What our clients say
These are direct quotes collected after project completion. We ask every client to rate the engagement once the model has been in production for at least 60 days.
We asked Proven Aics to build a demand-forecasting model for our 14 distribution centres. Before the project, our planners relied on spreadsheets and gut feeling. After 10 weeks the model was live, and within the first quarter our overstock costs dropped by 23%. The team explained every decision in plain English, which mattered a lot to our board.Helen Marchetti, operations director at Fieldway Logistics
Our compliance team was drowning in document reviews. Proven Aics built an NLP classifier that routes incoming regulatory filings to the right analyst and highlights the paragraphs that need attention. Processing time per document went from 40 minutes to under 8. That freed up two full-time equivalents for higher-value work.Rajan Deshpande, head of compliance at Northgate Capital
We hired them to detect surface defects on aluminium coils using camera feeds on our rolling line. The accuracy reached 96.4% within six weeks, which was above our 95% target. One thing I appreciated: they spent two days on the factory floor before writing a single line of code, so the model actually understood our edge cases.Tom Birkett, plant manager at Arden Metals
I was sceptical about AI for our small e-commerce brand, but the churn-prediction model they built paid for itself in the first month. We now send targeted retention offers to the customers most likely to leave, and our 90-day repeat purchase rate climbed from 31% to 44%. The dashboard they gave us is genuinely easy to read.Priya Okonkwo, founder of Claro Skincare
Case study: reducing energy costs for a hotel group
Pennine Hospitality Group
Pennine Hospitality runs seven hotels across Yorkshire and the Lake District. Their energy bill exceeded £1.8 million a year, and the building management systems in each property operated on fixed schedules that ignored occupancy patterns and weather.
We installed IoT sensors in 340 rooms and common areas, then trained a reinforcement-learning agent that adjusts heating, cooling and lighting in 15-minute intervals. The agent learns from occupancy data, booking forecasts and Met Office weather feeds. It took 12 weeks from sensor installation to the first hotel going live, and a further four weeks to roll out across all seven properties.
Within six months, the group had cut energy consumption by 19%, saving roughly £342,000 annualised. Guest comfort scores on post-stay surveys actually improved by two percentage points, because rooms reached the right temperature before guests arrived rather than after.
Case study: fraud detection for an online marketplace
Brickline Marketplace
Brickline is a peer-to-peer marketplace for second-hand electronics. Fraudulent listings and payment disputes were costing them over £90,000 a month in chargebacks and manual review labour. Their existing rule-based system flagged too many legitimate sellers, causing friction and lost revenue.
We built a gradient-boosted ensemble model trained on 18 months of transaction history, user-behaviour signals and device fingerprints. The model scores every new listing and every payment event in under 200 milliseconds. Listings above the risk threshold go into a short manual queue; the rest proceed automatically.
After three months in production, chargebacks dropped by 61% and the false-positive rate fell from 12% to 3.8%. The manual review team shrank from nine people to four, and those four now focus on the genuinely ambiguous cases rather than rubber-stamping obvious false alarms.