Decision intelligence for invisible losses in living systems

The worst losses are decided
before you can see them.

Loss in a living system behaves nothing like a machine breaking down: a herd, a shoal, a culture, a crop, a microbial reactor. The damage starts out of sight, the threshold keeps shifting, and there is seldom a single right answer. ruru forecasts that hidden risk, finds the optimal action within real limits, and makes the case so the call gets made: early, where it pays, before the loss is done.

Predict the risk · Optimise the response · Minimise the loss

How we choose a problem →
Risk over time the case for acting early
MOVING THRESHOLD LOSS BECOMES VISIBLE ACT NOW
Hidden risk What you can see The line that bites
Where it applies

Invisible, valuable, and treated as 'unsolvable'.

Each of these issues with living systems shares a common shape. In every one, the industry can already sense something: a THI reading, a pH trace, a spore count, a satellite bloom map. What none of it delivers is the optimal action, early enough, under real constraints. Dairy, coffee, biogas, microalgae, aflatoxins, algal blooms and migrating tuna: seemingly different, mathematically similar.

details
Where we start · desert & Gulf dairy

Heat-stress dairy

Where we start · desert & Gulf dairy

Heat-stress dairy

In hot, humid climates a confined herd sits above the heat-stress line much of the year. About half the milk loss is metabolic, and shows on no gauge. Forecast it pen by pen, then optimise cooling under a hard water budget where every litre is desalinated.

Method: adaptive probabilistic forecasting, constraint optimisation under a water budget.
~30% daily milk lost, summer vs winter
details
Coffee · post-harvest processing

Ferment & drying

Coffee · post-harvest processing

Ferment & drying

Cup grade is made or lost in fermentation and drying, and stays invisible until the coffee is cupped. A controllable, multivariable process with high-value buyers who carry the downgrade.

Method: multivariable process forecasting, control tuned to the target cup.
High-value specialty segment · controllable, multivariable
details
Biogas · anaerobic digestion

Reactor stability

Biogas · anaerobic digestion

Reactor stability

An invisible slide - volatile fatty acids, then pH, then a methane crash - tips a digester before the gas flow shows it, and recovery is slow and costly. A multivariable control problem with a consolidated operator base.

Method: precursor-state forecasting, constraint-based feed and dosing control.
Thousands of digesters · consolidated operators
details
Microalgae · ponds & photobioreactors

Culture crash

Microalgae · ponds & photobioreactors

Culture crash

A culture can crash without warning, and one event wipes a batch. An emerging, high-value production system where incumbents are immature - a textbook forecast-and-optimise fit.

Method: spatio-temporal forecasting of culture-crash risk, control of the response.
Emerging, high-value cultures
details
Grain · pre-harvest aflatoxin

Toxin routing

Grain · pre-harvest aflatoxin

Toxin routing

The toxin is invisible in the field and catastrophic downstream. The field-level forecast already exists; the open ground is the decision layer - routing each load to the handler that limits the damage.

Method: field-level toxin-risk forecast feeding handler-routing optimisation.
Forecast exists - the decision doesn't
details
Water utilities · algal blooms

Bloom response

Water utilities · algal blooms

Bloom response

A bloom is forecastable, but the response is un-optimised - and a bad one can shut a drinking-water supply. Utilities already monitor; the open ground is the optimised operational call.

Method: bloom forecasting feeding treatment and operations optimisation.
Can shut a water supply
details
Fisheries · migrating tuna

The moving shoal

Fisheries · migrating tuna

The moving shoal

A tuna stock is an invisible state that will not hold still - a warming ocean is pushing it across old boundaries faster than quota and effort can follow. Forecast where the fish will be, then optimise the catch and how fishing days are allocated under access and quota limits, before the value swims past the fleet that is licensed for it.

Method: spatio-temporal stock-distribution forecasting, allocation optimisation under quota and access limits.
The stock moves - the rules don't
details
Also fits the shape

More living-systems losses

Also fits the shape

More living-systems losses

The same signature recurs across agriculture and aquaculture: facial eczema in NZ dairy and sheep, biosecurity incursions, frost in horticulture, sea-temperature mortality in fish pens. Not every one shows all six signs. Where the loss is large and the budget is real, the decision layer earns its place there too.

One engine, many living-systems risks

Own one of these risks?

Start a conversation →
The mission

The real fight is before you know it.

ruru exists for a specific kind of loss: where prevention costs a fraction of recovery, where most of the damage has no cure, and where by the time the loss is measurable it is already done.

Half the losses we work on never show on a gauge. So the data has to beat observation, the maths has to be right, and the recommended call has to be trusted enough that the right person acts on a threat they cannot see. Getting all three to hold is harder than any single forecast.

Our test

We don't chase every problem.
We screen for one shape.

These are the six signs we look for. Few problems show all six, and we don't need them to. The more a problem shows, the more a decision layer moves the number rather than just dressing up a forecast.

01
The sign lags the harm
A hidden state moves first: the heat load, the liver dose, the infection. The measured or visible signal only confirms it once the damage is underway, so a sensor or a forecast alone is already too late.
02
The threshold moves
The level where it bites shifts with humidity, crop stage, litter, contacts, the individual animal. It's a model, not a dial you can set.
03
The damage compounds
Delayed and cumulative, and often it can't be undone. A missed window doesn't stay small, it grows.
04
The optimal action changes
There are several ways to respond, and the optimal one depends on the conditions. Sometimes the optimal move is to do nothing.
05
The fix competes for the scarce resource
Cooling burns the water you're trying to save; zinc costs money and risks toxicity; crews and gear are finite. A real trade-off, not a free lever.
06
Someone clearly carries the loss
A risk owner with the mandate and the budget to act: an operator, a levy body, an agency, a processor. Foresight without an owner changes nothing.

The more of these a problem shows, the more forecasting plus optimisation moves the number. Where the fit is partial, we say so.

The method

Forecast. Optimise. Communicate.

Where you have a forecast, we add the decision and the action: a real-time loop that doesn't just report, but produces results.

01 Predict

See the risk early

Forecast the hidden state before any sign shows: probabilistic, adaptive, and calibrated even when the data is sparse, late or blurred to a district.

OUR EDGE
02 Optimise

The best next move

Know your optimal action under the constraints that matter: the action, when, where, how much, or none at all. Re-optimised as the event unfolds.

03 Communicate

Get the call actioned

Turn the best next move into communication that engages the person who needs to act, so even a sceptical operator acts on a threat they can't yet see.

Forecast and optimisation are not a one-way pipeline. We loop between them, each improving the other, so the forecast is built to drive the decision. One shared engine across the losses we take on: heat-stressed dairy first, then coffee, biogas, microalgae, aflatoxins and algal blooms.

Already have the forecast?

See how we prove the decision →
The team

Built for hard decision problems.

Our scientists sit behind the forecasting, optimisation and uncertainty methods that others build on, with a combined 50,000+ citations on Google Scholar. We bring in further specialists as a problem demands, led by a CEO who turns advanced science into tools teams trust and use.

Dr Christoph Bergmeir
Core · Forecasting
Dr Christoph Bergmeir
Probabilistic & adaptive forecasting. Co-author of NeuralProphet; Clarivate top-1% forecasting papers.
Google Scholar ↗
Prof. Peter Stuckey
Core · Optimisation
Prof. Peter Stuckey
Constraint programming & optimisation. Creator of MiniZinc. AAAI Fellow · Google Eureka Prize.
Google Scholar ↗
Prof. Wray Buntine
Core · Uncertainty
Prof. Wray Buntine
Bayesian machine learning & uncertainty. Ex-NASA Ames, Berkeley, Google.
Google Scholar ↗
Dr Abishek Sriramulu
Core · Contextual AI
Dr Abishek Sriramulu
Graph, spatio-temporal & multimodal AI; adaptive dependency-learning GNNs.
Google Scholar ↗
Dr Frits de Nijs
Core · Optimisation
Dr Frits de Nijs
Risk-aware multi-agent & stochastic optimisation, reinforcement learning. 3rd, NeurIPS 2021 ML4CO.
Google Scholar ↗
Paul Shale
CEO · Product & Strategy
Paul Shale
Turns advanced science into tools teams trust and use. Law, finance, Harvard (disruption); startups across NZ, AU and the USA.
LinkedIn ↗Email

Domain experts set the constraints; the team builds the maths. They prove each method in the open literature first.

50,000+
citations across the team's work (Google Scholar)
Top 1%
Clarivate highly-cited in forecasting
1,000+
peer-reviewed publications, combined
The published methods behind ruru

ruru is not a new idea looking for proof. Each part of the loop is a field our scientists have led in the open literature - now aimed at invisible losses in living systems.

Probabilistic & adaptive forecasting
Forecast the hidden state before any sign shows - calibrated even when the data is sparse, late, or blurred to a district.
Graph & spatio-temporal AI
Track the threshold as it moves with humidity, crop stage, contacts and the individual animal - a model, not a dial.
Constraint programming & optimisation
Choose the optimal action under hard water, power and crew limits - including the discipline to do nothing.
Risk-aware & stochastic optimisation
Decide well under uncertainty, tuned to the best outcome rather than the lowest forecast error.
Bayesian ML & uncertainty
Put a trustworthy number on a risk no one can see yet, so the call can be defended.
Contextual & human-in-the-loop AI
Turn the call into evidence, confidence and an audit trail, so even a sceptical operator acts.

See the research & selected publications →

Want the science on your problem?

Talk to the team →
The research

We prove it before we promise it.

The optimisation-under-uncertainty at ruru's core is published work, not a pitch. Our team co-authored the Predict+Optimize benchmark for renewable-energy scheduling: the same forecast-then-optimise shape ruru applies to agricultural risk, which is why the method carries across problems.

Peer-reviewed · verified
Predict+Optimize Problem in Renewable Energy Scheduling ↗
IEEE Access, vol. 13, pp. 60064-60087, 2025
DOI 10.1109/ACCESS.2025.3555393
Four of ruru's core scientists are among the authors. A competition benchmark in energy scheduling, not agriculture.

Prove, not promise.

Before we ask anyone to build on faith, we replay their own history with our optimiser in the loop, same budget, same teams, a different allocation, and show, decision by decision, what it would have saved against what actually happened.

If the number holds, there's a position worth funding. If it doesn't, we recalibrate before we build. We never put a saved-loss figure on a slide that real data hasn't earned.

Selected publications · the methods behind ruru
  • Predict+Optimize Problem in Renewable Energy SchedulingBergmeir, de Nijs, Genov, Sriramulu, Abolghasemi, Bean et al · IEEE Access, 2025 · DOI ↗
  • Local and global trend Bayesian exponential smoothing modelsSmyl, Bergmeir, Dokumentov, Long, Wibowo, Schmidt · International Journal of Forecasting, 2025
  • MSTL: seasonal-trend decomposition for time series with multiple seasonal patternsBandara, Hyndman, Bergmeir · International Journal of Operational Research, 2025
  • Online guidance graph optimization for lifelong multi-agent path findingZang, Zhang, Harabor, Stuckey, Li · AAAI, 2025

Each scientist's full publication record is on their Google Scholar, linked in the team section above.

Read the Predict+Optimize paper →
Who we work with

If you carry the loss, we should talk.

ruru is built for the organisations that own a high-stakes, invisible risk, and have the mandate to act on it.

National-scale & desert dairy Coffee mills & exporters Biogas & digester operators Algae & bioreactor producers Water utilities Grain handlers & processors Levy bodies & co-ops Insurers