Early lameness detection · grazing dairy herds
Cow 214 will be visibly lame by Friday.
Draft her for the trimmer today, while it’s still a £30 problem instead of a £300 one.
A leg tag that reads gait, a model that finds the asymmetry days before the limp, and a draft list waiting in the parlour each morning. Built for pasture-based herds — where the camera systems can’t see.
Draft list
Tuesday 5 August
| Cow | Asymmetry | Limb | Conf. | Early | Do |
|---|---|---|---|---|---|
| 214 | rising | Left hind | 88% | 3 days | Draft |
| 087 | rising | Right hind | 81% | 2 days | Draft |
| 341 | rising | Left fore | 64% | 5 days | Watch |
| 156Post-trim d9 | settling | Left hind | — | — | Logged |
2nd
Costliest disease in dairy, after mastitis
£180–330
Commonly estimated cost of a single case, all-in
0–3
The mobility scale your vet already scores to
£1.50–2
Per cow per month, for the software
The signal
The limp is the last symptom.
A mobility score is a snapshot of what a person can see from the collecting yard. Stride timing, weight-bearing symmetry between steps and lying-bout structure all move first. It is the same gait-asymmetry analysis used on human athletes, run on a much slower one.
What moves first
Stride-timing irregularity · weight-bearing asymmetry between steps · increased lying time · shorter, more frequent lying bouts
What being late costs
Caught three days early, she is a foot trim. Found ten days late, she is milk loss, a missed service and a culling decision.
Read the numbers as a table
| Day | -12 | -11 | -10 | -9 | -8 | -7 | -6 | -5 | -4 | -3 | -2 | -1 | 0 | 1 | 2 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Asymmetry | 9 | 11 | 8 | 12 | 10 | 14 | 17 | 23 | 31 | 42 | 51 | 63 | 72 | 79 | 84 |
| Score | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 0 | 1 | 2 | 2 | 2 |
Illustrative. This is the shape we expect, drawn to explain the thesis — not measured data. Proving this curve on real herds is precisely what the first pilots are for.
How it works
Four things, in a line.
- 1
Tag
A six-axis IMU — accelerometer and gyroscope — on the pastern or in the ear. Ten to twenty pounds of bill of materials, running off a coin cell or solar.
- 2
Gateway
One LoRaWAN gateway on the barn covers the whole grazing platform. Farms have terrible connectivity; the radio choice is designed around that rather than in spite of it.
- 3
Model
Gait asymmetry from accelerometer time-series, plus behavioural signals. It runs as a service, so the tag stays dumb, cheap and long-lived.
- 4
List
A daily draft list with confidence scores, and treatment logging that writes straight back to the mobility record.
The tag
Cheap enough to put on every cow.
The economics only work if the hardware is close to disposable. Everything below is a target rather than a shipping spec — this is the part that has to be built, and the part that has been built before.
- Sensor
- 6-axis IMU — accelerometer + gyroscope
- Mount
- Pastern strap, or ear tag
- BOM target
- £10–20
- Power
- Coin cell or solar
- Radio
- LoRaWAN, sub-GHz
- Infrastructure
- One gateway per farm
- Coverage
- Whole grazing platform
- Commercial
- Hardware at cost, or leased
One gateway, sited once, reaches the whole platform. That is the reason this works on grazing herds and camera-based systems don’t: the cows spend most of the year somewhere a camera in the parlour race will never look.
Ground truth
The labels already exist.
The hard part of any health model is labelled outcomes. Here they are already being produced: vets and technicians mobility-score to a standard 0–3 scale as routine practice, and hoof trimmers keep lesion records — sole ulcer, white line disease, digital dermatitis. Collecting training data slots into work the farm already does, rather than asking anyone to do something new.
- 0
Good
Walks evenly, flat back, weight even on all four.
- 1
Imperfect
Uneven steps or a shortened stride. No one leg obviously affected.
- 2
Impaired
Clearly shortened stride on one limb. This is where most people first notice.
- 3
Severely impaired
Will not bear weight. Urgent.
Score 2 is where a person first notices. The model’s whole job is everything to the left of it.
And it doubles as the paperwork
Several UK retailer milk contracts now require documented mobility scoring.
Treatment logging is in the tool anyway, because you need it to close the loop on the model. Once it’s there, the compliance evidence is a by-product rather than a separate job — which is often the argument that gets a tool through the farm office door.
What a trimmer’s record already holds
- Sole ulcer
- White line disease
- Digital dermatitis
- Heel horn erosion
- Foot, date, and what was done
Every line of that is an outcome label with a date against it — produced by people who were never trying to train a model.
The landscape
Why lameness, and not the obvious thing.
Heat detection
Solved, and crowded.
Oestrus detection is what pays for the incumbent collars — Allflex, smaXtec, CowManager. Building another one means walking into a pricing fight over a job that is already done well enough.
Lameness
The acknowledged unsolved one.
Typically the second-costliest disease in dairy after mastitis, with a meaningful share of any UK herd affected at any moment, and each case commonly estimated at £180–330 once milk loss, fertility damage and culling are counted.
Prior art, in full
This has been tried. That’s the point.
IceRobotics, in Edinburgh, built a leg-mounted accelerometer — CowAlert — and was acquired by MSD Animal Health. So the thesis here is validated rather than novel, and anyone reading this page should know that before they read anything else on it.
What is actually different
- A decade of ML progress on time-series since that hardware was designed
- A far cheaper tag
- Pasture-based herds, where camera systems like CattleEye can't see
- Lameness first, rather than lameness bolted onto an oestrus collar
The model
£1.50–2 per cow, per month.
Hardware at cost or leased; the software carries the margin. A 200-cow herd comes to roughly £4–5k a year — deliberately priced well under what lameness is already costing the same herd.
Move the sliders to your own numbers. The case rate is the one you know better than we do.
What the subscription covers
- The morning draft list, with a confidence score on every cow
- Trim and treatment logging, written back against the animal
- Mobility-score evidence you can hand to a milk buyer
- Gateway and tags at cost, or leased with the software
Set it against your own herd
What you’d pay, next to what it costs you now.
200 cows × £1.50–2.00 a month. Hardware at cost or leased, on top.
50 cases a year × £180–330, counting milk loss, fertility damage and culling.
At this herd size and case rate, the lameness bill is 1.9–4.6× what the software costs.
Case cost range from published UK estimates · case rate is yours to set
The market
Top-down, with the usual analyst chaos.
A 3–5× spread on the same market, depending who you ask. Everyone agrees on sustained double-digit growth, and everyone puts software and services as the dominant segment — which is where this sits. Treat the level as noise and the direction as signal.
Grand View Research
$5.73bn → $14.82bn
12.8% CAGR
Market Research Future
$9.31bn → $24.78bn
Widest of the three
Verified Market Research
$1.70bn → $3.60bn
Counts a narrower market
Bottoms-up, for this specific product
Our arithmetic on top of published herd numbers. Treat it as rough — the point is the shape of the ladder, not the third significant figure.
£30–40m
a year
UK dairy, lameness layer only
About 1.6 million dairy cows across roughly 7,500 herds, at around £20 a cow a year. That is full penetration — a ceiling, not a forecast.
~£700m
a year
The developed dairy world
EU about 20m cows, US about 9m, New Zealand about 5m, Australia about 1.5m. Call it 37m at the same per-cow price.
several £bn
a year
If it grows into a platform
Halter charges $5–8 a cow a month for a broader job. Same herd base, different product — which is why the analyst top-downs land where they do. The numbers cross-check.
£15–30m a year — UK sheep
Around 14 million breeding ewes, at £1–2 a head a year. The ceiling is too low to attract commercial attention.
Sequencing
Dairy, then sheep, then somebody else’s hardware.
First
Dairy
Per-animal value supports hardware. A milking cow earns enough to justify a tag on her leg, which is what makes the unit economics survive.
Second
Sheep, once the tag gets cheaper
Footrot is a huge and essentially unserved UK problem. But a ewe is worth about £100, so only an ultra-cheap tag ever works here. It follows dairy; it can't lead.
Or instead
License the algorithm
The pure-software fallback: put the lameness model onto existing ear-tag makers' hardware and skip manufacturing entirely. Smaller margin, far smaller risk.
The moat
Every trimmed foot is a label.
A proprietary dataset of movement signatures tied to vet-confirmed outcomes. It compounds — the longer it runs, the further ahead of a standing start it gets, and none of it can be bought in.
Get involved
Three ways in.
What’s useful right now is conversation — with people who milk cows, with people who trim their feet, and with people who fund the gap between the two.
Dairy farmers
Fifteen minutes on the phone. No demo, no deck — we want to hear what lameness costs you.
Vets & hoof trimmers
You hold the labels. Mobility scores and lesion records are what turn sensor data into a model.
Grant bodies & investors
Early, honest, and specced in public. The prior art section is the one to read first.