We ask every client for a candid review after project wrap-up. Here are some of the responses.
We gave them 18 months of messy sales data and asked whether we could predict which wholesale accounts would reorder within 60 days. The model they built hit 83% accuracy in the first month. Our sales reps now prioritise outreach based on the model scores, and reorder rates climbed 12% last quarter.Helen Marsh
I was sceptical about AI for a company our size (22 employees). The team showed us a working prototype of an automated invoice-matching tool in three weeks. It now handles about 70% of our purchase-order reconciliation without any human input. The ROI paid for the project fee within five months.Rajan Patel
Good experience overall. They built a sentiment-analysis dashboard for our customer-service emails and it works well. The only reason for four stars instead of five is that the initial data-cleaning phase took longer than the estimate, which pushed the timeline by about ten days. Communication throughout was clear, though.Gareth Llewellyn
We needed a chatbot that could answer questions about our product catalogue (around 4,000 SKUs). The NLP model they fine-tuned handles roughly 60% of inbound queries without escalation. Our support team went from answering 200 chats a day to about 80, which freed two agents to focus on complex returns.Nia Okonkwo
The client handled around 1,200 container movements per month but had no reliable way to forecast weekly demand. Dispatchers over-booked hauliers some weeks and scrambled to find capacity others.
We ingested three years of booking data, weather records, port-schedule feeds, and UK bank-holiday calendars. After testing gradient-boosted trees against a simple ARIMA baseline, the ensemble model reduced mean absolute error by 34% compared to the spreadsheet method the team had been using.
The forecast now updates every Monday morning via an automated pipeline and feeds directly into their TMS. Haulier costs dropped by roughly £8,400 per quarter because the broker books capacity earlier at better rates.
About half our revenue each year comes from repeat engagements. Here is what clients tell us keeps them returning.
Every deliverable includes a summary written for non-technical stakeholders. No jargon-heavy slide decks that only the data team can parse. Board members and finance directors should understand what a model does and why it matters to revenue.
After the discovery phase, we quote a fixed price for the build. If we underestimate the effort, that is our problem. Clients have told us this removes the anxiety of watching billable hours climb during an open-ended engagement.
Every project includes 60 days of email support after deployment. If a model drifts or an integration breaks, we fix it. Two clients have extended this into ongoing retainers for quarterly model retraining, but there is no pressure to do so.