Case study

How Moose Knuckles quantified its first three AI use cases

An opportunity audit run on the brand's own data: three months of support tickets, the live product catalogue, the data architecture. Three prioritised, costed and sequenced use cases, in a company with no AI in production at the start.

Published 30 July 2026

Industry

Premium retailLuxury outerwear, Montréal

Scope

3 areas auditedCustomer service, product content, BI

Tandem service

AI opportunity auditQuantified on the client's own data

Output

RoadmapConservative ROI, cost, phased plan

The challenge

No AI in production, and no visibility on where it would pay off

Leadership wanted a rigorous, numbers-first read on the real opportunities, not a generic technology pitch. Three pain points stood out. Customer service first: a queue where fewer than one ticket in two was genuine demand, with first-response times running into hours at seasonal peaks. Product content next: three months from concept to a live product page, while roughly half of all images were only colourway variants of a single master shot. And data last: business users lost 5 to 10 hours a week rebuilding reports by hand.

1 in 2

tickets that were genuine demand

A saturated support queue, with first-response times running into hours at seasonal peaks.

3 months

from concept to a live product page

Roughly half of all images were only colourway variants of a single master shot, reshot in studio.

5-10 h

a week lost rebuilding reports

Data stayed behind the BI team, and a large majority of dashboard users had stopped logging in.

The company

Moose Knuckles, Montréal

Moose Knuckles

Moose Knuckles is a premium luxury outerwear brand headquartered in Montréal, selling through wholesale, direct-to-consumer stores and e-commerce, with China as its primary growth market.

Extreme seasonality concentrates the majority of revenue in a short winter window. That makes fast, data-driven decisions on inventory, content and customer service directly material to margin: on a cycle that short, a late call cannot be recovered.

Headquarters

Montréal, Canada

Industry

Luxury outerwear

Channels

Wholesale, retail, e-commerce

Growth market

China

Our approach

Quantify on the client's data, not on assumptions

The audit ran in two phases, stakeholder interviews and data collection, then analysis and prioritisation alongside the client's own teams. The principle: build every recommendation on Moose Knuckles' real data, so every figure is defensible and owned by the client.

Interviews and data collection

A first phase with stakeholders, to scope the work from the ground up rather than from assumptions.

Analysis on real data

Three months of support tickets, the live product catalogue, public reviews, operational KPIs and the existing data architecture.

AI-assisted classification and sizing

Volumes were classified and quantified to surface recurring patterns and size each opportunity, with explicit assumptions.

Three prioritised, sequenced bundles

Each with a conservative ROI model, an implementation cost and a phased, reversible build plan: the client validates value at every step before committing further.

Results

Three quantified opportunities, ready to build

The audit produced conservative, sourced numbers for each of the three use cases. These are identified opportunities, not gains already realised. On customer service, a first-line and overflow agent: analysing 7,612 tickets across 10 categories showed that around 32 % of real demand is automatable today, a deliberately conservative floor, while defects, warranty and disputes stay with human agents. Cost per handled ticket drops from the order of a few dollars to a few cents on that share, which reads as capacity returned to the team rather than headcount reduction. On product pages, generating colourway variants from a single real master shot makes around 3,750 images a year addressable: unit cost falls from roughly a hundred dollars in studio to a few cents, and the three-month cycle compresses into weeks. On conversational BI, a natural-language querying layer over the existing data warehouse, built in three phases sharing one backend, with payback identified at four to seven months on recovered time alone.

use cases quantified, with cost and ROI

tickets analysed to size the potential

of real demand automatable, a conservative floor

4-7 months

payback identified on conversational BI