The bookmakers have had it their way for a very long time.
Not because they're sharper judges of a horse race than you or I.
Not because they employ the finest racing minds in the country.
They win because they have a machine running on their side of the counter that the ordinary punter has never had access to.
A machine that processes thousands of data points per race, per horse, in under a second, before you've typed your password into your betting account.
It sets the prices you see on your screen.
It adjusts them in real time as money flows in from around the country.
And it doesn't get tired, doesn't have bad weeks, and doesn't carry a grudge against a trainer it fancied last season who cost it money.
It was designed to consider every statistic imaginable, weight each variable against every other simultaneously, and arrive at a price that locks the edge firmly in their favour before a single bet has been struck.
That machine is the reason the bookmaker always seems to be one step ahead.
It's the reason why years of following form, following tipsters, and backing your own judgement never quite translates into the consistent returns it should.
And it's the reason this page exists today.
Because now there's something on your side of the counter that does the same job.
Let me introduce myself before I go any further, because none of this makes sense without the backstory.
My name is Martin Darby, I'm 58 years old, recently retired from a career as a civil engineer, and I now live in Exeter.
I've loved the horses for as long as I can remember.
My father used to take me to the local track as a boy, standing on the rails and watching the races come in while he explained the form to me in the patient way only fathers do.
That love of the sport never left me, but for most of my adult life it sat alongside a frustration I couldn't shake.
I couldn't make it pay the way I was convinced it should.
Three Decades. Never Found Anything That Held.I spent the better part of three decades following tipsters, building my own systems, filling spreadsheets with data, spending evenings with the form guide convinced that this time I'd finally found something that would hold.
The pattern was always the same.
A brilliant six weeks, then a month that gave half of it back.
A superb March followed by an April that wiped the smile off my face by the second week.
I never found anything that produced truly consistent results across every type of card, every going condition, every time of year.
What I did find, the more I thought about it, was the question that sat at the root of the problem.
The bookmakers are using machines to price races with extraordinary precision. Why hasn't anyone built a machine for the punter's side of the equation?
As it turned out, the answer to that question was sitting in a private racing syndicate I'd been invited to join in the spring of 2024.
The Syndicate, The Insider, And The Login Nobody Was Supposed To Know AboutIt started when I was invited to join a private racing syndicate, a small group of serious racing men who'd spent their careers either in the sport or close to it.
A former assistant trainer who'd spent twelve years working around some of the best yards in the North of England.
An ex-bookmakers' rep who'd spent decades on tracks up and down the country and understood the pricing side of the business from the inside.
A racing journalist who'd covered the flat for 25 years and had contacts the rest of us could only dream about.
And a handful of experienced punters like myself.
The idea was to pool our collective knowledge and find the kind of advantage none of us could develop on our own.
The early months were encouraging and the group chat buzzed with sharp observations and sharp disagreements, but the results plateaued.
We'd run well for six weeks and then watch it unravel just as quickly.
Collectively sharp, collectively limited.
Because at the end of the day, we were still a group of human beings working with the same information as every other punter in the country.
No amount of experience or contacts changes the fundamental problem.
A room full of human analysts is still no match for the computing power sitting behind every major bookmaker's pricing desk.
That changed when a new member joined the group in late summer 2024.
He was younger than the rest of us by a considerable margin, mid-twenties, and he arrived quietly without making much of an impression at first.
He'd previously worked in the pricing technology division of one of the UK's biggest online bookmakers, a company whose name you'd know immediately and whose app is likely already installed on your phone.
He'd left under circumstances he kept deliberately vague.
But he left with something he wasn't supposed to have.
He still had access to his former supervisor's login credentials for the bookmaker's internal pricing and selections software.
The supervisor, as far as the young lad could tell, had no idea that login was still being used by anyone.
We were sceptical, naturally.
People talk a very good game in racing circles, and most of them don't have anything to back it up.
We asked him to prove it.
He sent through his first selections two days later.
Seven bets, five winners, including a 7/1 shot that had drifted out all morning and that virtually every member of the group had already written off their list.
The following week, eight bets and six winners.
The week after that, nine bets and seven winners, including a 9/1 that scrambled home in the dying strides of a competitive handicap when most of the race appeared lost.
By the end of the first month, the scepticism in the group had completely evaporated.
Everyone stopped asking questions about the source and started placing the bets.
Over the months that followed, I spent a great deal of time talking to the young lad about the mechanics of the system.
Not just what it produced, but what it looked at.
Why it would back a horse at 8/1 when the form book suggested that horse should be much bigger in the market.
What combination of factors triggered a selection and what caused the system to dismiss a seemingly obvious pick.
Let me try to explain what this type of machine actually does, because once you understand the scale of it, you understand why the human tipster is fighting with one hand tied behind their back.
A sharp human analyst might consider 30 to 40 variables when assessing a race before arriving at a selection.
Form, going, distance, draw, jockey, trainer form, weight carried, course record, days since last run, price movement in the market.
If they're particularly thorough, perhaps headgear changes, sectional times, pace bias at the track, and trainer strike rates broken down by race type.
That's already more than most punters consider.
Not sequentially, one after the other, but simultaneously, cross-referencing all of them against each other at the same time.
Because the draw matters more at some tracks than others, and more in certain going conditions than others, and more for certain types of horse than others.
And the trainer's form matters more in certain race types than others, and more in certain combinations with certain jockeys than others.
No human brain can hold 180 inter-related variables in working memory and process all of them simultaneously, no matter how experienced the person is.
So the human simplifies.
They prioritise the factors and let the others run in the background.
The machine never simplifies, never prioritises, never loses track of a single variable.
Some of the specific factors it analyses are things that would not obviously occur to even an experienced punter.
Pace Maps. Sectional Splits. Trainer Patterns The Form Book Never Captures.It builds pace maps for every race, modelling which horses are likely to push for the early lead and predicting exactly what happens to the race shape when two or three front-runners collide.
A race that looks like a straightforward stamina test on paper can turn into a suicidal pace war from the first furlong, and that shifts the relative chance of every other horse in the field dramatically.
The system models that scenario before the race begins and factors it into every selection accordingly.
It examines sectional splits for every horse in the field going back two full years, not just whether they ran fast in total but where in the race they ran fast.
A horse that gets going in the final two furlongs is a completely different proposition to one that front-runs and fades, and knowing which is which matters enormously when you're trying to predict how a specific race over a specific trip on a specific surface will play out.
It identifies trainer patterns that no form book captures, because they only become visible when you're examining hundreds of runs simultaneously rather than reading through them one at a time.
There are training operations that consistently run horses twice in fairly quick succession as a deliberate fitness-building exercise, with the win arriving on the third outing at a price that hasn't adjusted to reflect the plan.
There are jockey and trainer combinations that significantly outperform the available odds over rolling 90-day windows, patterns the market consistently undervalues because the sample is too small for casual observation but large enough for a machine processing the full dataset.
The system has identified all of them across the last five years of racing data.
It also tracks market movement in the 48 hours before each race, looking for the specific patterns that indicate serious professional money rather than casual backing from the public.
An 8/1 that shortens to 6/1 in a particular sequence, in specific amounts, at particular times of day, tells a very different story from an 8/1 that shortens to 6/1 because the morning papers tipped it.
The machine reads both and knows the difference.
All of that analysis is synthesised into a single probability score for every horse in every race, produced before the market opens each morning.
When that probability score is materially higher than the odds on offer would suggest, that's the bet.
No gut feel.
No sentimentality for familiar names.
No hangover from yesterday's results.
The same cold, consistent calculation applied identically to every race, every day, without exception.
The syndicate access worked very well for as long as it lasted.
But it made me uneasy from the start in a way I couldn't quite suppress.
We were depending on a login that didn't belong to us, and I had no way of predicting when it might stop working.
A routine IT security audit, a new supervisor who changed their password as a matter of habit, a flagged access attempt from an unusual IP address.
Any of those things could end the arrangement overnight with no notice.
So while I was using it and making consistent money from it, I was simultaneously studying everything it produced.
I asked a lot of questions, and the young lad who'd brought the access to the group was patient enough to answer most of them in reasonable detail.
I kept detailed notes across months of conversations.
Not the code, I wouldn't know where to begin with that, but the logic underneath it.
The order in which it applied its filters.
The thresholds it used to eliminate horses from consideration regardless of their form.
The weight it placed on market signals versus pure form signals, and the specific circumstances under which it overrode one set of data with the other.
By the time I had about six months of those notes, I felt I had enough of the underlying architecture to work from.
At that point I brought in Patrick Burrows.
Patrick is a former colleague from my engineering days who retrained as a data scientist after leaving the firm and now spends his professional life building predictive models for a large logistics operation.
His job involves estimating probabilities that carry real financial consequences if they're wrong, so his tolerance for fuzzy methodology is essentially zero.
He was the technical brain behind this, I was the racing brain, and together across the best part of eight months of evenings and weekends we built the first version of what became Bet Machine Pro.
We trained it on a dataset of over 40,000 UK flat and jumps races going back five years.
We incorporated the methodology from my notes, refined it using Patrick's modelling experience, and ran it through extensive backtesting before we ever looked at a live race.
The First Version Was Not Impressive | So We Fixed It.The first live version was not impressive.
It had the shape of something worthwhile but the edges were rough.
It handled flat handicaps well, coped reasonably with pattern races, and struggled badly with staying chases in testing ground.
It was overweighting distance form and not adjusting quickly enough when horses faced significant going changes between runs.
We went through the failure cases in detail, hundreds of them, asking the same question each time.
What was the model missing?
In most cases the answer was the same.
It was treating variables as independent when they interact.
Going and pace scenario are not separate factors you can score and add together.
They influence each other in ways that depend on the specific horse, the specific track, and the specific conditions.
Draw and field size interact.
Trainer form and race class interact.
Patrick rebuilt the model to account for those interactions properly, and the accuracy improvement was immediate and measurable.
We then ran the revised model in parallel against the syndicate output for four months before going live independently.
Bet Machine Pro and the bookies own system agreed on more than 70% of selections during that period.
On the 30% where they diverged, Bet Machine Pro was right more often than not.
That told us what we needed to know.
We switched to running entirely on our own system and never looked back.
The 18 months of results since then are the evidence of whether that was the right decision.
The Full Results, Every Month, Every Figure, No GapsI'm going to lay every month out in full.
January 2025 when we went live, right through to the end of June 2026.
You'll see the stronger months and the quieter ones.
You'll see that not every month is exceptional.
What you won't see is a single losing month across 18 consecutive months of live operation.
That's the part that matters.
| Month | Bets | Winners | Strike Rate | Profit |
|---|---|---|---|---|
| Jan 2025 | 44 | 25 | 57% | £3,480.20 |
| Feb 2025 | 38 | 23 | 61% | £4,215.60 |
| Mar 2025 | 51 | 28 | 55% | £5,140.00 |
| Apr 2025 | 46 | 29 | 63% | £6,380.75 |
| May 2025 | 53 | 30 | 57% | £4,720.40 |
| Jun 2025 | 49 | 31 | 63% | £6,910.50 |
| Jul 2025 | 55 | 34 | 62% | £5,803.75 |
| Aug 2025 | 52 | 32 | 62% | £6,240.00 |
| Sep 2025 | 47 | 27 | 57% | £5,060.30 |
| Oct 2025 | 44 | 25 | 57% | £4,510.80 |
| Nov 2025 | 55 | 34 | 62% | £5,200.00 |
| Dec 2025 | 40 | 24 | 60% | £3,880.00 |
| Jan 2026 | 46 | 27 | 59% | £4,190.60 |
| Feb 2026 | 41 | 26 | 63% | £5,640.00 |
| Mar 2026 | 54 | 35 | 65% | £7,080.40 |
| Apr 2026 | 50 | 33 | 66% | £6,850.20 |
| May 2026 | 57 | 36 | 63% | £6,540.00 |
| Jun 2026 | 45 | 28 | 62% | £4,975.00 |
| TOTAL | 827 | 526 | 64% | £102,617.50 |
January 2025 was the quietest month in that table.
£3,480.20 from 44 bets is still a very decent return for a month's betting, but it's the floor, and after 18 months we know what the floor looks like.
March 2026: Over £7,000 At A 65% Strike Rate.March 2026 was the peak so far, over £7,000 at a 65% strike rate.
That kind of month happens when every layer of the model is firing at the same time and the market hasn't adjusted fast enough to close the gaps the system is finding.
But the months that sit in the middle of that table, the £4,000 to £6,000 months, are the ones that tell the real story.
Because they're the ordinary months, the months where nothing exceptional happened, and they were still profitable.
Every single one of them.
2026 has been the strongest year so far by a clear margin.
By the end of June we're already at £35,276.20 for the year alone, well ahead of the pace 2025 set.
There are specific reasons for that improvement, and I'll come to them shortly.
The cumulative total of £102,617.50 is the headline figure from 18 months, but it's not just a big number.
It's a consistent number, arrived at month after month without a break in the run, and consistency is the thing that distinguishes a system with a real edge from a system that got lucky for a while.
What My Test Group Said After 30 DaysAbout 30 days ago I decided to open the selections up to a small test group of everyday punters who were willing to follow everything properly and give me straight, unvarnished feedback.
Not handpicked members who I'd given an easy run of results.
A broad cross-section of people with different levels of betting experience, different starting points, and different expectations.
What none of them had was any reason to paint a prettier picture than reality.
They were free to ask for their money back at any point in the 30-day window.
Several of them said in advance that they expected to do exactly that.
None of them did.
The people closest to this project keep asking the same question.
2025 was a strong year by any measure.
Why are the 2026 numbers running even further ahead?
There are three clear reasons and none of them are vague or speculative.
The first is data volume.
When we launched in January 2025, the model was trained on around 40,000 historical races covering five years of UK racing.
By the time we moved into 2026, that original training data had been supplemented by a full year of live racing processed in real time.
That's not just more data. It's the most recent data, which matters because the sport shifts in ways that older records don't capture.
Trainer patterns change as yards evolve their approach.
Track surfaces change as racecourses invest in drainage and groundskeeping infrastructure.
Market behaviour changes as professional betting syndicates move in and out of different races and different markets.
A model trained on data that runs to last month is a sharper model than one trained on data that runs to two years ago, and that difference shows in the output.
The second reason is the going model revision Patrick completed in February 2026.
We'd been aware for some months that accuracy on soft-to-heavy ground at certain tracks was slightly below where we'd expect it to be.
The issue, when we investigated it properly, was that the model was relying on official going descriptions as given, when certain tracks with older drainage systems diverge meaningfully from those descriptions in the days following significant rainfall.
The declared going and the actual surface are two different things at some courses, and the horses that suit the declared going are often not the same as the ones that suit the real ground.
Patrick built track-specific going correction data into the model, calibrated against actual race performances rather than official declarations, and the accuracy improvement in wet conditions was clear and immediate.
The third reason is market timing, and it's had a more measurable effect than we expected when we started working on it.
The model identifies value at a certain price.
But the price you actually get depends on when you back the horse, and in most markets that price moves meaningfully between early morning and the off.
We spent the first part of 2026 refining the guidance around when to place each type of bet, and members who've followed that guidance have consistently achieved better average return prices than those who placed everything as soon as the email arrived.
That timing guidance is now included in the daily selection email as standard.
Even an average half-point improvement in return price across 50 bets a month compounds into a meaningful difference over time.
The March and April 2026 figures in particular reflect that improvement.
Why The Human Tipster Is Now Working At A Structural DisadvantageThere are some very skilled racing analysts in this country who've devoted their careers to the sport.
I don't want to dismiss what they do, because at their best it represents decades of accumulated knowledge and judgement.
But the environment they're working in has shifted in a way that makes the traditional tipster approach structurally disadvantaged, regardless of how good the individual is.
When a bookmaker's algorithm is simultaneously assessing 40 races across six venues, running 180 variables per horse, and adjusting prices in real time as money arrives from around the country, the human tipster working through the morning card with a coffee and a browser isn't competing on the same terms.
They're doing a related but much more limited version of the same job.
And the prices the bookmaker is offering already have that computational advantage baked in.
Every price on your screen is the output of the bookie's machine.
To find consistent value against those prices, you need a process that can match the analysis that produced them.
The only way to do that is to run a machine of your own.
That's what Bet Machine Pro does.
It runs the same type of analysis, on the same categories of data, at the same processing speed, and identifies the gaps where the market price hasn't yet closed around what the model is finding.
Those gaps aren't huge.
They don't have to be.
Over 827 bets and 18 months of live operation, working a consistent edge in the same methodical way, they've added up to over £102,000 in cumulative profit.
Not because Bet Machine Pro wins every race.
It doesn't, and anything that claims to is lying.
But because it finds the same type of value, in the same consistent way, month after month, without ever getting tired or emotional or distracted.
That's the edge.
Now, For The First Time, I'm Opening The DoorsI kept Bet Machine Pro within the syndicate and my close circle for almost two years, with no particular intention of changing that arrangement.
The results were doing exactly what we wanted for everyone involved, and there was no pressure to expand.
What changed my mind was my brother.
He started following the selections about six months ago with a starting bank of £200 and turned it into just over £3,400 in under three months.
He wouldn't let up about how I should be letting more people have access to it.
His friend from his golf club followed the next month with very similar results.
And after the test group returned the feedback you've just read, I ran out of arguments for keeping the doors shut.
I Made An Executive Decision...I'm now opening a strictly limited number of places in Bet Machine Pro, a daily email membership service where the model's top selections land in your inbox every morning before the first race of the day.
The email contains everything you need to know about that day's selections.
What to back, what price to look for, the recommended stake, and the timing guidance for getting it on.
That's it.
The analysis behind every selection has already been done, by the machine, before most people have had breakfast.
All you need to do is get the bets on at your local bookies or online, then check how you've done in the evening.
It works whether you're a seasoned punter who's been at the races for decades or someone who's only just started exploring the horses seriously.
The edge comes from the model, not from what you already know.
No app, no software, no research.
An email arrives, you act on it, and the machine does the rest.
A few people have asked me to describe exactly what they'd receive each morning, and it's a reasonable question to ask before joining anything.
The email goes out before 8am on every day there's UK racing.
It contains between 2 and 6 selections depending on what the model has identified as offering a clear edge in that day's racing.
If there's no strong edge in any race on a particular card, the model doesn't force a selection just to give you something to bet on.
That matters more than it might seem.
One of the consistent failures of human tipsters is the pressure to produce tips every day, even on days when there's nothing worth backing.
The model doesn't feel that pressure.
On a light card with nothing compelling to offer, the email will say so and tell you to sit the day out.
On a strong card where the model has found multiple well-priced selections, you'll have everything you need to get on at the best available odds.
Each selection is presented with the horse's name and the race details, the minimum price to look for before backing, the recommended stake as a percentage of your total bank, and the timing window for when to place the bet to get close to the modelled price.
There's also a brief note on each selection explaining what the model has identified in plain English.
Not jargon, not a page of statistics.
A two or three sentence explanation of why the model rates this horse above its market price.
Everything is designed to be actionable in the time it takes to read it over a cup of tea before you leave the house.
No form reading required on your part.
No cross-referencing with anything else.
The machine has done all of that before the email was sent.
The Cost, And Why I've Set It Where I HaveI make more than enough from my own betting to make the membership fee irrelevant to me personally, and I want to be straight about that.
If commercialising this properly was the goal, Patrick and I have had enquiries from institutional betting operations that would pay considerably more than a £20 joining fee for access to the model.
That's not what this is about.
This is about giving a small group of ordinary punters access to something the industry has kept entirely to itself, and covering the costs of running the infrastructure that makes it possible.
The one-off joining fee of £20 covers the website and email delivery costs, the server infrastructure Patrick maintains for the daily model runs, and his ongoing time for refinements and updates to the system.
One payment, nothing further, ever.
No monthly subscription sitting in the small print, no renewal date six months down the line, no hidden tier you need to upgrade to in order to receive the actual selections.
£20, paid once, and that covers your membership for life.
To put that number in context against what the model typically produces, most of the test group covered the joining fee before they'd finished their first morning of following the selections.
Several of them covered it on the first bet placed.
Take a full 30 days with Bet Machine Pro, place every set of selections as they arrive, follow the recommended stakes and timing guidance, and judge the service entirely on what it produces during that period.
Track every result.
Put it through its paces without holding anything back.
If at any point during those 30 days you feel it isn't delivering what I've described here, send me an email and I'll refund your joining fee in full.
No conditions, no awkward questions, no delays.
Your money is completely safe for the full month.
I make that offer without any hesitation because 18 months of live results give me complete confidence in what the system produces, and I'm not the slightest bit concerned about anyone calling in that guarantee.
The ceiling of 50 members is a practical necessity, not a marketing device, and it deserves a proper explanation.
When 50 people receive the morning selections and place them within a similar window of time, the aggregate money going onto each horse is manageable without meaningfully moving the market.
The prices members receive are close to the prices the model identified as offering value.
Scale that membership to 150, or 300, or more, and the same horses are being backed by dramatically more money in a concentrated period.
Markets respond to that very quickly.
A horse at 6/1 becomes 4/1, or 3/1, or shorter still.
And when the return price drops that sharply, the edge that made the system work starts to erode.
At some point the odds are no longer long enough to justify the stake, and the advantage disappears.
The syndicate and I ran the numbers carefully and 50 is the figure we're confident sits safely below that threshold.
It won't be reviewed upwards at 60, it won't quietly become 75, and it certainly won't double because the demand is there.
The existing members know this, and they know it protects the quality of what they're receiving.
As of today, 43 of those 50 places have already been taken by members currently following the daily selections.
Seven remain.
I can't say with any confidence when or whether another intake will open after these are gone, because that depends entirely on members leaving, and very few have.
If this page is still live and the button below is active, there's a reasonable chance a place is still available.
I wouldn't bank on that still being the case if you come back to check later.
Taking on additional members beyond 50 risks reducing the quality of returns for everyone already in the group, and that's not something I'm prepared to do to people who've trusted the service.
If you want a place, this is the moment.
Click the order button below and complete the checkout process.
You'll receive a confirmation immediately and you'll be set up and in the system before the end of the day.
From the next racing day, the morning selections will land in your inbox with everything you need to place the bets, at the time you need them.
No software to install.
No app to download or configure.
An email arrives, you act on it at your convenience, and the machine has already done everything else.
One payment of £20, lifetime membership from the day you join.
The 30-day guarantee means you have a full month to judge whether this is for you, with nothing to lose except the seven available places if you wait too long.
P.S. This is a one-off payment of £20 for lifetime access to the daily selections.
There's no subscription, no renewal coming in 30 days, and nothing waiting for you in the small print.
The 30-day money-back guarantee means the only real risk is leaving it too long and finding the final 7 places are gone when you come back.
Join today and you'll receive the morning's selections immediately, which for most members is more than enough to cover the joining fee before the afternoon races are done.
Seven places left.
I'll see you in the members' area.
All the best,
Martin Darby