read by someone the client never met, and to design accordingly.
Escalation is the product
Much of the commentary on assisted service focuses on how well a system answers. The more consequential design choice is how quickly and gracefully it stops answering.
Escalation triggers should be explicit rather than inferred. A client expressing frustration. A client whose recorded history places them above a defined value threshold. A topic that is bespoke, legal, medical, financial or privacy-sensitive. A direct request for a person, which should be honoured without friction. A question the system cannot answer with confidence, which is not the same as a question it can answer plausibly.
The handover itself deserves as much attention as the trigger. A client who is passed to an associate should not feel they have failed a test or been downgraded to a queue. The associate, in turn, should receive context rather than a transcript: what the client wants, what has already been said, what has been promised. A handover that forces the client to repeat themselves has undone the benefit of the automation entirely.
Voice is edited, not generated
Luxury houses spend considerable effort on how they speak: the register of a letter, the restraint of a caption, the precise distance between warmth and familiarity. That voice is an asset, and it is easily flattened.
Two failures recur. The first is generic phrasing โ language that could have come from any retailer and quietly signals that no one at the house was paying attention. The second is over-familiarity, where a system mistakes enthusiasm for service and addresses a long-standing client with the tone of a marketing email.
The practical answer is editorial rather than technical. Approved phrasing for common situations. A list of claims that may never be made โ about provenance, delivery dates, repair turnaround or availability. Human review for any outbound communication above a defined sensitivity or value threshold. A named editor inside the house who owns the voice and can change it, in the same way a creative director owns a collection.
Measure the relationship, not the automation
Response speed and automation rate are the easiest metrics to report and the least informative. A house can drive both up while quietly eroding the thing it was trying to protect.
More useful measures sit closer to the relationship: client retention and repeat purchase, resolution at first contact, the proportion of escalations that were handled without the client noticing a seam, complaint rates and their severity, and whether associates trust the system enough to use it. Time saved is worth counting only if it is redeployed โ into selling, into client contact, into the work that justified the headcount.
None of this is an argument against assistance. It is an argument for treating it as an appointment rather than a feature. The houses likely to manage the transition well will describe their systems the way they describe a junior colleague: a narrow remit, a clear script, and a standing instruction to fetch a senior the moment something matters. Discretion cannot be added at a later release. It has to be designed in from the first conversation, which is precisely the standard the Global Luxury Awards exist to recognise.
