You have the data. Now get the answer out of it.
Custom models, built on your own data. The answers a chatbot can’t reach, because they have to be modelled, not read.
For the team that has the data, and still makes the same decision by gut.
- A PhD specialist on your problem
- A straight call on whether ML is the right tool
- Runs where your data is, and it stays yours
You tried AI on this, and it couldn't answer.
You’ve got the images, the sensor feeds, the years of records - and a decision you still make by hand. If a chatbot couldn’t help, it’s because it was built for a different job: a chatbot reads documents; a model learns from your measurements. The answer is already in your data. It just has to be modelled out.
Your data today
With a model on it
You tried a chatbot on it, and it couldn't answer the question.
A model built for the question, trained on your own data.
A chatbot answers what's written down; this learns what your measurements mean.
A person eyeballs every image, part or reading to make the call.
The model makes the routine call, and flags the ones a human should see.
Years of data sit in storage, doing nothing.
That history becomes a forecast, a risk score, or an early warning.
You'd need thousands of labelled failure examples to start.
Anomaly methods learn from normal operation alone - no failure library needed.
The model sounds just as sure when it's wrong.
Every output carries a measure of confidence, so you know when to trust it.
The data was always enough. It needed the right model on it.
Models for the answers that live in your data.
Custom models on your own images, video, sensors, signals and records - built from first problem framing through to production, with an early, honest call on whether machine learning is even the right tool.
Computer vision
Teach a model to see what your people check by eye, in images and video.
- Detect, classify and segment - defects, cracks, wear, objects and assets
- Pixel-level inspection that shows where the fault is, not just that there is one
- Works where the thing you're looking for is rare and examples are few
Sensors, signals and time series
Turn a measurement stream into something you can act on.
- Acoustic, vibration, environmental, flow, load and telemetry data
- Denoising, calibration, event detection and forecasting
- Built for how real sensors behave: drift, dropout, devices that disagree
Anomaly detection
Find what's abnormal without ever defining abnormal first.
- Trains on normal operation alone - no labelled history of failures needed
- Condition monitoring, quality assurance, unusual transactions and fraud
- Learn what normal looks like, then flag everything that isn't
Forecasting and uncertainty
Model what happens next, with an honest measure of how sure it is.
- Demand, capacity, risk and pricing, with a confidence range on every result
- Bayesian and statistical rigour - is the result real, or is it noise?
- Uncertainty quantification is our specialism, and the subject of a pending US patent
A specialist on your problem, who owns it end to end.
Dr Vincent Russell holds a PhD in applied mathematics (University of Auckland, Dean's List) and has spent 7+ years building ML and statistical models. He's co-inventor on a pending US patent for AI-based uncertainty quantification, and was lead ML engineer on an air-quality programme running in production.
The public-benchmark result above (roughly 10,800 photos across 12 products, one model, trained only on good parts) shows the approach; everything a procurement or legal team needs is in our Trust Centre
- PhD-led, so you get an answer even from data that defeats off-the-shelf tools.
- Runs where your data is - edge, on-premise or cloud - and your data stays yours, never used to train a model.
- Every model flags its own uncertainty, so your team knows when to trust it.
- A straight answer on whether ML is even the right tool - sometimes a simpler fix wins, and we say so early, before the spend.
We crunch through complex data to generate insights, and RIVER made it simple to work through, while keeping everyone aligned on our core value proposition.
Alex OsbornCTO, trev · 2024Prove it's feasible first, then build.
Feasibility
Weeks, fixed scope
Is there signal in the data, what accuracy is realistic, and is ML even the right tool? You get a written assessment and a baseline. A no arrives in weeks, not month nine.
Prototype
Next
A model trained on your own data, measured against the target we agreed at feasibility, with results you can inspect case by case.
Production
Then, and ongoing
Deployed into the environment it has to live in - edge, on-premise or cloud - with monitoring, versioning and retraining built in, so it stays honest as conditions change.
The investment
A short, fixed-scope study: is there signal, and is ML the right tool?
Or start on the Retainer from $5,000 a month. If something isn’t delivered, that month is free.
Every engagement includes
- A straight call on whether ML is the right tool for the problem
- A model trained on your own data, with results you can inspect
- A confidence measure on the model's outputs, so you know when to trust it
- Deployment where your data lives - edge, on-premise or cloud
- Monitoring, versioning and retraining, so it stays honest over time
- Your data and your model, yours to keep and never used to train anyone else's
The questions people ask when they’re ready to move.
It depends on the problem more than the volume. Some inspection tasks work from a few hundred photographs of normal operation; some forecasting needs years of history.
- Feasibility answers it for your case before you commit to a build - that’s the point of the first step.
- Representative data matters more than a huge pile. We ask for the messy cases too, because that’s where models fall over.
Often not. Asking a team for a thousand labelled photos of every way a part can fail is what kills projects at the data stage. Much of this work avoids that entirely.
- Anomaly methods train on normal examples, which you already have - no library of past failures needed.
- Where labels are genuinely needed, we design the smallest labelling effort that will do the job first.
Yes. Models can run on embedded and edge hardware, including sites with no connectivity, as well as in the cloud.
- Real-time inference on-site, for unreliable connections or latency-critical work.
- Your data stays yours, in the environment you choose, and is never used to train a model. The Trust Centre carries the detail your audit will ask for.
Then we say so, plainly, and point you to the simpler thing. Some problems are better solved by a rule, a threshold or a better sensor - and that call gets made early, in feasibility, not after months of spend.
A straight answer on whether ML is even the right tool is part of what you’re paying for.
An LLM answers questions found in documents; this models the answers found in measurements. They’re different disciplines, and this is the one for data an LLM can’t read.
- LLM and agent work reads and reasons over text, records and documents - our ERP and advisory work.
- This work learns from images, sensors, signals and numeric history, where the answer has to be modelled out of the data.
We do both, so a model lands inside a working process rather than beside one.
Easy to start. Fast to prove. Built to scale.
Tell us about your machine learning, and we’ll take it from there. Prefer to look first? Ask Moana, or take the free Readiness Check.
Ideas from the field.
Guides and research from a team that ships it.
Take
AI-First Sales: Back Your Team, Don't Replace It
The tempting shortcut is to hand sales to AI. The one that works is AI doing the research, the first reply and the nudges, so your people spend their time on the conversations that close.
TakeHighlight
Five frontier models in three days. The only thing that survives is your eval set
Anthropic, Google, Meta, Alibaba and OpenAI all shipped a frontier model in the first three days of September. The teams that switched in a day had one thing in common: 30 to 50 real examples from their own work, scored the same way every time. How to build yours this afternoon.
PerspectiveHighlight
Your AI vendor's pricing has a balance sheet under it
The five biggest US cloud companies are spending close to US$700 billion this year, most of it on AI, and the gap between that and what they earn is borrowed. Prices and terms can move for reasons that have nothing to do with you. Four questions to put to every AI vendor at renewal.
Teams like yours, in their own words.
A decade of work, across health, sport, finance and the public good.
RIVER learnt our business inside out, and found growth we hadn't seen ourselves. That is what changed our ability to scale.
Jay HarrisonFounder, Edison Health · 2022We crunch through complex data to generate insights, and RIVER made it simple to work through, while keeping everyone aligned on our core value proposition.
Alex OsbornCTO, trev · 2024Our business runs on what RIVER built. They have been central to our technology and our growth, and it keeps driving results.
Mark BramwellDirector, Commercial Realty · 2024RIVER listens, understands what we actually need, and gives us our time back. That is what a good partnership looks like.
Tony NaiduDevelopment Manager, Auckland Badminton Association · 2025RIVER helped us validate our market and launch our platform, and they haven't stopped since. They've been with us every step.
Mike CollinsFounder, iMatch Sport · 2025RIVER has been a trusted partner from the start, and they exceeded what we expected. We're already seeing the difference their work is making.
Ellen LekkaProgramme Specialist, UNESCO & SVSG · 2025The team at RIVER has been an incredible partner from day one. Their ongoing support helped make our success possible.
Scott TownshendFounder, trev · 2022RIVER approached every detail with care. Every decision was made with consideration and purpose.
Daemon CoyleGM, Mental Health Foundation · 2021Our new brand is bold and effective, and customers love it. The website has driven a real increase in leads.
Marcus SinGM, Cleanly · 2020We have always been ambitious about what our site should do, and RIVER take on the hardest parts of it. They make it happen, no matter how complex.
Chris HartFounder, Real Groovy · 2021RIVER are responsive and always ready to adapt. We have never had to chase them, and their support has been critical as our needs changed.
Supriya Kulkarni-PadhyeM&E Coordinator, Oceania Football Confederation · 2020We love our new brand. It reflects who we are as a team and resonates with our customers. Thank you for bringing our vision to life.
Tracy BergeGM, Greentree Advisers · 2021