Trapped in a lift, undergraduate students Harry Lucas and Liam Ellul immediately struck up a conversation about “the future and how it was going to happen”.

The conversation ended with the makings of a business idea and ultimately saw the creation of start-up Seer Insights.

“We were originally looking at hyperspectral imaging for broadacre farming and drawing inferences from that kind of data,” Ellul said.

This type of aerial imaging typically uses drones to overfly paddocks or crops and detect – for example – their health.

The plan was to do this for vineyards but after talking to members of the industry, Ellul and Lucas weren’t confident there was a market for it.

“We were looking at drones as a data collection system,” Lucas said.

It turned out drones aren’t much use for data collection in vineyards because it can be hard to spot grapes under the canopy of the vines.

But it also turned out that collecting data wasn’t a burning issue among viticulturists.

“We went into one viticulturist’s office and he pretty much said, ‘We have so much data. We don't need more data - we need to know what our yield is’,” Lucas said.

Until now, yields have mostly been calculated by hand. Grape growers will count bunches in various parts of the vineyard and then mathematically extract an estimation of the expected yield of the entire crop.

The estimate is used to calculate production and transport requirements, but can be out by up to 30 percent – which has significant financial implications for the supply chain, Lucas said.

The industry says inaccurate yield estimates cost it about $200 million a year.

Seer Insights’ answer to the problem is to digitise the data collection – growers record bunch counts in a mobile app rather than on paper – and then to apply machine learning algorithms to the data to more accurately work out the expected yield.

The company is testing its GrapeBrain system this growing season with various wineries and contract grape growers in South Australia.

They hope these tests will prove the system’s efficacy and provide valuable feedback on any tweaks needed before the system can be made commercially available.

The system itself is built on top of a Microsoft stack which the co-founders have stitched together.

The architecture decision in some ways isn’t surprising – they recently won an entrepreneurial competition at the University of Adelaide that was sponsored by Microsoft.

The prize included time at Microsoft’s Innovation Centre in Adelaide and they receive ongoing guidance and support from the technology giant.

With proof of concept testing underway, Seer Insights has only recently begun looking for funding to scale up the business.

“It initially started off for the love [of it],” Ellul said.

“We were operating through necessity. We tried to keep it as lean as possible.”

That included not asking the wine industry for any cash – a strategy Ellul believes has worked well for Seer’s development.

“Because we’re new to the business, when I was going into a large multinational [winery] and asking them for a bunch of their data, I was treading very lightly around asking them for petrol money,” Ellul told Information Age.

“It ended up that not expecting money from the wine organisations was a good thing.

“We've been getting a lot more support internally from these companies now because we’ve displayed that we're serious about this and we're doing it because we think we can get something happening.”

GrapeBrain is not the only initiative that is trying to solve the problem of yield estimation.

UNSW researchers are driving cars and robots mounted with cameras along the rows of grapes in vineyards in a bid to more accurately estimate yields.

They hope to be able to provide accurate estimates even before the vines begin to bud; Seer Insights, by contrast, is predicting estimates up to three months before harvest.

“Ideally wineries would be loving it if you could predict it a year in advance but that gets quite complicated,” Lucas said.