AI in the Yard and Office

How is AI changing dairy farming?

Dairy is where livestock technology is most mature, and it is not close. The combination of high-value animals, twice-daily milking routines, and relatively large herd sizes creates the conditions where automation and monitoring can deliver genuine returns. If you are running 200 cows on a conventional parlour, the maths on robotic milking is now well established. If you are running 60 cows on a hill farm, it is a different conversation entirely.

Robotic milking systems

The two dominant systems in the UK are the Lely Astronaut A5 and the DeLaval VMS V310. Both are voluntary milking systems: the cow walks in when she chooses, a robotic arm attaches the cluster, and the system milks her without human intervention. A single unit handles 60 to 70 cows, milking each two to three times per day based on individual yield and lactation stage.

Roughly 5 to 10 percent of UK dairy herds now use robotic milking. That figure is concentrated in larger operations, typically 200 cows and above, where the economics work. A single unit costs GBP 120,000 to 180,000 installed, and most farms need at least two. The payback period is 7 to 10 years, though farms regularly report that the real value is not just labour saving but the data the robots generate.

These are not dumb machines. Each milking event records individual cow yield, milk conductivity (an early indicator of mastitis), somatic cell count, fat and protein ratios, blood in milk, and milking speed. Over time, the system builds a profile of each animal. When a cow's milk conductivity spikes three days before she would show clinical mastitis symptoms, the system flags her. When yield drops against her expected curve, you know before she goes off her feed.

The quality of life argument is real. Dairy farming is one of the most demanding enterprises in agriculture. Two milkings a day, 365 days a year, no weekends. Robotic milking removes the 4am start and the 4pm finish. For family farms where the next generation is deciding whether to come home, that matters as much as the financial return.

Precision feeding

Alongside robotic milking, precision feeding systems like the Lely Vector and DeLaval OptiFeed automate diet mixing and delivery. The Vector, an autonomous feed pusher, mixes and distributes rations multiple times per day based on group requirements. This is more than convenience: consistent, frequent feeding improves rumen health, increases dry matter intake, and can lift yield by 5 to 10 percent versus twice-daily feeding from a mixer wagon.

Feed is typically 60 to 70 percent of the cost of milk production. Even a 3 to 5 percent improvement in feed conversion efficiency on a 200-cow herd translates to thousands of pounds per year. The challenge is the capital cost on top of an already significant investment in milking robots.

The farms that get the most from robotic milking are the ones that use the data, not just the automation. If you install a robot and never look at the reports, you have bought a very expensive milking parlour.

What can smart collars and ear tags actually detect?

The monitoring collar and ear tag market has matured significantly in the past five years. The leading system in UK dairy is Allflex SenseHub (owned by MSD Animal Health), which combines a neck collar or ear tag with a base station and cloud platform. The tags monitor rumination time, eating behaviour, activity levels, and rest patterns in real time.

Heat detection

This is where monitoring pays back fastest. Visual heat detection by a stockperson achieves 50 to 60 percent accuracy in most herds. SenseHub and similar systems claim 90 percent accuracy and above, detecting the characteristic increase in activity and decrease in rumination that signals oestrus. For a 200-cow dairy herd where each missed heat costs GBP 250 to 350 in extended calving interval, the maths is straightforward. The system costs GBP 80 to 120 per animal, and most farms recoup the investment within 18 months through improved conception rates alone.

Health alerts

Beyond heat detection, these systems flag health issues 1 to 3 days before clinical signs become visible. A drop in rumination time, a change in lying behaviour, reduced eating: these patterns precede visible illness. Early intervention means lower treatment costs, better recovery rates, and fewer culls.

Camera-based lameness detection

CattleEye, a Northern Ireland company acquired by MSD Animal Health, uses cameras mounted above cattle walkways to automatically score gait and detect lameness. The system analyses video footage of cows walking and assigns mobility scores without a stockperson needing to stand in the rain watching cows file past. It detects lameness 2 to 3 weeks earlier than typical human observation.

This matters because lameness is one of the most costly and prevalent welfare issues in UK dairy. Prevalence runs at 30 percent or higher in many herds. Each case costs GBP 180 to 300 when you account for reduced yield, treatment, extended calving interval, and premature culling. A 200-cow herd with 30 percent prevalence is losing GBP 10,000 to 18,000 per year to lameness alone. Catching cases two to three weeks earlier reduces severity and cost per case significantly.

Calving sensors

Moocall produces a tail-mounted sensor (GBP 250 plus subscription) that detects the rhythmic tail movement associated with the onset of calving and sends an alert to the farmer's phone. For spring-calving herds where cows are calving outdoors, or for farms with high-value pedigree animals, avoiding one dead calf pays for the sensor many times over. It is simple, targeted technology that solves a specific, expensive problem.

Beef and sheep

Breedr (UK-based, raised over GBP 10 million) is building a livestock management platform focused on beef and sheep. The concept is straightforward: digitise individual animal records, track weight gain against feed costs, benchmark performance, and eventually connect to markets. It is early, but it addresses a genuine gap. Most beef and sheep farmers manage their data in notebooks, spreadsheets, or not at all.

What about sheep, pigs, and poultry?

The honest assessment is that livestock AI is most mature in dairy, developing in beef, and least mature in sheep, pigs, and poultry. The economics explain why: dairy cows are high-value individual animals milked twice daily in controlled environments. Sheep are low-value animals spread across hills. The data density and return on investment per animal are fundamentally different.

Sheep: virtual fencing and EID

Nofence is the most interesting technology for sheep and cattle in extensive systems. It uses GPS collars (GBP 200 to 250 per collar, plus subscription) to create virtual fence boundaries. When an animal approaches the virtual boundary, the collar emits audio cues. If the animal continues, it receives a mild electric stimulus. DEFRA has approved its use in England.

The applications for conservation grazing, targeted mob grazing, and managing stock on land without physical fencing are significant. For hill farmers managing sheep across thousands of acres of common land, the ability to contain stock without building or maintaining miles of fencing changes the economics of grazing management. The limitation is cost: at GBP 200+ per collar, it is only viable for cattle or high-value breeding ewes, not a commercial flock of 500 mules.

Electronic Identification (EID) compliance is quietly creating the data infrastructure that future sheep AI will run on. Every sheep in the UK carries an EID tag by law. Readers are getting cheaper, software is improving, and the move from paper records to digital flock management is slowly happening. It is not AI yet, but it is the foundation.

Pigs: under severe pressure

SoundTalks, acquired by Boehringer Ingelheim, uses microphones mounted in pig houses to detect respiratory disease from cough patterns. The system analyses audio continuously and flags buildings where respiratory sounds increase, triggering earlier treatment and reducing spread. It is genuine applied AI, using sound pattern recognition in a way that a stockperson walking through a 1,000-pig unit physically cannot.

But the UK pig sector is in crisis. The national herd has contracted by roughly 40 percent since 2020, driven by a combination of low prices, high feed costs, labour shortages, and the impact of losing EU processing workers post-Brexit. Average pig farm business income sits around GBP 4,000. Investing in monitoring technology when the business itself is barely viable is not a realistic conversation for most pig producers.

Poultry: environmental control

ChickenBoy is an autonomous robot that moves through poultry houses monitoring temperature, humidity, ammonia levels, and bird behaviour. Poultry production is already highly controlled and data-driven compared to other livestock sectors, so the technology is building on an existing foundation of environmental management rather than starting from scratch.

The poultry sector's challenge is different from other livestock: the technology exists, but margins are so tight per bird that the investment case has to work at scale. A free-range egg producer running 16,000 birds can justify environmental monitoring that a small flock of 500 layers cannot.

The maturity spectrum

If you map livestock sectors by technology maturity, dairy sits at the top: robotic milking is commercially proven, monitoring systems have clear ROI, and precision feeding delivers measurable yield improvements. Beef cattle are next, with monitoring collars becoming viable for larger suckler herds and finishing units, though the returns per animal are lower than dairy. Poultry is highly automated at scale but the AI layer is thin. Pigs have some genuinely intelligent monitoring (SoundTalks), but the sector economics make adoption academic. Sheep are at the bottom: high animal numbers, low value per head, dispersed across challenging terrain, with almost no commercial AI solutions that make economic sense at typical flock sizes.

This is not a criticism of any sector. It is a reflection of the economics. Technology investment follows return on investment, and return on investment follows the value per animal, the frequency of human-animal interaction, and the controllability of the environment. A dairy cow worth GBP 1,500, milked three times a day in a shed with power and Wi-Fi, is a fundamentally different proposition from a hill ewe worth GBP 80, checked twice a week on a fellside with no mobile signal.

What farm management software are UK farms actually using?

This is where we need to be particularly honest. The gap between what agritech marketing promises and what is actually deployed on working UK farms is wide. Most farm management software today is digital record-keeping. It replaces paper, which is valuable, but it is not artificial intelligence in any meaningful sense.

The established platforms

Gatekeeper is the dominant arable farm management platform, used by over 15,000 UK farms. It handles crop planning, spray records, field operations, compliance reporting, and environmental stewardship mapping. It is solid, reliable software that does what it says. It is not AI.

Farmplan is the most widely used farm accounting package. It manages VAT returns, BPS claims, and financial reporting. Many farms also use it for livestock recording and herd management. It works well as an accounting tool. The "intelligence" is in the reporting, not in predictive analytics or machine learning.

Muddy Boots (now part of TELUS Agriculture) focuses on supply chain compliance and food safety, linking farm records to retailer requirements. Hectare runs the largest online livestock marketing platform in the UK, replacing the physical auction mart for some transactions. AgriWebb provides livestock management with a focus on individual animal recording and weight tracking.

AHDB Farmbench is a free benchmarking tool from the Agriculture and Horticulture Development Board that lets farmers compare their costs of production against anonymised industry data. It is genuinely useful and genuinely free, which makes it more widely used than many paid alternatives.

The honest assessment

Very little genuine AI is deployed in UK farm software today. When vendors say "AI-powered", they typically mean basic decision-support rules: "your nitrogen application rate for this field based on soil type and previous crop is X." That is useful, but it is a lookup table, not machine learning. It is the same calculation an agronomist does in their head, just automated.

The exceptions are narrow and specific: image recognition for crop disease identification (Plantix, Agrii's Omnia), satellite-derived biomass mapping (Yara N-Sensor, RHIZA), and the monitoring systems already discussed (CattleEye, SoundTalks). These use genuine machine learning. But they are point solutions, not integrated farm management platforms.

The frustration farmers express most frequently is fragmentation. The milking robot has its own software. The monitoring collars have their own app. The feed system, the accounts, the compliance records, the environmental data: all separate systems, separate logins, separate data silos. Nothing talks to anything else. A farmer might be logging into six or seven different platforms to manage one business. The integration layer, the thing that would actually make AI useful by connecting all this data, barely exists.

How is environmental compliance driving tech adoption?

The transition from the Basic Payment Scheme to Environmental Land Management schemes is the single biggest policy shift in UK agriculture in a generation. It changes the fundamental question from "how much land do you farm?" to "what environmental outcomes are you delivering?" That shift creates enormous demand for monitoring, measurement, and evidence tools.

The ELMS landscape

The Sustainable Farming Incentive (SFI) pays GBP 22 to 58 per hectare for specific actions: soil health assessments, herbal leys, hedgerow management, integrated pest management, nutrient management. Countryside Stewardship covers more targeted habitat creation and restoration. Landscape Recovery is the premium tier for large-scale environmental projects.

The critical change is that payment increasingly depends on evidence of outcomes, not just evidence of activity. It is no longer enough to say you planted a hedgerow. You need to demonstrate what it delivered in terms of biodiversity, carbon sequestration, or water quality improvement. That evidence requirement is where technology becomes essential.

Carbon and sustainability tools

The Cool Farm Tool, developed by the Cool Farm Alliance (a consortium including Tesco, Unilever, PepsiCo, and others), is used by over 12,000 farms globally. It calculates greenhouse gas emissions from farming activities, covering soil, livestock, energy, and inputs. It is free and increasingly required by major retailers as part of supply chain sustainability reporting.

The Farm Carbon Toolkit is a UK-developed calculator that maps carbon flows across the whole farm, including sequestration in soils, hedgerows, and woodland as well as emissions. Agricarbon (UK-based) combines soil carbon measurement with carbon credit verification, attempting to create a route for farmers to generate income from soil carbon sequestration. The carbon credit market for agriculture is still immature and the prices per tonne are low, but it is developing.

Sam (Sustainable Agriculture Metrics) is a sustainability platform used by several UK food supply chains. Soilmentor is a soil health monitoring app designed for farmers to track soil biological activity, infiltration rates, and organic matter over time. It makes soil assessment accessible without needing a laboratory for every test.

Biodiversity monitoring

BirdNET, developed by the Cornell Lab of Ornithology and Chemnitz University, uses AI to identify bird species from audio recordings. Point a cheap smartphone or dedicated recorder at a hedgerow or field margin for 24 hours and BirdNET will tell you which species are present and how active they are. For ELMS applications that require evidence of biodiversity outcomes, this is a practical, low-cost solution.

Agrimetrics, backed by Rothamsted Research, is building a data marketplace that aggregates agricultural and environmental data from multiple sources. The vision is that by combining soil data, weather data, satellite imagery, farm management records, and environmental monitoring into a single platform, the industry can move from fragmented records to integrated intelligence. It is early days, but the concept is sound.

The irony of ELMS is that it demands digital evidence from an industry where half the farms do not have reliable broadband. The policy runs ahead of the infrastructure.

Where is the labour gap and can AI help?

The UK lost an estimated 60,000 to 80,000 seasonal EU workers after Brexit. The Seasonal Worker Visa scheme, capped at roughly 45,000 places, does not fill the gap. Minimum wage sits at GBP 12.21 per hour and rising. For horticultural operations that rely on hand-picking, the labour equation has fundamentally changed.

Fieldwork Robotics (a University of Plymouth spinout) is developing robots for harvesting soft fruit: raspberries, strawberries, and other crops that currently require manual picking. The robots use computer vision to identify ripe fruit and robotic arms to pick without bruising. It works in trials. It is not yet deployed at commercial scale, and the speed per punnet is still slower than an experienced human picker.

Saga Robotics produces Thorvald, an autonomous platform that can be fitted with different tools. Its most deployed application in the UK is UV-C light treatment for strawberry powdery mildew, running through polytunnels at night to reduce fungicide use. It is not replacing a picker; it is replacing a sprayer and doing a better job.

FarmDroid from Denmark offers autonomous seeding and mechanical weeding for row crops like sugar beet and onions. Because it GPS-maps exactly where each seed is planted, it can return weeks later and mechanically weed between and within rows with millimetre precision, eliminating herbicide use. Garford (UK-based, Lincolnshire) makes camera-guided inter-row hoes that achieve similar precision at higher speed.

The key insight is that automation is most urgently needed in horticulture and soft fruit, but adoption is limited by cost and maturity. A single robotic strawberry picker that costs GBP 500,000 and picks slower than a human does not solve the problem yet. The economics have to close further. For livestock, the labour gap is different: it is about the relentless daily grind of milking, feeding, and checking stock. Robotic milking addresses the first. Precision feeding addresses the second. But checking stock in a field in February is hard to automate.

There is also a mental health dimension to the labour shortage that does not get discussed enough in technology conversations. Farming has some of the highest rates of suicide and depression of any occupation in the UK. The Farm Safety Foundation reports that a farmer dies by suicide roughly every week. Much of this is linked to isolation, relentless workload, financial pressure, and the feeling of being trapped. Technology that genuinely reduces workload, that means you do not have to be in the parlour at 4am every single day, that gives you a weekend off for the first time in years, has a value that goes beyond its financial return. The quality of life argument for automation is not soft. It is life and death for some farm families.

What stops farmers from adopting this technology?

It is tempting to frame technology adoption as a mindset problem: farmers are resistant to change, slow to adopt, suspicious of new things. That framing is lazy and wrong. The barriers are structural, financial, and infrastructural.

Connectivity

Only around 50 percent of rural premises have access to superfast broadband (30 Mbps+). 4G mobile coverage in agricultural areas is patchy at best, with significant not-spots across Wales, Scotland, and upland England. A monitoring system that requires a cloud connection to function is useless in a field with no signal. Project Gigabit and the Shared Rural Network are improving things, but slowly. Many farm buildings are steel-framed sheds that kill Wi-Fi signals. The practical reality of getting data from a cow's collar to a farmer's phone involves more infrastructure than the technology vendors acknowledge.

Cost versus income

This is the fundamental issue. The technologies described in this article range from GBP 250 (a Moocall sensor) to GBP 350,000+ (a two-unit robotic milking installation). The income available to pay for them varies just as dramatically.

A lowland grazing farm earning GBP 18,000 cannot invest GBP 150,000 in robots. A pig farm earning GBP 4,000 cannot invest in anything. The technology conversation in agriculture is really two separate conversations: one for the top-quartile farms with income and scale to adopt, and another for the majority who cannot. Most agritech marketing is aimed at the first group and irrelevant to the second.

Age and skills

The average age of a UK farmer is 59. This is not about stereotyping older farmers as technophobic. It is about the rational calculation that a 62-year-old farmer three years from retirement is unlikely to invest in a system with a 10-year payback. Succession uncertainty compounds this: if nobody is coming home to take on the farm, there is no reason to invest in its technological future.

Data fragmentation and trust

Every technology vendor builds their own walled garden. Lely data does not flow into Farmplan. SenseHub does not connect to Gatekeeper. The farmer ends up as the integration layer, manually transferring information between systems or simply ignoring the data because accessing it requires too many logins and too much time.

Data ownership is a legitimate concern. Farmers ask, reasonably, who owns the data their robots and collars generate. If it sits on a vendor's cloud, can it be shared with processors, retailers, or government without consent? Can it be used to price insurance, benchmark against competitors, or enforce compliance? These are not paranoid questions. They are the questions any business owner asks before handing operational data to a third party.

The NFU and AHDB have both published guidance on farm data principles, emphasising farmer ownership and control. But in practice, most vendor contracts give the technology provider broad rights to use anonymised and aggregated data. When a farmer's milking robot data feeds into a vendor's benchmarking platform, the farmer may benefit from the comparison, but the vendor benefits from the dataset. The power dynamic is not balanced, and farmers know it.

The missing middle

There is a significant gap in the market between six-figure robotic systems designed for large commercial operations and free DEFRA spreadsheets designed for compliance. The affordable, integrated, genuinely intelligent farm management platform for the farmer earning GBP 18,000 to 50,000 does not exist yet. That is where the real opportunity sits: not in adding another point solution, but in building the layer that connects everything a farm already has and makes it useful.

The technology is not the hard part. The hard part is making it work on a real farm, with real margins, real broadband, and a real person who has been up since five and does not have time to debug a Bluetooth connection in the rain.