Everyone's feed budget is wrong. That's not the problem.
A feed budget is a forecast, and forecasts made under uncertainty are wrong by construction. The useful question isn't how to make it right — it's how wrong it can be before your decision changes.
By Erin —farmBI1 September 2026
It's June. Ask ten sheep and beef farmers whether they run a feed budget. Two will start talking to you about getting their feed conversion efficiency down from 32kg dry matter per kilo of production to 25kg dry matter by getting a better winter rotation nailed. One of them will tell you about how they chucked it all into ChatGPT and it is all looking pretty good. Two of them will mutter about a spreadsheet they have been using. The other five will pretty happily tell you they don't use feed budgets.
Now go back to those five farmers with feed budgets at the end of August. Is their June-built feed budget still trustworthy?
Well… winter was a bit drier than we expected so we got better utilisation of the kale crop, which meant the R1s are a bit better than expected, and it lasted a bit longer. No feral storms through lambing, so good survival and a few more on the ground than we thought.
Is that feed budget a failure?
Nope. That is actually why it was built.
A feed budget is a forecast. Forecasts made under uncertainty (i.e. farming!) are wrong. Not occasionally, but inherently. The moment you think you can predict the weather is the moment you should stop farming and start selling a calendar with the next two years of rainfall mapped out.
The trouble starts when a budget is treated as a plan rather than a projection. A plan that turns out wrong feels like a personal failure, so it gets quietly abandoned and nobody builds another one until the banker prompts you at the next review.
The question that matters
Decision research has a useful reframe here, and it comes from outside agriculture, from forecasting and decision analysis. It goes something like this: don't ask what the answer is, ask what would have to be true for your decision to change.
You don't need a feed budget that's right. You need to know how wrong it can be before you would change what you do.
When you apply that to feed, it changes what you are looking for. You are not chasing the last decimal point of pasture growth for July. You are looking for the size of the gap between where you are, and the point at which you would change what you are doing — sell early, buy more, feed supplement, plant a forage crop.
The practical version
Run the budget. Then run it again using "plausible downside" pasture growth rates. Run it again using "plausible upside" rates. If both versions leave your plan intact, you are awesome and can relax and roll with the punches. If the alternative versions diverge, you've found the line in the sand you need to monitor against.
The TAB pays long odds on exact score predictions — the false precision trap
The forecast doesn't need to be exact. It needs to tell you whether the plausible range crosses a decision threshold.
A specific, definitive number carries a lot of weight. But how was it derived? When an outcome depends on several uncertain things all being true, its probability depends on their combined — and often conditional — probabilities.
The TAB will give you much shorter odds on the All Blacks winning than on them winning 31–17. Same game, same teams — but the exact score requires a whole stack of uncertain things to line up.
Farm forecasts are no different. You are told your lambs will average 42.3kg liveweight on 15 January and the schedule will be $8.05/kg CWT. Perfect, sign me up. But think about what has to be true for that number to land. Pasture growth has to be roughly as we expect it. Processor demand has to develop as expected. Lamb growth rates follow the assumed curve. The currency has to hold where we expect it. Putin needs to take a breather.
Imagine, just for illustration, that we thought there was a 75% chance of each of four important assumptions landing inside the range we used. The chance of all four lining up is not 75%. If they were independent, it would be:
0.75 × 0.75 × 0.75 × 0.75 = 32%
A one in three chance. Yet a model is presenting a definitive answer up to two decimal places. That is the trap.
The calculation can be precise while the answer is not. A single number carries no information about how much it should be trusted. It's the reason confidence — how much weight you should put on an answer, given what had to be assumed to produce it — should be visible rather than hidden. 42.3kg with high confidence and 42.3kg with low confidence are not the same forecast.
Two farms, one number
Or, bringing it back to the feed budget context — a closing August average pasture cover of 1450kg DM/ha for two farms, Farm A and Farm B. Farmer A is a committed soul who does a pasture measuring round each month, seeing most paddocks each time. Farmer B reckons the June opening cover was pretty close, based on where they have usually been in the past, so can't be too far out by the end of August. Those two farms produce exactly the same figure — one from regular monitoring, one from an educated guess at opening. It will look identical on the page while meaning entirely different things.
Same headline number, two different amounts of information. Range widths are illustrative.
Same headline. Different amounts of information. Farm A can act on that number now. Farm B should be asking what cheap information would narrow the band before committing to decisions — which might be as simple as getting some covers from across a proportion of the farm.
What this doesn't mean
It doesn't mean precision is pointless, or that measuring is a waste of time. Narrower bands — higher confidence — are better; they let you act earlier, because you really know how close that line in the sand is. But a wide band is not a reason to give up on the exercise. It is still information about where you are, and about what it would be worth knowing.
Field notes is written while building farmBI. If you'd argue with any of this, I'd like to hear it — that's most of what the pilot is for.
