Kirk Thompson Jordan Thompson
Hire Us Immediate start
All work

Data Intelligence · 2024

03

Pulse
Analytics

ML Analytics Dashboards Development
Pulse Analytics dashboard

See what's coming
before it arrives.

A fast-growing retail group had invested heavily in BI tooling but was still making decisions on last week's data. By the time their dashboards showed a problem, it was already costing them revenue. They needed intelligence that was genuinely predictive — not just a better view of the past.

KIJO built Pulse — an ML-powered analytics layer that sits on top of existing data infrastructure and adds genuine predictive capability. Anomaly detection, forecast models, and proactive alerts — all surfacing through a dashboard their operations team actually wanted to use.

Client type Retail / Multi-site
Engagement Retainer · 10 weeks
Services Development + Consultancy
Delivered Q4 2024

Reporting the past,
missing the future.

The existing BI stack was excellent at describing what had already happened. Revenue reports, stock reports, footfall reports — all accurate, all too late. Stock-outs were identified after they'd cost sales. Demand spikes were noticed after the peak had passed. The data existed to have predicted every one of these events — but it wasn't being used predictively.

Backward-looking data

24-48 hour reporting lag meant operational decisions were always made on stale data — too late to prevent issues.

No anomaly detection

Unusual patterns in revenue or stock required manual analysis to identify — problems weren't found until they were obvious.

Alert fatigue

Existing monitoring produced hundreds of low-quality alerts daily — the team had learned to ignore them.

ML-powered
early warning.

Pulse adds a predictive layer to their existing data stack without replacing it. Three ML models run continuously: demand forecasting, anomaly detection, and supply risk scoring. Each produces a small number of high-confidence, actionable alerts — surfaced through a clean operations dashboard and a daily digest to leadership.

Demand forecasting

Time-series model predicting demand at SKU level with 92% accuracy — 6 hours before real-time systems would flag it.

Anomaly detection

Unsupervised model detecting revenue and stock anomalies in real time — with contextual reasoning explaining each alert.

Smart alert layer

Confidence-weighted alerting reducing noise by 94% — only surfacing anomalies worth acting on, with suggested actions.

92%

Forecast
accuracy

6hrs

Earlier
warnings

4×

Decision
speed

10wks

From brief
to live