Turning Sheep EID and Weigh Crate Data Into Decisions

Most sheep farmers with a stick reader and a weigh crate already hold the records that would tell them which lambs are falling behind, which field is carrying a worm burden and which ewes are costing them money. The numbers are there, sitting in the weigh head or a phone app as a list of tag numbers and weights. What’s missing is the hour it takes to turn three or four weighings into gain figures and look for patterns. An AI chat assistant can do that hour’s work in about two minutes, using an export you can already produce.

This post sticks to one approach: take the CSV from the sheep EID weighing software you already have, give it to a general AI tool, and act on three things it can find. Those are lambs that aren’t growing, groups whose growth suddenly drops, and ewes whose lambs never reach the trailer on time.

What you probably already own

Before buying anything, check what your kit can export. Most UK setups fall into one of these:

All of them can give you a file with at least three columns: EID, date and weight. That’s enough. If you also record a management group or field (most indicators let you set a “session” or “group” name before you start), you get a lot more out of it.

The AI side costs very little. ChatGPT Plus is around £20 a month, Claude Pro is around £18, and Microsoft Copilot is included with some Microsoft 365 business plans. The free tiers can handle a few hundred rows once a month, though they cap how many files you can upload. Any of them works. Pick whichever you’ll actually open.

Getting the data into shape

Export each weighing session as a CSV, then put them into a single spreadsheet with one row per weighing. It should look something like this:

EID,Date,Weight_kg,Group,Dam_EID
826 012345678901,2026-05-12,17.2,Twins_F7,826 012345600112
826 012345678901,2026-06-02,22.9,Twins_F7,826 012345600112
826 012345678901,2026-06-23,24.1,Twins_F7,826 012345600112
826 012345678918,2026-05-12,19.0,Twins_F4,826 012345600147
826 012345678918,2026-06-02,24.8,Twins_F4,826 012345600147
826 012345678918,2026-06-23,30.2,Twins_F4,826 012345600147

You’ll only have the Dam_EID column if you link lambs to ewes at birth or at marking. Shearwell’s lambing tags and the “mothering up” functions in the Datamars and Gallagher apps both do this. Without it, you can still do the first two jobs below. The third needs it.

Clean the file before it goes anywhere. Delete double reads (two weights for one tag on the same day, usually from a lamb that backed out of the crate), and check the crate was zeroed. A crate that drifted 1.5 kg between sessions will make an entire mob look like it stalled. Weigh at a similar time of day each session, ideally after the lambs have stood off feed for a couple of hours, or gut fill will add noise of 1 to 2 kg per lamb.

Job one: pick out the lambs that aren’t growing

Upload the file and ask something plain:

This is lamb weight data from EID weighings. For each lamb, calculate daily liveweight gain (DLWG) between each pair of weighings and overall. List any lamb whose overall DLWG is under 180 g/day, sorted worst first, with its group. Show the formula you used.

The last line is important. These tools are generally good at arithmetic on a clean table, but asking for the formula lets you check one lamb by hand. Take the first lamb above: 24.1 minus 17.2 is 6.9 kg over 42 days, which is 164 g/day. If the AI says 164, you can trust the rest of the column. If it says something else, the dates have probably been read as text, and you can tell it so.

On a 600-lamb crop, the output tends to look like this:

Lambs with overall DLWG < 180 g/day (12 May – 23 June, 42 days)

EID               Group      Start  End   DLWG g/d  Last-period g/d
826 012345678955  Twins_F7   15.8   20.9  121       57
826 012345678940  Twins_F7   16.4   22.1  136       62
826 012345678901  Twins_F7   17.2   24.1  164       57
826 012345679210  Single_F2  21.0   28.0  167       190
...
41 lambs flagged: 33 in Twins_F7, 6 in Twins_F4, 2 in Single_F2

The bottom line of that output is the useful part. One or two slow lambs in a group usually means an individual problem: a poor milker for a dam, joint ill, an orf-affected mouth. Those go in the pen for a closer look next time you’re at the handling system. When 33 of the 41 flagged lambs are in one field, the problem lies with the field.

Job two: catch a worm problem before it shows

Look at the “last period” column for Twins_F7. Those lambs were growing at a respectable rate in the first three weeks and then fell to roughly a third of it. Ask the AI to compare groups by period:

Give me the average DLWG per group for each weighing interval, and flag any group where gain fell by more than 30% from one interval to the next.

Group       12 May–2 Jun   2 Jun–23 Jun   Change
Twins_F4    271 g/d        257 g/d        -5%
Twins_F7    268 g/d        154 g/d        -43%
Single_F2   305 g/d        296 g/d        -3%

Grass growth in late June on neighbouring fields doesn’t account for a 43% drop in one mob while the others hold steady. Nematodirus usually hits earlier than this, but in June a sharp fall in growth with no scouring yet strongly suggests a building trichostrongyle burden (Teladorsagia and Trichostrongylus). It’s exactly what SCOPS guidance tells you to watch for.

The data doesn’t replace a faecal egg count, but it tells you where to take one. A composite FEC from 10 fresh samples in Field 7 costs around £25 to £40 through most vets or a postal lab. If it comes back at, say, 850 eggs per gram, you’ve confirmed the problem before any lamb has scoured or died.

You also get a payoff on wormer costs. Rather than drenching all 180 twins, use targeted selective treatment. Moredun’s Happy Factor approach treats only the lambs that are below their expected gain and leaves good growers untreated, which keeps a refugia of susceptible worms on the pasture. Ask the AI:

For Twins_F7 only, list lambs whose DLWG in the last interval was below 200 g/day. Output just the EIDs as a list I can load as a draft file.

In this example, about 110 of the 180 lambs come back. At roughly 60p a dose for a 30 kg lamb on a white or yellow drench, you save around £40 on that mob in one round. More importantly, you aren’t selecting for resistance in the 70 lambs that were coping. If you have an auto-drafter, load the EID list as a draft file and the crate sorts them for you. With a stick reader alone, set the list as an alert file on the SDL440S or XRS2i and it’ll beep at each lamb that needs a dose.

Job three: find the ewes to cull

This is where linking dam to lamb pays off. Once lambs have had a weaning weight or an 8-week weight, ask:

Using Dam_EID, total the 8-week weight of all lambs for each ewe. Also give me each ewe’s own body weight from the October weighing. Calculate kg of lamb at 8 weeks per kg of ewe, and list the bottom 10%.

Kilograms of lamb per kilogram of ewe is a fairer measure than lambs reared, because a 90 kg ewe rearing one 24 kg single isn’t the same as a 65 kg ewe rearing twins at 22 kg each.

Ewe 00203 is the type that slips through a visual cull: big, sound in the mouth, in good condition at tupping. That condition comes from not feeding her lamb. If she shows up in the bottom 10% two years running, with no recorded mastitis or lameness to explain it, she should go. Ewe 00112 might be a marginal case. Ask the AI to cross-reference her lambs against Job two, because if both of them were in Field 7 the shortfall probably came from worms rather than milk.

A quick way to sense-check the list is to have the AI print the bottom 20 with each ewe’s age, if your software records it, and her result from last year. One poor year for a gimmer rearing her first set of twins is very different from a five-year-old ewe that’s been in the bottom 10% three times.

Where this goes wrong

Most failures come from the data, not the AI. Missed weighings leave gaps that skew gains, because a lamb weighed twice, 60 days apart, can hide a three-week stall in the middle. Lambs that lose a tag and get a replacement show up as two half-records. Ask the AI to list any EID with only one weighing, then reconcile those against your replacement tag register before you trust the averages.

Then there’s the question of what you’re uploading. A file of tag numbers and weights isn’t personally sensitive, but your CPH number and holding name don’t need to be in it, so strip them out. If you’re on a paid plan, turn off the setting that lets the provider use your chats for training.

Finally, keep an eye on what the tool is actually calculating. If a gain figure looks implausible (a lamb gaining 600 g/day, or losing weight across a whole group), ask it to show the rows for that animal. Usually it turns out to be a typo in the weight or a date in the wrong format.

For the wider picture of how weight data fits alongside lameness scoring, body condition and the newer sensor-based tools, our guide to AI livestock health monitoring covers what’s worth adding once this routine is working.

To make a start this week, export last year’s weaning weights and this spring’s marking weights, put them in one sheet with the tag number in column A, and ask for the bottom 10% by gain. You’ll know within ten minutes whether the lambs you’d already been worried about are the ones on the list, and whether there are others you’d missed.