Skip to content
Free Consultation
Industries / E-commerce
🛒 E-commerce

800 Returns a Week.
Handled in 6 Hours, Not 4 Days.

Returns, catalogue work, and customer queries grow with your revenue. Scale them with headcount and growth eats your margin. We automate the volume — with consistent decisions your policy actually intended.

14–17%
Average online return rate
as share of total sales
(NRF 2024 data)
4 days → 6 hrs
Returns SLA achieved
on our e-commerce project
€155K
Annual savings on
returns processing project

Growth That Makes Operations Worse, Not Better

E-commerce volume amplifies. When operations are manual, more orders means more returns, more queries, and more inconsistencies — at scale.

📦
Returns Volume That Scales With Revenue

At a 15% return rate, a €10M business handles 1,500+ returns per €1M in sales. Assessing condition, applying policy, and issuing refunds by hand is a department that grows with revenue.

Inconsistent Policy Decisions

The policy says 30 days, full refund on unused items — but the edge cases (worn once, missing tag) get decided by whoever's on the queue that day. Same situation, different answers.

📉
4-Day SLA Killing Repeat Purchases

67% of shoppers check the returns policy before buying (NRF). A slow resolution erodes trust, even when it's correct. The customer who waited 4 days for a refund doesn't come back.

💳
Fraud and Chargeback Losses

Manual review can't spot patterns across thousands of transactions. The customer with 7 returns in 60 days doesn't stand out in a queue. Automated detection flags it before the refund goes out.

How the Returns System Works

AI Agents · E-commerce

800 Returns a Week — Without 4 Full-Time Staff

An online fashion retailer handling 800+ returns a week with four full-time staff, a 4-day resolution time, inconsistent policy calls, and no view of fraud. We built a returns system that checks item condition from photos, applies policy automatically, flags fraud, and sends only the genuine edge cases to a person. One person now does what four used to.

Before Manual review Inconsistent decisions 4-day SLA
After Return received Vision condition check Policy engine + fraud check 6-hour resolution
Annual savings
€155K
Resolution SLA
4 days → 6 hrs
Headcount
4 FTEs → 1

3 FTEs reallocated (€38K × 3 = €114K) + estimated fraud/chargeback reduction (€41K). Payback: ~10 weeks.

See this and 8 more case studies →

Consistent Decisions, Faster Resolutions, No Extra Headcount

The goal isn't removing humans from hard decisions. It's removing them from the 90% of decisions that aren't actually hard.

Vision-based condition assessment

It reads the customer's photos and judges condition — unused, worn, damaged, missing tags — against your policy. Every decision is logged with the image and the rule applied.

Automated policy engine

Your policy becomes rules you can change yourself — no developer needed. Refund, exchange, store credit, or reject, applied the same way on every return.

Fraud and anomaly detection

It flags the patterns a queue hides — frequent returners, inconsistent claims, mismatched details — before the refund runs, then sends them to a person with full context.

Human exception queue

Genuine edge cases — unusual damage, disputes, policy gaps — go to one reviewer with everything attached: images, history, and policy reasoning. People handle only what the system couldn't.

What Runs the Returns System

GPT-4 Vision
Judges item condition from photos
Python / FastAPI
Runs returns intake and the policy engine
LangChain
Walks each return through your policy
Redis
Keeps the queue and statuses live
PostgreSQL
Stores returns history and the audit log
React (exceptions UI)
Where your reviewer sees each case

How many returns are you processing a week?

Share your returns volume, your team, and your main SLA challenge. We'll map what's automatable, what needs a human, and what payback looks like.

Estimate my returns savings