Retail and E-commerce

Retail systems that hold up when the traffic arrives

We build the forecasting, support and data systems behind a retail operation, so the catalogue stays accurate, the support queue stays short and peak weeks stop being a fire drill.

Start a project
The sector

Retail runs on decisions made in advance

A retail business commits money before it knows what customers will want. Stock is bought ahead, shelf space is allocated, campaigns are booked. The record of what actually happened sits in a storefront platform, marketplace channels, a warehouse or 3PL system, a payment processor and a finance ledger, and those systems rarely agree on the same number.

The other half of the job is what happens after the order. Where is my parcel, the size is wrong, the card was charged twice. That volume is seasonal, repetitive and mostly answerable from systems you already run. It gets handled by people reading order records aloud, and it spikes exactly when the team has least room.

Villaex works on both halves. We build the pipelines that put sales, stock and returns into one modelled warehouse, the forecasting that turns that history into a per SKU plan, and the support agents that answer from live order data instead of guessing. We have shipped these as products, so we build for the week of the year when everything arrives at once.

What makes it hard

The constraints that shape retail and e-commerce systems

The catalogue is never clean

Product data arrives from suppliers, marketplaces and a long tail of manual edits. Titles disagree, variants duplicate, attributes are missing. Every forecast, search result and support answer inherits whatever is wrong in that catalogue.

Promotions poison the next forecast

A discounted week looks like demand, so the model reorders for a price that will not run again. New lines have no history at all, so their plan is a guess dressed up as a number.

Support volume is spiky

Order status, sizing and returns questions arrive in bursts around campaigns, sales and delivery delays. Staffing for the peak wastes money the rest of the year, and staffing for the average means the queue grows on your busiest days.

Returns eat the margin quietly

A return touches the storefront, the warehouse, the payment processor and the ledger, and each system logs it differently. Without a joined up record, nobody can say which products, sizes or suppliers are actually costing the business money.

What we build

Systems we ship for retail and e-commerce teams

Demand forecasting per SKU and location

A model trained on your own sales, stock and seasonality, producing a daily forecast per product and location. Promoted weeks are modelled apart from baseline, new lines borrow from similar ones, and reorder alerts fire while there is still time.

A warehouse the whole business trusts

Pipelines from the storefront platform, marketplace channels, the warehouse or 3PL system, the payment processor and the ledger into one modelled warehouse, with tested transformations and lineage, so sales, stock, refunds and margin reconcile in one place.

Support agents grounded in order data

Chat and voice agents wired into your orders, shipping and payments APIs, so they answer with live tracking, start a return, email a label, and hand the rest to a person with the transcript attached.

Catalogue and product data cleanup

Pipelines that normalise supplier feeds, match duplicate variants, fill missing attributes and draft product copy, with a review queue so a merchandiser approves changes rather than trusting a model to write to the live catalogue.

One return record, joined up

Returns joined from the storefront, the warehouse, the payment processor and the ledger, with reason codes normalised, so a merchandiser can read returns by product, size, supplier and channel and finance can reconcile refunds against payouts.

Proof

Screens from systems in production

app.shelfwise.io/worklist/reorder
Below par
Search⌘K
PM

Reorder worklist

2 criticalForecast refreshed 06:15 · 14-day coverSort: Urgency
ProductStoreOn hand / parSuggested

Aurora Oat Milk 1L

8412-OAT-1L

Kingsway
18/72
96
Order

Valemount Cold Brew 330ml

6620-CB-330

Kingsway
44/60
24
Order

Harlow Sourdough Loaf

3391-SD-800

Fairlane
26/40
18
Order

Marchetti Passata 700g

5107-PS-700

Kingsway
31/90
72
Order

Nord Sparkling Water 6pk

2244-SW-6PK

Fairlane
58/64
12
Order

Kestrel Free-Range Eggs 12

9038-EG-012

Kingsway
12/48
60
Order

Bramble Greek Yoghurt 500g

7715-GY-500

Fairlane
39/50
16
Order
7 of 18 · 298 units queued · cut-off 16:00Send purchase order
Predictive Analytics · Retail

Demand forecasting that cut stockouts by a third

A multi-brand retailer ran on week-old reports and gut feel, leaving shelves empty in some stores and overstocked in others. We unified their sales, inventory, and seasonality data and shipped a forecasting model surfaced in a real-time operations dashboard, turning a guessing game into a planned one.

quantvue.io/explore/revenue-by-region
ExploresRevenue by region

Net revenue by region

Weekly · USD · 12 Dec 2025 to 12 Mar 2026

WestMidwestNortheast
$120k$90k$60k$30k$0
W10 · West $118.4k
Dec 15Jan 05Jan 26Feb 16Mar 09
RegionOrdersNet revenueΔ prior
West18,402$1,284,910+12.4%
Midwest12,865$842,377+6.1%
Northeast9,143$611,204−2.8%
South7,690$498,552+3.7%
Mountain5,218$342,088+1.2%
Unified Data Platform · Retail

From conflicting reports to one source of truth

Sales, inventory, and customer data lived in systems that never agreed, and a planned AI project was blocked because the data wasn't usable. We built a unified platform with automated pipelines into a modeled warehouse, tested transformations, real-time dashboards, and governance. It became the foundation for everything next.

The services behind this work

AI AnalyticsWe build AI analytics that go beyond dashboards: forecasting what's next, flagging anomalies before they cost you, and answering questions in plain language, turning your data into decisions, on a foundation engineered for trust.
Data EngineeringGreat analytics and reliable AI both rest on the same foundation: clean, unified, trustworthy data.
Customer Support AgentsVoice agents that handle common support issues end to end: grounded in your knowledge, connected to your systems, and escalating to a human with full context for the cases that need one.
Process AutomationEvery business runs on repetitive, rules-heavy tasks that quietly consume your team's hours.
Enterprise IntegrationsYour business runs on a dozen systems that don't talk to each other.
Custom SoftwareOff-the-shelf tools force your processes into someone else's mold.
What we engineer around

Card data stays inside the payment processor's scope, so checkout and refund flows move tokens and never raw card numbers.. Customer records carry consent and deletion obligations, so exports, marketing tables and analytics models are built to honour a deletion request.. Storefronts are built against accessibility guidelines, because a checkout a screen reader cannot complete is a checkout that loses orders.. Marketplace and carrier APIs come with rate limits and contractual data rules, so pipelines back off and retry rather than hammer an endpoint..

FAQ

Questions we get before we start

Still unresolved? A 30-minute conversation with an engineer usually settles it faster than another page of copy.

Usually an audit of what your storefront, marketplace channels, warehouse system, payment processor and ledger actually hold, then one pipeline that puts those records in one place. Forecasting follows, at category level where history is thin and per SKU as the record builds. Something is running early, not at the end.

It will if you let it answer from memory. We ground the agent in your orders, shipping and payments APIs, so a tracking answer is read from the system rather than generated. Anything outside that grounding is escalated to a person with the transcript and order attached. You set the topics it is allowed to act on.

No replacement. Most of what we build sits alongside the systems you already run and reads from their APIs. It runs in your own cloud account, and the code, migrations and runbooks are yours, so your team can take it on or keep us on maintenance. Replacement is a separate conversation, and we would rather have it after the data layer is working.

That is the case we design for, not the exception. Jobs, agents and pipelines are load tested against your own busiest historical days, with queues, retries and rate limits set so a spike degrades slowly. Each marketplace and storefront sits behind one adapter with contract tests, so an API change fails loudly and an exception goes to a person rather than passing quietly.

Building something for retail and e-commerce?

A 30 minute conversation with an engineer is usually enough to scope it honestly.

Start a project