Ask a flotation crew why the collector setpoint sits where it does and the answer usually comes back to the bad days. Hard ore shows up, recovery dips, and somebody has to explain it at the morning meeting. So the dose gets set for the toughest feed the plant expects to see, and it stays there.
Most of the time that feed isn't in the circuit. For hours on end the ore is easier than the setpoint assumes, and the extra reagent goes straight into cost. Reagent optimization is about closing that gap without giving up recovery.
Feed changes faster than people can react to it. Grade and hardness move between shifts, sometimes within one, as trucks pull from different benches or reclaim from a stockpile built in layers. By the time this morning's assays reach a metallurgist, the ore they describe has already gone through the cells.
Padding the setpoint is a sensible response to that delay. Extra collector feels like cheap insurance, until a year of it turns into one of the bigger lines in the operating budget and nobody can say how much was needed.
The idea is easy to state. Match the dose to the ore that's in the circuit right now and keep adjusting as it shifts. Doing that by hand on a busy shift is close to impossible.
A model trained on the plant's own historian data can carry that load. It reads feed indicators and circuit response, forecasts what a change in ore will do to recovery, and recommends the lowest dose that holds it. When the feed turns difficult the dose goes up. When it eases off the dose comes down, and that second half is where the savings live. Recommendations sit on top of the existing control system, so operators stay in charge.
A gold operation producing around 100,000 ounces a year had been adding collector at a fixed rate. NTWIST layered real-time feed insight and predictive dosing over the site's existing control. Reagent costs dropped by up to 30%, and the site saw a gold recovery improvement of 1.6% or more.
A system like this doesn't need to beat the crew every hour to pay for itself. It needs to catch the stretches where a fixed setpoint is wrong, and those stretches add up. At a base-metals refiner where NTWIST applied the same approach, the model increased potential profit during 42% of operating hours.
No new hardware went in. The work ran on the plant's existing instruments and data. Our Mine-to-Mill 2.0 blog covers the upstream side, where grade volatility starts. More deployment results like this one are on our case studies page.
Does reagent spend swing month to month while recovery stays flat? And is there a year or more of historian data on feed and reagent addition? If both answers lean yes, there's probably money sitting in the setpoint. It's often a quick win in mineral processing optimization because the data already exists.
That's the problem the MineMax suite was built around. DynaMax and MillMax work together to read feed variation in real time and recommend the dose that fits it, on top of the control a site already runs. Our guide to AI mining optimization explains that decision layer, and our blog on reducing emissions while increasing throughput covers the footprint side.
To put a number on your own circuit, try the Mining ROI Calculator or book a Clarity Call with NTWIST.
Reagent optimization means matching reagent dose to the ore in the circuit right now, instead of holding a fixed dose set for the hardest expected feed. It cuts reagent use on easier ore and protects recovery when feed gets tough.
It depends on how much a plant overdoses today. At a gold operation producing about 100,000 ounces a year, NTWIST's predictive dosing cut reagent costs by up to 30% over fixed-rate dosing.
It doesn't have to. Predictive dosing raises the dose when feed gets difficult and lowers it when feed is easier. In one NTWIST deployment, gold recovery improved by 1.6% or more while reagent costs fell.
No. NTWIST's MineMax suite sits on top of a plant's existing control system and uses data the site already collects, so the work needs minimal infrastructure change.
Most plants already have enough in their process historian. A year or more of reagent addition rates plus feed and circuit data is usually a solid base.
Pramodh Chandrashekar is the Marketing Manager at NTWIST, where he leads content and communications for the company's mining and manufacturing work. He writes about how industrial AI is used in mine-to-mill operations and production planning, drawing on NTWIST's deployment results and its engineering and sales teams.
He is a three-time TEDx speaker and is known in India's green entrepreneurship space. A lawyer by training with a BBA LLB (Hons.) degree, he also wrote Bleeding Markets, a beginner-friendly book on finance.
LinkedIn: Pramodh Chandrashekar