Curaes · your AI premium curated customer experience

Your AI should know your client. It shouldn't show it.

The problem

AI chatbots and recommenders reveal every guess they make and push product after product. For a premium brand, that feels like being watched, and clients pull away.

Try it yourself See the matrix
Curaes in action
MAISON VALE · CHAT12:41
Hello, anything new in silk?
AI assistant · draft, not sent
Hi Sarah!! 👋 We noticed you viewed the silk scarf 7 times 👀 Congrats on the baby on the way! 🍼 Only 2 left, 20% OFF!
AI assistant paused · Mei alerted
Mei, your relationship manager, has joined
Good afternoon, Ms Tan. It's Mei. We do have a few lovely silk pieces in. Shall I keep an hour for you this week?
The matrix, live
Pushy
Creepy
Generic
Concierge ✓
knows more →↑ shows more

Waiting for a message…

Why you should care

Creepy personalisation costs you the client.

When AI reveals what it knows, clients feel watched. Clicks fall, trust drops, and a premium brand starts to feel mass-market.

See the research
Clicks fall

Personalised ads based on covert tracking feel invasive and get fewer clicks.

Aguirre et al. 2015, Journal of Retailing
Pushback

Over-personal messages make people resist, even when the offer is good.

White et al. 2008, Marketing Letters
Value drops

Premium brands sell distance and service. Hard pushes make them feel mass-market.

Kapferer & Bastien 2009, The Luxury Strategy
Why now

AI chatbots already talk to your clients. No one checks them first.

CRM tools now write and send personal messages on their own. Clients expect personalisation, and regulators are watching how it's done.

Why this is urgent
Already live

Generative AI in CRM and chat sends personalised messages at scale, with no human review.

Expected

71% of consumers expect personalisation; 76% are frustrated without it.

McKinsey & Company 2021
Regulated

The EU AI Act bans manipulative AI. Singapore's PDPA limits data to what clients agreed to.

EU AI Act Art. 5 · PDPA 2012
The solution · the Curaes matrix

Know a lot. Show a little.

A simple guardrail framework for any AI that talks to customers. Every message lands in one of four boxes. Curaes keeps yours in the green one, and calls a human when it can't.

How much it shows ↑
Shows a lot, knows littlePushy
Shows a lot, knows a lotCreepy
Shows little, knows littleGeneric
Shows little, knows a lotConcierge ✓
How much it knows →

Four things we score

1
ShowDoes the message reveal how we know? "You viewed this 7 times."
2
PersonalHow private is the topic? A past purchase is fine. Health, family or money is not.
3
PushHow many product pushes this client got this week.
4
SourceDid the client give us this data, or did we buy or guess it?
The scoring rule
Show level × Personal level = Creepiness
Every push over the limit doubles it. As creepiness rises, trust falls twice as fast.
Green · 1–2AI sends
Amber · 3–6AI softens, then sends
Red · 7+AI stops. A human takes over.
How the scoring works
  • Show and Personal are each rated Low (1), Medium (2) or High (3).
  • The weekly push limit starts at 2 product recommendations per client.
  • Data the client never gave us, or a private life event, is an automatic red.
  • On red, the AI chatbot and product pushes pause, and the relationship manager or sales associate is alerted on their phone.
  • The "trust falls twice as fast" link is the core hypothesis the research will test with real clients.
Try it

Client on the left. Manager on the right.

Tick what the AI knows and how often it has pushed products. Watch the chatbot stop itself and call a human.

What the AI knows about Ms Tan

Client view · brand chat
Manager view · Mei's phone
Next step

Pilot Curaes on one channel.

We run alongside your current AI for a month and show you every message it would have stopped.

Talk to us on LinkedIn View on GitHub
The researcher

Jasmine Cheong

AI hyper-personalisation safety research: helping premium brands personalise with AI without crossing into creepy. Curaes is an ongoing project.

Connect on LinkedIn
Sources
  1. Aguirre, E. et al. (2015). Unraveling the personalization paradox. Journal of Retailing, 91(1).
  2. White, T. B. et al. (2008). Getting too personal. Marketing Letters, 19(1).
  3. Kapferer, J.-N. & Bastien, V. (2009). The Luxury Strategy. Kogan Page.
  4. McKinsey & Company (2021). The value of getting personalization right—or wrong—is multiplying.
  5. EU AI Act (2024), Art. 5 · Personal Data Protection Act 2012 (Singapore).