Tiny AI system doubles sugarcane yields in Maharashtra

In Baramati, Pune district, sugarcane farmers are using AI-generated daily alerts to decide when to irrigate, where to fertilise and where to scout for pests. The project has produced striking numbers. It has also produced a set of caveats that rarely travel with them.

The initiative, called Farm of the Future, is run by the Agricultural Development Trust (ADT) Baramati with Microsoft and Oxford University. Microsoft partner Click2Cloud built the farmer-facing app, Agripilot.ai, which delivers alerts in English, Hindi and Marathi.

How it works

Weather stations on participating farms carry wind, rain, solar, temperature and humidity gauges above ground, with soil sensors measuring moisture, pH, electrical conductivity, potassium and nitrogen below.

That data is combined with satellite and drone imagery and historical crop records, then analysed to produce simple daily instructions: water more, spray fertiliser, scout for pests. A satellite map pinpoints exactly where.

The value is in that specificity. Conditions vary between neighbouring fields, so a fixed irrigation or fertiliser schedule applies the same treatment to plots that need different things.

What the numbers actually say

ADT unveiled the project at its January 2024 Krushik farmers' festival, covering about a dozen crops including sugarcane, tomato and okra. The headline sugarcane result shown there came from a demonstration plot at the Krishi Vigyan Kendra research station: stalks 30 to 40% heavier at harvest, 20% more sucrose, less water and fertiliser, and a crop cycle cut from 18 months to 12.

That was recruitment material, not a farmer-field result. Some 20,000 farmers signed up after seeing it. From those, 1,000 were selected, and an initial cohort of about 200 planted roughly one-acre test plots in mid-2024.

Reported outcomes since then vary widely by source. ADT and Krishi Vigyan Kendra put the yield gain at 40%, a figure Satya Nadella has repeated. A Microsoft report cites more than 20% higher production, a 25% cut in fertiliser costs through spot fertilisation, and an 8% reduction in water use. Other accounts put water savings at 30% or even 50%. Farmer Seema Chavan, whose video Nadella shared, says her costs fell about 30% and her yield rose 30 to 40% on 1.5 acres.

Nearly all of these figures come from Microsoft's or ADT's own communications. There is no independent or peer-reviewed evaluation in the public record.

The parts usually left out

The AI needed correcting. Microsoft's own account notes that ADT agronomists review every AI-generated alert before it reaches a farmer, and that in the first six months, 10 to 20% of recommendations were edited for accuracy. Researchers expected the system to need minimal human intervention only once the crop cycle closed.

Someone is paying for the hardware. Each trial farmer paid a one-off ₹10,000 for soil testing and training. ADT contributed roughly ₹75,000 per farmer in equipment and other costs. That is about ₹85,000 per acre of setup, heavily subsidised.

What happens next

The programme is expanding from 1,000 farmers towards 50,000 across Maharashtra. In August 2025, the Indian Sugar and Bio-energy Manufacturers Association signed an MoU with ADT Baramati and agri-tech firm Map My Crop to launch a National AI-ML Network, targeting cane productivity above 100 tonnes per hectare and sugar recovery above 11%. Click2Cloud is separately working with the Chhattisgarh and Uttar Pradesh governments.

The open question is not whether precision advice improves outcomes on a research plot. It is whether ₹85,000 of subsidised hardware per acre, plus an agronomist checking the machine's homework, holds together at fifty thousand farmers.