We build machine-learning models, computer-vision pipelines and predictive-analytics tools for mid-sized UK companies. You bring the data; we turn it into decisions.
Four practices, each with its own engineering team. Pick one or combine several into a single engagement.
We train regression and classification models on your historical records, typically 12 to 36 months of transactional data. The output is a forecast API your existing software can call. Turnaround from first data handover to a production endpoint runs about six weeks for a single-target model.
From quality-control cameras on a production line to document scanning for insurance claims, we design detection and segmentation models that run on-premise or in the cloud. Most projects start with a 500-image labelling sprint so we can benchmark accuracy before scaling up.
Customer-support ticket routing, contract clause extraction, sentiment monitoring across social channels. We fine-tune transformer models on your domain vocabulary so the system learns the jargon your staff already use, rather than forcing everyone to write in textbook English.
Before any model gets built, we audit your data estate: where it lives, how clean it is, what gaps exist. The deliverable is a 20-page report with a prioritised roadmap, cost estimates and risk flags. Some clients use the audit alone and hand the build to their in-house team.
Substitute AI Vision started as a two-person consultancy helping local retailers forecast stock levels. Five years later the team has grown to 28 engineers, data scientists and project managers, all based at 74 Ullrich Ridge, Manchester M1 5AN, Greater Manchester, United Kingdom.
We work exclusively with UK businesses. That keeps us in the same time zone, under the same data-protection rules and close enough for on-site workshops when a video call won't cut it. Our longest-running client, a logistics firm in Leeds, has been with us for four years. The shortest engagement was a three-day data audit for a Nottingham brewery that wanted to know whether AI was worth the investment (it was, but only for one of the three use cases they had in mind).
Every project starts with a scoping call, usually 45 minutes, where we figure out whether the problem is genuinely a good fit for machine learning or whether a simpler statistical approach would serve you better. We turn away roughly one in five enquiries because the data isn't there yet. Honest advice up front saves everyone time.
"They told us our second use case wasn't ready for AI and saved us about £40,000 in wasted development. The demand-forecasting model they did build paid for itself within three months."
Rachel Ng, operations director, Pennine Logistics
"Our insurance-claims team now processes documents in a third of the time. The NLP model picks out clause references and flags anomalies before a human even opens the file."
David Hargreaves, CTO, Albion Underwriting
"The data audit alone was worth the fee. We thought we needed a recommendation engine; turns out a simple clustering model on our existing CRM data did the job."
Sunita Patel, founder, Leaflet Box