Most berry growers are running their harvest planning on a coin flip. Not because they’re not trying — their teams work hard, their agronomists are skilled, and they’ve been counting fruit by hand for years. The problem is the method. Manual counting, even done well, produces ~50% forecast accuracy. That’s the industry norm. And when your marketing company needs weekly volume commitments to pass on to the supermarket, 50% is a serious commercial problem.

This post looks at why accuracy stalls at 50%, what actually drives better forecasting, and how the gap is being closed.

Why manual counting hits a ceiling

Manual yield estimation typically involves walking fields, counting fruit on sample plants, and extrapolating to the whole block. The method is labour-intensive, slow, and limited by three things:

1. Sample size. Counting every plant on a large commercial farm is impossible, so teams count a small sample. The smaller the sample, the larger the margin of error.

2. Human variability. Different people count differently. One person calls an immature berry mature; another doesn’t. Counts taken at different times of day, in different light, by different people produce different numbers.

3. No dynamic adjustment. Manual counts give you a number at a moment in time. They don’t account for how fast the crop will ripen from that point, how weather events over the next six weeks will affect harvest timing, or how your specific variety ripens compared to published averages.

Add these three limitations together and 50% accuracy starts to look optimistic.

What actually drives forecast accuracy

Good yield forecasting is not just about counting more accurately. It’s about modelling ripening dynamics correctly. The factors that matter are:

• Phenological stage counts. Knowing how many flowers, immature berries, white fruit, and mature berries are on the plant at any given time — not just a total fruit count.
• Variety-specific ripening curves. Every berry variety ripens differently. Generic averages are next to useless if you’re growing a specialist variety that behaves differently from the published norm.
• Growing degree days (GDD). Temperature accumulation drives ripening faster than calendar days. A warm week can accelerate your harvest by five to seven days — which most static forecasts won’t predict.
• Farm- and block-level data. Microclimates within a single farm produce meaningfully different ripening rates. Block-level data captures this; farm-level averages mask it.

How GreenView AI addresses each of these

GreenView AI uses two waves of AI analysis on video footage captured by a GoPro camera mounted on farm machinery. The first wave counts fruit at each phenological stage — flowers, immature fruit, white fruit, mature fruit — faster and more consistently than any human team. The second wave applies the appropriate ripening curve for your specific variety and location, then models it forward using growing degree days and current weather data.

The result: forecasts accurate to 90% or better, updated weekly as new footage is uploaded.

The system has 300+ ripening curves built from 35 growing seasons of global data. Every farm, every block, every microclimate gets its own personalised model. It’s not using a generic average — it’s forecasting your crop, on your land.

What 90% accuracy is actually worth

For a grower with commercial marketing obligations, the difference between 50% and 90% forecast accuracy is not an abstract number. It’s the difference between:

• Fulfilling your volume commitments to your marketer — or facing a rejection event
• Scheduling the right number of picking crew — or paying for idle labour or scrambling at peak harvest
• Committing to cold chain logistics at the right time — or losing fruit to delays
• Protecting your commercial relationships — or eroding them one bad season at a time

One Bitwise Agronomy customer described 50 rejection events per year costing them 55,000 kg of fruit — a loss of over $660,000, before picking and logistics costs.

Getting your forecast right is one of the highest-leverage improvements any large berry operation can make. The tools to do it exist. The question is whether you’re using them.