Guardian Dogs, Herding Dogs and Alpacas

#Fifty-three per cent of the successes are dead

The Cheetah Outreach Trust has been placing Anatolian Shepherd livestock guarding dogs on South African farms since 2005. By the end of June 2025 it had provided 406 dogs — the majority, though not all, in South Africa. Of those, 340 were deemed successful and 66 were rehomed because of behavioural problems. The Trust reports that as an 84% successful placement rate (Cheetah Outreach 2025).

Now read the next line in the same report. Of those 340 successes: 32% are still actively working, 13% are retired, 53% died on duty — snakebite, injury, confrontation with a predator, or old age — and 2% moved out of cheetah range.

So the 84% is a non-removal rate, cumulative over twenty years. It is not an efficacy rate, and it says nothing about whether anybody's lambs stopped dying. Because more than half the "successes" are dead, the number of those 406 dogs standing in a flock on any given morning is roughly a third of the placements.

The previous lesson ended on what killing predators actually achieves. This is the alternative, and it has its own arithmetic — which nobody selling it to you will do out loud.

#What the South African evidence actually says

There are three bodies of evidence and they answer different questions.

The efficacy survey. Rust, Whitehouse-Tedd and MacMillan surveyed livestock guarding dog placements across 94 South African farms between 2005 and 2011. Perceived depredation ceased entirely in 91% of placements, with a gross mean annual saving of USD 3 189 per farm (Rust et al. 2013). That is the number everyone quotes. The word carrying the weight is perceived — farmer self-report, from farmers who accepted a dog and kept it, asked by the people who gave it to them.

The same paper reports the failure modes, and a course that quotes only the benefit is selling rather than teaching: behavioural problems in 16% of dogs, and 22% dying prematurely, primarily from snake bites. Treat that first number as soft. The paper sits behind a paywall, the research behind this lesson worked from second-hand summaries of it, and they disagree — one puts behavioural problems at 16%, another at 28% with 17% of dogs removed from their farms. The 22% premature mortality is common to both, and so is the conclusion you budget on: roughly one dog in five does not complete its working life.

The before-and-after trial. McManus and colleagues followed 11 commercial livestock farms in the Eastern Cape through one year of lethal control and then two years of non-lethal control — guarding dogs, alpacas and protective collars together. Total cost per head fell from USD 23.41 to USD 9.60 to USD 5.92, and mean stock losses from 13.6% to 4.4% to 3.7% (McManus et al. 2015). Roughly a four-to-one improvement by the second year.

State the limits in the same breath, every time. There is no control group — the same eleven farms are compared with themselves across successive years, so any change in rainfall, predator abundance, lambing management or reporting behaviour is tangled up with the intervention. The farms were self-selected volunteers who had agreed to stop lethal control, and they knew they were being measured. The lethal-control baseline year is the one whose losses were least likely to be verified and most likely to be over-reported. It is directionally valuable and it is not proof.

The programme data. Behind Cheetah Outreach's numbers sits a method: monthly monitoring visits for the first twelve months, then twice yearly for dogs under two and annually after that, and snake-aversion training, which the Trust credits with preventing the high snakebite mortality of the project's early years. Those are not extras bolted on to a dog. They are the cost of the method. A smaller set in the same report is the reality check: of 25 dogs originally placed in one project area, 13 were still active at the end of 2025, and three died defending livestock in the first half of that year alone.

#Costing a dog against its failure rate

The only published South African-relevant capital figures come from the McManus trial and are in US dollars of the 2014/15 trial period: a guardian dog at USD 553 up front plus USD 432 a year, an alpaca at USD 860 plus USD 98 a year, and protective collars at USD 3.50 each. Treat them as a shape rather than a quote — they do not appear in the paper's published abstract, so they are unconfirmed against the full text.

One dog over a ten-year working life is USD 553 + (10 × USD 432) = USD 4 873.

Now apply the failure rate. With 16% showing behavioural problems and 22% dying prematurely, plan on roughly one placement in five not completing. To hold a working dog in the flock continuously for a decade you must therefore fund about 1.25 dogs' worth of purchase, bonding and the whole first-year monitoring cycle — call it USD 6 090 over the decade — plus a gap, the months between one dog failing and the next puppy being bonded and useful, during which the flock is unguarded and your losses revert.

Against a perceived saving of USD 3 189 per farm per year, even that inflated cost pays back inside about two years. Do not fall in love with that ratio: both sides of it are weak numbers. What the arithmetic is good for is the shape of the decision — the puppy is a small fraction of the cost, and the replacement provision is the line most farm budgets leave out entirely.

#Alpacas and donkeys: what is and is not established

The honest South African position on alpacas is short. The McManus trial bundled guarding dogs, alpacas and protective collars into one non-lethal package, so it establishes something about the package and nothing about alpacas on their own — no effect size can be pulled out of it. What it does give you is a cost: USD 860 plus USD 98 a year in 2014/15 dollars, a higher capital cost and a much lower running cost than a dog.

Guardian donkeys are in the same position. Work on donkeys as livestock guardians in an African savanna has been published (Eur. J. Wildl. Res. 2026) but it sits behind a paywall and its findings, sample size and even country have not been read here. No South African trial isolating the effect of a guardian donkey or alpaca from the herding it usually comes with has been found.

There is also a mechanism question worth a thought before you spend, and it is a hypothesis rather than a finding: a lone camelid's anti-predator behaviour is an alarm-and-confront response that evolved against solitary Andean canids, while black-backed jackals commonly hunt in pairs or family groups. Whether one alpaca is the right instrument against a coordinated pair is exactly what your own before-and-after records could answer, and no published South African trial has.

#The herding dog is a different animal doing a different job

A trained herding dog is the cheapest labour multiplier available to a small mixed operation, and it is directly regenerative — because in planned grazing the bottleneck is almost never the fencing. It is the number of people available to move animals calmly, often. One person with a good dog does the frequent, low-stress moves that one person alone cannot, which is the difference between a grazing plan on paper and a grazing plan that happens.

Two breeds carry an organised South African structure. The Border Collie works close, gathers and fetches, and is exceptionally biddable — with a correspondingly high need for work and stimulation, which is why a bored one becomes a problem. The Kelpie is more independent, more tolerant of heat and distance, and will back sheep in a race. The South African Sheepdog Association runs registration and trials for both, and is where to find dogs, handlers and a realistic idea of what a started dog can do.

Herding also sits on the right side of the damage-causing-animal law. Under the Norms and Standards for the Management of Damage-Causing Animals, GN 512 in GG 40236, commenced 30 August 2016 under NEMBA 10 of 2004, deterrent methods including fencing, collars and herding require no permit, while cage traps, foothold traps, poison collars, darting, call-and-shoot, poison-firing apparatus, denning and hounds require authorisation preceded by a severity assessment and inspection report (summary of GN 512). A dog that moves your sheep is a management tool; a dog run at a jackal is a controlled method. The current text and status of that instrument were not confirmed in the research behind this lesson, and your provincial ordinance sits on top of it — confirm both with your provincial conservation authority before you rely on either.

#What to do on Monday

If you already run a guardian dog and have no loss records for the seasons before it arrived, start them today — otherwise you are funding a decade of something you cannot evaluate. If you are considering one, write the replacement provision into the budget before the purchase price, and check whether your boundary makes a loose 50 kg dog somebody else's problem. And if moving animals is already costing you a labour unit you do not have, the cheaper animal to buy first is the herding dog — as long as you never leave it out there after dark.