TREN

E-commerce · Scope to be defined during discovery

AI System for Matching Similar Products by Visual Features

We design an AI system that reviews product images, category data and stock records with their source and recency details. The system prepares an in-stock similar-product suggestion but does not treat it as a final decision. If products look similar but serve different functions, the suggestion does not become an action; a product manager makes the final assessment.

Representative product discovery ops panel with a work queue, checks and an audit trail: in-stock similar product queue — **In-stock similar product queue** Representative interface — contains no real data. Product and supplier names are coded. The panel does not publish similar products or display them to customers.

The problem

In day-to-day operations, product images, category data and stock records arrive in separate records and files, preventing the team from reviewing the same case in a single view.

As a result, the link to source data may be lost while preparing an in-stock similar-product suggestion, leading to repeated checks and uncertainty about the basis for the decision.

Incomplete automation may overlook products that look similar but serve different functions, while unsuitable automation carries the risk of a misleading suggestion.

What we set out to improve

Time taken for the in-stock similar-product suggestion draft to be ready for review
Correct routing of records to the appropriate specialist queue when products look similar but serve different functions
Logging of suggestions amended or rejected by the product manager

How the system works

01 Retrieve product images, category data and stock records from an authorised source
02 Verify the source system, record identifier and recency details
03 Transform the fields into the target process schema
04 Prepare an in-stock similar-product suggestion draft
05 Compare the draft with business rules and source records
06 Check for the exception: products look similar but serve different functions
07 The product manager approves, amends or rejects the draft
08 Write the input, suggestion, amendment and final decision to the audit log

Methods we used

Source-traceable data extraction for product images, category data and stock records
Structured output that constrains the in-stock similar-product suggestion fields through business rules
An exception gate for products that look similar but serve different functions
A queue showing the source and suggestion together for the product manager
Duplicate-processing prevention and rollback records to manage the risk of a misleading suggestion

Where people stay involved

The product manager owns the final decision. The workflow pauses when products look similar but serve different functions, and the supporting records are presented to the specialist. The specialist can amend or reject the suggestion, or stop the process.

Data and security

Only the fields required for the task are processed from product images, category data and stock records. Source-system permissions are preserved, and personal and commercially sensitive data is minimised before being passed to the model. Every read, suggestion, human amendment and write to the target system is logged.

Who this suits

A good fit

E-commerce teams that carry out repeated reviews
Companies with defined source ownership and approval responsibility
Organisations that want exceptions to remain under human oversight

Not a good fit

When source data is not current
When the decision owner and rollback process have not been defined
When the existing method is sufficient for the low volume of cases

Frequently asked questions

What data is used?

The system uses product images, category data and stock records. Exact connections, fields and retention limits are defined during the access review.

How are exceptions handled?

The workflow pauses when products look similar but serve different functions. The record is sent to the product manager's queue with its source evidence.

How is unsuitable automation constrained?

Because of the risk of a misleading suggestion, no final action is taken without human approval. The source, suggestion and human decision are logged separately.

Project led by:DijitalPi

✦ PI ASSISTANT · ARTIFICIAL INTELLIGENCE

online

Ask the DijitalPi AI Assistant your question

No need to fill in a form and wait for a reply — write your question and let Pi, trained on DijitalPi's 20 years of know-how, answer within seconds.

Instant answers, 24/7
Service-specific context — no generic replies
Can connect you to a free consultation if you wish

Hi! 👋 I'm Pi — DijitalPi's AI assistant. Write whatever you'd like to know about AI System for Matching Similar Products by Visual Features and I'll answer right away.

Pi Assistant answers are for information purposes; for a proposal get in touch.

ASK AI ABOUT DIJITALPI

Let an AI explain what DijitalPi does.

Opens your chosen assistant with a ready research prompt. It reads the site live and answers.

ChatGPT Claude Gemini Grok Perplexity