Nearly New Cashmere · Customer service
AI-led customer service that knows its best customers.
Every customer email now arrives with a reply already drafted from that customer's full history. In a blind test on 100 real emails, the drafts were as good as or better than the team's own reply 93% of the time.
93%
of drafts as good as or better than the team's own reply, tested blind on 100 real customer emails
77,000+
customers, each with one record built from five systems and reconciled to Shopify within 0.2%
Every hour
new emails drafted from 8am to 6pm, every day, for the team to check and send

The business
Nearly New Cashmere restores and recycles discarded cashmere and sells it online, alongside its own recycled cashmere range. It has grown from fairs into a family business with an Instagram community of more than 200,000, and most of its sales now happen online.
Its customers come back again and again, so every reply is part of a long relationship, not a one-off transaction.
The challenge
Answering one email well meant opening five systems: Shopify for orders, Klaviyo for marketing history, Smile.io for loyalty, Loop for returns and Loox for reviews. Then the person replying had to know the answer: the returns policy, which items are one size, where to send cashmere for credit.
Much of that knowledge lived in people's heads. As the inbox grew, there was rarely time to piece it all together before replying.
What we built
One record for every customer. We joined all five systems into a single view, one row per customer, rebuilt every night and checked against Shopify's own figures. When a draft is written, the latest order and delivery status is read live from Shopify, so it is never out of date.
A specialist for every kind of email. Each email is sorted by type and handed to one of 13 drafters: returns, delivery, sizing, loyalty, gift cards, repairs and more. Each was written from real threads and works from a written policy with a named owner, so a draft quotes the policy rather than inventing one. What the team knows that is written nowhere else lives in one shared document they can add to just by asking.
Automation where AI is not needed. Joining the data, refreshing it and checking every draft against the brand's own vocabulary are plain rules, not AI. A guardrail that depends on the model remembering is not a guardrail. AI does the two things that need judgement: reading what the customer wants, and writing the reply.
How we know it works
Before anyone relied on it, we tested it blind. We took 100 customer emails the team had already answered, chosen at random across nine weeks, drafted a reply to each, and compared it with what the team actually sent.
93% of drafts were as good as the team's reply or better, and 29% were better. After fixes, none contained a wrong fact or a promise the business would not make, and every one passed the brand voice check. In a handful of cases the tester glimpsed the team's reply first, so the score may flatter it slightly.
The testing did not stop at launch. Every Monday, last week's drafts are scored against what the team sent, so any slip shows up within a week.
In practice
A complaint about a damaged jumper arrived. The draft waiting for it already showed that this customer had placed 35 orders worth more than £30,000. The reply was no longer just an answer to a question. It was a chance to keep one of the business's most valuable customers.
What changed
The team's job shifted from finding information to reviewing it. A draft is waiting in the inbox, labelled, with anything to check listed at the top. Nothing is sent without a person, and anything that moves money, such as a refund, stays theirs to do.
The bigger change is what the reply can do. Every reply knows who it is talking to, so a routine query becomes a chance to strengthen the relationship with the customers who matter most.
What comes next
Next, drafts will treat the business's top customers as a priority, putting their messages first, and will welcome back customers who have not ordered for a while. After that, a phone screen that shows the whole customer the moment they call.
Each is a new use of the same customer record, not a new system.
Which process is growth making harder for you?
Bring it to a first conversation. We will tell you honestly whether AI, automation or a better-connected system is the right answer, and what it could be worth.
Book a call