// Demandscape website — Platform (how the digital twin is built) function PIcon({ name, size = 22 }) { return ; } // Placeholder for imagery / data-viz to be supplied later function Placeholder({ label, minHeight = 240, tone = 'light' }) { const dark = tone === 'dark'; return (
[ {label} ]
); } // ── How Demandscape works — 4-stage pipeline schematic ─────────────── function PipelineSchematic() { React.useEffect(() => {if (window.lucide) window.lucide.createIcons();}); const chipStyle = { fontFamily: 'var(--font-mono)', fontSize: 12.5, letterSpacing: '.02em', color: 'var(--ds-navy)', background: 'var(--ds-grey-50)', border: '1px solid var(--ds-grey-200)', padding: '.5rem .8rem', whiteSpace: 'nowrap' }; const stageLabel = (tone) => ({ fontFamily: 'var(--font-mono)', fontSize: 13, letterSpacing: '.14em', textTransform: 'uppercase', color: tone === 'dark' ? 'rgba(255,255,255,.65)' : 'var(--ds-grey-500)' }); const iconBox = (tone) => ({ width: 48, height: 48, display: 'flex', alignItems: 'center', justifyContent: 'center', background: tone === 'dark' ? 'rgba(255,255,255,.10)' : 'var(--ds-blue-soft)' }); const cardTitle = (tone) => ({ fontFamily: 'var(--font-display)', fontSize: 'clamp(1.4rem,1.7vw,1.7rem)', lineHeight: 1.1, letterSpacing: '-0.01em', margin: 0, color: tone === 'dark' ? '#fff' : 'var(--ds-navy)' }); const cardBody = (tone) => ({ fontFamily: 'var(--font-sans)', fontSize: 16, lineHeight: 1.55, margin: 0, color: tone === 'dark' ? 'rgba(255,255,255,.82)' : 'var(--ds-navy)' }); function Card({ tone = 'light', stage, icon, title, children, footer }) { const dark = tone === 'dark'; return (
{!dark &&
}
{stage}

{title}

{children}
{footer}
); } function Arrow() { return (
); } return (
[ How Demandscape works ]

Creating a digital twin of a customer base

800k customers 12k products 25M baskets per yr
}> Loyalty-card basket data containing the choices customers make repeatedly over time, against a cost. 1.5M weights

Developed from original research at the University of Oxford.

}> We train a constrained deep learning model which combines a deep-learning with an economic model of customer choice. The model learns the purchase drivers behind every choice — a living twin of your customer base, not a record of the past. “What will customers do if…?”

}> Simulate any scenario, including ones that have never happened before, to predict customer demand for every product and every customer type. ); } // ── Section visuals: bracket-framed Lucide diagrams ────────────────── const visFrameBare = { width: '100%', aspectRatio: '1 / 1', display: 'flex', alignItems: 'center', justifyContent: 'center' }; function ArchetypesDiagram() { const cols = ['var(--ds-blue)', 'var(--ds-purple)', 'var(--ds-amber)', 'var(--ds-green)', 'var(--ds-red)', 'var(--ds-blue-700)']; return (
{cols.map((c, i) =>
)}
); } function WelfareDiagram() { const base = [1, 1, 1, 0]; return (
{base.map((c, i) =>
{c ? : }
)}
); } function PurposeDiagram() { const goods = [['milk', 'var(--ds-blue)'], ['egg', 'var(--ds-amber)'], ['croissant', 'var(--ds-purple)'], ['apple', 'var(--ds-green)'], ['carrot', 'var(--ds-red)']]; return (
{goods.map(([n, c], i) =>
)}
); } function PlatformPage({ onNav }) { const { Button, Card, Badge, BracketFrame, Stat } = window.DemandscapeDesignSystem_0e6a32; React.useEffect(() => {if (window.lucide) window.lucide.createIcons();}); return (
{/* ── HERO ─────────────────────────────────── */}
Platform

How Demandscape builds a digital twin of your customer base

{/* ── PIPELINE ─────────────────────────────── */}
{/* ── PURCHASE DRIVERS ─────────────────────── */}
[ Purchase drivers ]

What drives your customers, discovered from data, not assumed from theory

We provide Demandscape with no product characteristics and no customer attributes. The model infers what matters entirely from the decisions customers made and their context.

The result is a set of purchase drivers that are used to motivate the digital twin.

Purchase drivers visual
{/* ── CUSTOMER ARCHETYPES ──────────────────── */}
[ Customer archetypes ]

The pure expressions of how different shoppers decide

We can map the purchase drivers to identify a set of customer archetypes which represent the range of behaviours within the customer base.

These archetypes exist as real customers, although most real shoppers are a weighted blend of multiple archetypes.

A synthetic panel of these archetypes reproduces the behaviour of the full customer base.

Archetypes can be mapped directly onto any existing segmentation framework so that insights are readily interpretable and actionable.

{/* ── CUSTOMER WELFARE ─────────────────────── */}
[ Customer utility ]

An early signal of which customers are at risk, before churn shows in your data

As a natural output of the digital twin, we can calculate utility for different customer types. It is a measure of how well your current range, pricing and format choices are meeting underlying needs.

For example, a set of price increases or changes in range, might trigger a decline in utility for certain groups, even as spending remains the same. This is a warning signal of customer attrition.

The model can be used to evaluate interventions to maintain customer utility and reduce chances of attrition.

{/* ── CUSTOMER PURPOSE ─────────────────────── */}
[ Customer purpose ]

Finding ways to increase share of wallet

The customers within the digital twin are driven by distinct shopping purposes which reflect the reality that they split their spend across multiple stores.

Within the model we can identify how different customer's purposes can be extended to increase their spend within the store.

{/* ── TECHNICAL FOUNDATIONS ────────────────── */}
[ Technical foundations ]

Built on leading AI research

Demandscape was developed from original research at the University of Oxford, combining an economic model of consumer choice with a deep learning architecture. The result is a constrained deep learning model (CDLM) which can capture the complexity of grocery purchase behaviour across a full retailer range; structured to respect the economic principles that govern customer decision making.

Demandscape is built on JAX and runs on NVIDIA GPU infrastructure. Training data for our most recent deployment covers approximately 350 million data points: one year of transaction history across 800,000 customers, 12,000 products, and 20 million weekly shops.

And the model is designed to scale to significantly larger data sets.

JAX
NVIDIA
); } window.PlatformPage = PlatformPage;