Spiky demand
Traffic is not a smooth curve. It is eleven ordinary months and a handful of hours that pay for them, and those hours are when the system is least tested.
eCommerce & retail
The storefront is the easy half.
Most retail platforms look fine until something happens. A campaign lands, a product goes viral, a supplier feed changes shape, and the part of the system nobody demonstrated in the pitch turns out to be the part that decides your quarter. Front ends are well understood. Inventory truth, order state and the back office are where retail businesses actually break, and that is where our ecommerce software development starts.
The question to answer first
“Will this hold when we do ten times the traffic on one day of the year?”
2
Published client reviews from retail and ecommerce businesses
The pressure
Four things that decide a retail quarter.
Inventory truth
The number on the site, the number in the warehouse and the number in the accounting system are three different numbers. Overselling is a systems problem wearing a customer-service costume.
Channel sprawl
Web, marketplace, retail, wholesale and social each want their own view of the same catalogue, and each has its own idea of what a product is.
Margin pressure
Acquisition costs rise, returns are expensive, and the operational efficiency of the back office is now the thing that decides whether growth is profitable.
The peak day
What we typically do in the six weeks before your biggest hour.
Capacity is not a setting. It is a piece of work that happens before the campaign, against the number you actually expect, with the results written down. The timings below are typical, and confirmed in writing once scope is agreed.
01
Week minus 6
Agree the number
Peak concurrent sessions, orders per minute and payload sizes, taken from last year plus the growth you are planning for. Everything after this is measured against that figure.
02
Week minus 5
Load test to failure
Not to the target, to the point where it breaks. You need to know what breaks first and what it does when it goes, because that is what you will be managing on the day.
03
Week minus 4
Fix the first bottleneck
Often a database connection pool or an unbounded query behind a category page, and rarely the thing anyone predicted.
04
Week minus 2
Degrade deliberately
Recommendations, reviews and personalisation are switched off under load, by design, with a flag. The checkout keeps working while they are gone.
05
Week minus 1
Rehearse the incident
Someone runs the runbook while someone else breaks things. The point is to find out who has the access they will need at two in the morning.
06
The day
Watch orders, not CPU
The metric that matters is completed orders per minute against forecast. Infrastructure graphs are diagnosis, not detection.
The work
What we build for retailers
The storefront is the part everyone demonstrates. These are the parts that decide whether growth is profitable.
Commerce platforms
Custom where the standard cart genuinely does not fit, and honestly assessed when it does. If a standard platform is the better answer, we will say so before you spend anything.
Order and inventory systems
One authoritative view of stock across every channel, with reservations, backorders and returns modelled properly rather than patched.
Back-office tooling
The screens your operations team lives in all day. Usually the highest-return software a retailer builds and usually the last thing it funds.
Product information management
One catalogue, many channels, with the transformations each channel needs handled once rather than in five exports.
After the launch, we noticed that users and customers had a better experience and gave positive comments about the site’s usability. The number of people leaving the site quickly went down, more time was spent looking at products, and fewer ordered were abandoned.
Read the full review
Non-negotiable
Four rules we do not trade away for a deadline
- Peak is the spec
- Capacity is designed against your worst hour, not your average one, and it is load-tested against that number before the campaign, not after.
- Idempotent orders
- A customer who double-taps checkout on a flaky connection must not be charged twice. This is a correctness requirement, not an edge case.
- PCI scope
- Card data is kept out of your systems wherever possible. The cheapest compliance is the compliance you never took on.
- Graceful degradation
- When the recommendation service is down the site still sells. Non-essential systems must not be able to take the checkout with them.
Where AI actually lands
Four places it earns its cost, and one line we do not cross.
Worth doing
- Demand forecasting and replenishment, where a small accuracy gain moves working capital.
- Search and merchandising, which is a retrieval problem long before it is a model problem.
- Customer service triage, answering the routine order-status questions so your people handle the rest.
- Catalogue enrichment, turning supplier feeds of varying quality into consistent product data.
Where we stop
Pricing without guardrails. An automated pricing loop with no floor and no human review is a well-documented way to sell your inventory at a loss overnight, and it has happened to larger companies than yours.
Evidence
We have no retail case study to publish yet. Two published client reviews come from retail and ecommerce businesses, including Sahara Centre, whose ecommerce user experience and interface we redesigned.
How we deliver it
The services this work draws on.
Web App Development
Fast, accessible applications that behave correctly on a bad connection.
Backend Development
APIs, data models, and the boring reliability that everything above depends on.
UX and UI Design
Interface design grounded in how the work actually gets done, not how it demos.
Other sectors we work in
Next step
Tell us what your biggest day of the year looks like.
A senior engineer reads every enquiry and replies within one business day with an honest view of what it will take, including when a standard platform would serve you better.
