About Services Tools Blog Contact us

Customer Insight Tagging

Fills in a fixed set of customer profile details from till purchases and email correspondence, using a model run locally and a list of footwear customer groups.

  • Classification & Tagging
  • Footwear retail
  • Pipeline
  • Productised
  • Hybrid
  • 2026

The problem

A mailing platform cannot build useful audiences from raw till lines, and the signal for who a customer is sits in the inbox. Opening each thread by hand is too slow. So is tagging thousands of customers one at a time.

What it does

A fixed list of allowed values covers age, spend, footwear type, channel, season and life stage. A background job loads one customer with their purchases and emails, and asks a model to fill each blank detail in a set format. Brand names are copied from the supplier on each purchase line. Several computers can share the work.

Outcomes

  • Each customer carries audience tags such as boots, mid range and store regular.
  • Mail threads are labelled for complaints, wholesale enquiries and delivery chases.
  • The brand list comes from the till and not from the model.
  • A split queue lets several machines share the work.

Simplified mockup

Interactive preview with sample data. Nothing is sent to a server.

Nora Hale · C-18420

Subscribed
Customer
Nora Hale
Ashford · account opened 2019 · 5 purchases this year
Store card Family account 3 open threads
Purchases this year
DateItemSupplierPrice
12 MarField bootNorthfield£89
2 AprWalking shoeHarbour£64
18 MayChelsea bootNorthfield£110
9 JunChild trainerMillbrook£32
21 AugRain bootHarbour£48
Insight fields Not tagged
AgeEmptyWaiting
SpendEmptyWaiting
FootwearEmptyWaiting
ChannelEmptyWaiting
SeasonEmptyWaiting
Life stageEmptyWaiting
BrandsEmptyTill only
Tagging run
Ready to tag
  1. Load the customer
  2. Read the till lines
  3. Attach the mail
  4. Fill empty cohort fields
  5. Copy brands, no model
  6. Write the tags

Correspondence · Nora Hale

3 threads
Size complaint 4 Sep

The Northfield field boot in a 7 pinches across the toe after a morning on the marsh. I would like the wide fit if you have it in store.

Used for fit and footwear type. Not a brand guess.

Wholesale enquiry 22 Jul

Three Counties Outfitters has three shops. I would like a trade account for autumn boots, paid on 30 days.

Life stage stays family shopper. Trade is a thread label, not a cohort value.

Delivery chase 11 Aug

Order 44821, Harbour Lane wellingtons, was promised for Tuesday collection at the Ashford shop. Please confirm the van.

Channel evidence for store regular.

Insight workers

Shard 1
1,184
Tagged · other machine
Shard 2 · this machine
640
Tagged · 220 left
Shard 3
1,105
Tagged · other machine
Shard 4
940
Still waiting
Customers tagged per shard
Dry run
Nothing is written until you leave dry run

How it works

  1. Pick customers by number or by slice, and skip unsubscribed addresses unless told otherwise.
  2. Load the purchases, make the dates readable and attach recent emails.
  3. Ask the model to fill each blank profile detail, with a second model as backup.
  4. Copy brand names from the purchase lines without asking the model.
  5. Save the tags on the customer and log how long it took.
  6. Offer a practice run that changes nothing, and run the same job from a dashboard with a progress window, which can use Anthropic Claude instead.

Pipeline

flowchart LR
  A[Customer] --> B[Purchases]
  A --> C[Correspondence]
  B --> D[Ollama]
  C --> D
  B --> E[Brand list]
  D --> F[Cohort tags]