Mender AI
Work

Prediction at national scale, and tools for businesses with nobody to build them.

The first two were built inside large UK companies. The last two were built for clients, whose names are withheld. Both matter: the modelling is where the method comes from, and the client work is where it meets a business that has no technical team.

Demand forecasting In-house, FTSE-listed retailer

Predicting demand across tens of thousands of product lines

Forecasting algorithms for a major UK retailer, estimating how much of each line would sell, where, and when — the numbers that buying, allocation and markdown decisions were then made against. At that scale a forecast is not one model but a system: seasonality, price changes, newness, and lines with almost no history all have to be handled differently, and the output has to be trusted by people who are not statisticians.

Scale
Tens of thousands of product lines, forecast continuously
Used for
Buying quantities, allocation, markdown timing
Applies to
Any business holding stock, planning production, or committing to orders in advance
Risk and pricing models In-house, UK consumer lending

Prediction where the decision has money attached

Credit risk and affordability modelling, and prediction of where competitors' prices sit against market data. Lending is a useful discipline to have learned in: a model that cannot be explained is worthless however accurate it is, because the decision has to stand up to a regulator and to the customer it affects. The same standard applies to a small business deciding which orders to accept — you need the number and the reason behind it.

Built
Risk and affordability prediction; competitor price modelling from market data
Constraint
Every output explainable, not just accurate
Applies to
Credit terms, order acceptance, pricing against a market you cannot see directly
Verification portal Client build, employee screening agency

One place to see where every candidate's documents stand

Candidate paperwork moved between HR, the verifying team and the candidate by email, and the answer to "where is that one up to?" meant asking someone. The portal gives each of the three roles its own view of the same case file, with a status on every document and a record of who changed it. Nothing is chased by phone and nothing sits unread in an inbox.

Roles
HR, verifier, candidate — separate access, one case file
Built with
React front end, role-based access, document status tracking
Status
Reviewed and approved by the client; deployment in progress
Knit chart converter Client build, knitwear manufacturer

Artwork to machine-ready stitch grid, without colour drift

High-resolution client artwork had to be reduced by hand to a low-resolution bitmap for the knitting machine. Every downscale introduced colours that were not in the design, and each one was found and corrected by eye. The converter reduces the image against the supplied colour legend, so no new colour survives the process, and it refuses any output that exceeds the machine's stitch ceiling. A second converter handles scarves, where the width is capped far lower.

Constraint
346 × 509 stitches per garment panel; 120 stitches across for scarves
Input
Client artwork plus the colour legend already supplied with it
Status
Working tool, proven on live client designs
What else fits

Nearly every job we take on is one of four shapes.

The sector changes constantly. The shape almost never does. If one of these sentences sounds like something said in your business, that is the kind of job this is.

Guessed, when it could be predicted
“We order what we ordered last year, and hope.”

A forecast built on your own trading history, with an honest statement of how confident it is.

Reorder levels, production quantities, next week's staffing, which bookings get missed, which orders carry risk.

Retyped, when it could be read once
“It arrives on paper and somebody types it in.”

A pipeline that reads the document and returns structured output in your own template.

Delivery notes, site and inspection forms, supplier invoices, handwritten job sheets, timesheets.

In someone's head, when it could be on a screen
“Only Dave knows where that order has got to.”

One shared view with a status on every item, and a record of who changed what.

Jobs across a workshop, cases in progress, who holds which document, what the drivers actually delivered.

Worked out by hand, when it could be calculated
“Every quote comes out differently depending on who wrote it.”

A calculator that works from your real inputs and your own margins, the same way every time.

Quotes and estimates, job costings, artwork and file prep, delivery rounds, rotas.

The four shapes are why the sector matters less than it looks. A knitwear factory and a screening agency had nothing in common except that both were doing one of these by hand.

Something here look like your problem?

Describe the job See what it costs