Let me ask you something.
When was the last time you felt properly in control of your betting?
I don't mean a lucky Saturday when two outsiders came in and you walked round the house feeling ten feet tall.
I mean week after week, when a losing day stops rattling you because you know exactly why each bet was placed.
That feeling, knowing you're acting on hard information rather than hope, is what Senator Bet Bot hands you every morning.
And I want to show you how it does it.
It's Friday the 2nd of October 2026, a little after 7 o'clock, and today's selections have already gone out to every member.
The software finished its overnight run at 5:52 this morning, while I was still asleep and the house was dark.
By the time I'd let the dog out and filled the kettle, it had read every race on today's British card and thrown most of them back like undersized fish.
What's left is a short list, usually 2 or 3 horses, and on some mornings fewer than that.
Nothing on that list got there because somebody liked the look of it.
Every name on it had to survive a process that rejects far more horses than it keeps, and I'll show you exactly what that process looks like further down this page.
I'll also tell you where the whole thing came from, because it didn't begin in a laboratory or a tech company.
It began with a greengrocer's pencilled notebook and a conversation after five-a-side on a wet Tuesday night.
Right now, in the first days of October 2026, my AI betting tool is running better than at any point since the first line of code was written.
It's finding patterns buried so deep in the numbers that no tipster working with a pen and a form guide would ever stumble across them.
And it does all of this each morning before most people have finished their first cup of tea.
The result is 2 or 3 of the most carefully calculated horse racing selections you'll find anywhere in this country.
These aren't hunches, and they're not the sort of thing you'd get from someone who glanced at the runners over a bacon sandwich.
They come from a system that doesn't get tired or emotional, and doesn't lose its nerve after a bad afternoon at Haydock.
It doesn't care who trains the favourite or what the pundits were saying on the television last night.
All it cares about is what the evidence says, and whether that evidence is strong enough to risk money on.
In June, Senator Bet Bot delivered a profit of £4,957.40.
That came straight after May, which finished at £7,085.10 and still stands as the best month Senator Bet Bot has ever produced.
Then summer arrived properly, with evening meetings on most days of the week and fields full of horses the software had plenty of fresh data on.
August came in at £6,958.30.
And September, which closed on Wednesday night, added £5,874.90.
Across the first 9 months of 2026, the total now stands at £52,445.10.
All of it has been banked in active betting accounts, and every penny of it can be withdrawn tax-free whenever I choose.
I'm not telling you that to impress you, because figures on a web page are easy to type and I know you've seen plenty of them before.
I'm telling you because I want you to understand where those figures come from, and by the end of this letter I think you will.
A plumber from West Yorkshire cleared £1,900 over the summer while working full time.
A hairdresser from Tyneside told me last week she'd paid for a family holiday in Cornwall out of her summer winnings.
And a retired civil servant from Kent has followed the bets since the start of July without a losing month.
None of them are professional gamblers, just ordinary people who were happy to let the model do the heavy lifting.
"Tom, I've stuck your morning email to the fridge every day since July, because it's the only betting advice my wife has ever agreed with. We hardly talk about the horses at all these days, which tells you how calm the whole thing has been."
Gordon Tate, Hexham
"I kept a spreadsheet of every selection for 10 weeks before I staked a penny, because I've been burned by tipsters who only remember their winners. My spreadsheet and Tom's figures matched line for line, so I started betting in August and I haven't looked back since."
Wendy Barraclough, Skipton
"Who'd have thought the most relaxing part of my week would be placing a bet on my phone in a supermarket car park at half past six in the morning? I drive a lorry on early starts, and the email is already waiting for me when I pull in for my first break."
Neil Ashby, Kettering
"My only regret is the 3 years I wasted chasing tipsters on social media before my brother forwarded me this page. If you're where I was, stop reading reviews and give it a month, like I eventually did."
Pauline Dewhurst, Morecambe
There's more feedback further down the page, but if you've got a few minutes to stay with me now...
I want to tell you the full story of how Senator Bet Bot came to be.
Because it didn't start the way you might expect.
My name's Tom Cavendish, I'm 54, and I've spent the last 26 years teaching maths at a secondary school in Chesterfield.
For 19 of those years I've also been the person who builds the school timetable, which is a job most teachers would rather eat chalk than go near.
If you've never seen one being built, imagine squeezing 1,400 lessons a week into 41 rooms, around 83 teachers who each have their own part-time days and awkward requests.
It takes me most of August every year, and the first version is always wrong.
But somewhere around the fifth or sixth attempt it clicks, and I've always loved the moment when a problem with far too many moving parts suddenly holds together.
Horse racing has been the other great puzzle in my life for as long as I can remember.
It started with my grandad, Walter, who ran a fruit and veg stall on Chesterfield market and never once let a Saturday go by without a bet.
He kept a battered black notebook in the inside pocket of his coat, and he called it his hard-luck book.
Every time he watched a horse get trapped on the rails or finish full of running with nowhere to go, he'd write its name down in pencil.
He didn't always win when it did, but he won often enough that my nan had stopped complaining about the notebook by the time I was 10.
I'd sit on an upturned crate behind the stall on a Saturday afternoon, listening to the commentary on his little radio, while he checked the runners against his pencil marks.
I didn't know it then, but that little book was the first racing system I ever saw, and a version of it lives inside Senator Bet Bot today.
For me, racing has never been just a way to make money, although that's obviously part of it.
I've always been drawn to problems with too many variables, the kind where most people give up because there's too much to hold in your head at once.
And for years, I threw myself at it the way most serious punters do.
I spent evenings with form books spread across the kitchen table.
I kept meticulous records on spreadsheets that eventually ran to thousands of rows.
I tracked trainers, jockeys, track quirks, going changes, stall positions, seasonal trends, market movers, and every other angle I'd read about or stumbled across myself.
By 2019 the main spreadsheet had grown to 38,000 rows and took nearly 4 minutes to open on our old laptop.
My wife, Alison, once asked me whether it would be quicker to go to the races and ask the horses.
She had a point, but I wasn't ready to hear it.
And I was good at it, better than most people who go about it the same way.
But I was never quite good enough.
Something was always slipping through, whether it was a factor I'd weighted too heavily or a race I should have left alone because I'd talked myself into a position I didn't really have.
Anyone who tells you they've completely removed feeling from their betting is either lying or hasn't had a bad enough run yet.
After a string of losers, I'd start tinkering with my approach to try to win back what I'd dropped.
After a good run, I'd get overconfident and start taking risks I'd normally avoid.
I remember one Saturday in the spring of 2021 when I backed 6 horses I'd never have touched on a calm day, purely because the first bet of the afternoon had lost and I wanted the money back by teatime.
Not one of them won, and I sat in the car outside my brother's house for 10 minutes before going in for his birthday tea, because I didn't want anyone to see my face.
It was a cycle I couldn't break out of using willpower alone.
It was something that didn't have any emotional state to manage in the first place.
The school timetable had taught me that lesson years earlier, if only I'd been paying attention.
I never solve a timetable in my head, because my head gets tired and starts cutting corners by about 11 o'clock at night.
I solve it with software that checks every possible clash, every single time, without caring how late it is or how badly I want to go to bed.
It took me an embarrassingly long time to see that my betting needed exactly the same treatment.
That realisation sat with me for a while before I did anything about it.
And then, in late 2023, I had a conversation that sent everything I was doing off in a completely new direction.
I was at a five-a-side football night, nothing glamorous, just a bunch of blokes who'd been playing together for years and always ended up in the pub afterwards.
It was a wet Tuesday in November, and I'd spent most of the game in goal because my knees don't get on with the artificial pitch any more.
One of the lads, a software developer I'd known for about four years, started talking about what he'd been working on.
He was building machine learning models for a logistics company, the kind of systems that predict delivery delays before they happen, based on hundreds of inputs updated in real time.
He told us his model could look at a van leaving a depot in Doncaster and say, with surprising accuracy, whether a parcel would reach Stoke on time.
It weighed up traffic, weather, the driver's route history and dozens of other signals that no human dispatcher could keep track of at once.
I didn't want a model that looks at 5 factors and gives you a tidy ranking.
I wanted one that could cope with the real mess of what happens in a race, with all its moving parts and contradictions.
I asked him, half seriously, half as a throwaway idea, whether something like that could be built for betting.
He thought about it for a moment and said yes, but the data pipeline would be the hard part.
By the end of it, we'd sketched out the rough shape of what would eventually become Senator Bet Bot.
The sketch was drawn in biro on the back of a beer mat, and it now sits in a frame above my desk, slightly stained and completely unreadable to anyone but the two of us.
His name's Mark, and he doesn't follow racing at all, never has.
But he understood data architecture better than anyone I'd ever spoken to, and he saw straight away that the problem I'd been wrestling with by hand was exactly what a well-built AI model could tackle.
What sealed it was the moment I showed him my spreadsheet on my phone in the car park afterwards.
He scrolled through it for about 30 seconds, looked up and said, "You've already done the hard bit, you've just been doing it with a teaspoon."
We agreed to give it a go as a side project, with no deadline hanging over us, just to see what was possible.
The first few months were the least glamorous part of the whole journey.
We worked on it 3 evenings a week, usually at Mark's kitchen table, with his cat walking across the keyboard at least once a night.
Mark built the data ingestion framework while I focused on defining what data actually mattered.
That sounds simple, but it turned out to be the slowest part of the whole build.
Horse racing throws off an extraordinary amount of data, across every race on every day of the calendar year.
Not all of it is useful, and some of it actively misleads you if the model leans on it too heavily.
We spent months on the data selection question alone, deciding what should go in and what should be thrown away.
Some inputs only matter under certain conditions, so they had to be tagged and switched on only when those conditions turned up.
Mark would stress-test the architecture to see whether the model was learning real patterns or just memorising noise.
That distinction, between real patterns and statistical noise, turned out to be the most important design decision we made.
The best example of noise came in January 2024, about 2 months into the build.
One of our early test versions suddenly developed a strong liking for horses with short names.
For a few days I thought we'd stumbled on something extraordinary, and I even scribbled down a list of 4-letter horses running that weekend.
Mark dug into it and found the pattern came from a quirk in how one of our data sources shortened long names, which meant a batch of winners had been stored twice.
The model hadn't learned a single thing about racing, it had learned about a typing error.
We laughed about it at the time, but it frightened me, because that's exactly how most betting systems fool the people who build them.
They look brilliant in backtesting and then fall apart in live conditions because they've essentially just memorised the past rather than understood it.
We went to enormous lengths to avoid that.
Being a teacher, I ended up thinking about the problem the way I'd think about exams.
You don't test a pupil with questions they've already seen the answers to, because all you learn is how good their memory is.
So we locked away a full season of races that the model was never allowed to look at, the way you'd seal a mock paper in an envelope and put it in the headteacher's safe.
Only when a version of the model had been built and frozen did we open the envelope and let it sit the exam.
Every setting we built in had to show predictive power on races it had never seen before, not just the ones it had been trained on.
Most of our early versions failed that exam, and a couple of them failed it so badly that Mark deleted them the same night.
Version 7 did better again, and it's the ancestor of everything that runs today.
We still use the sealed envelope for every change we make, and nothing reaches members until it has passed.
It slowed us down a great deal, but it meant that when we finally started running the system live, it held up.
It wasn't perfect, because nothing is in its early days, but it held up well enough to tell us we were onto something real.
People often picture an AI betting tool as one big brain that stares at a race and spits out a winner.
Senator Bet Bot doesn't work like that, and I think it's worth explaining how it does work, because it's the reason I trust it with my own money.
It's really a team of smaller models, each trained to answer one narrow question about a race.
One of them only thinks about pace, and how the race is likely to be run from the stalls to the line.
A third reads the trainer's past behaviour, while a fourth watches the betting market move overnight.
There are 9 of these specialist models in total, and none of them is allowed to pick a bet on its own.
Their answers go to a final model that Mark calls the referee, which decides how far to trust each specialist for this particular type of race.
On a 5-furlong sprint on fast ground, the referee leans heavily on the pace specialist, because pace decides most of those races before they've properly started.
On a 3-mile chase in the mud, it leans much more on fitness and the trainer's record with stayers.
The referee learned those weightings for itself, from thousands of past races, and it adjusts them every night as new results come in.
Once it has weighed everything up, it produces a single number for every runner, which is the software's own estimate of that horse's chance of winning.
The bot replays each race 20,000 times in simulation, with small random changes each time, like a slow start or a horse getting shuffled back on the home turn.
If a horse only wins in a narrow band of those imaginary races, the software treats it as fragile and lets it go.
If it keeps winning, or keeps finishing close, across thousands of different versions of the race, it becomes a candidate.
Mark calls this the stress test, and it has thrown out plenty of horses I'd have happily backed in my spreadsheet days.
All told, the program weighs up 214 separate pieces of information for every runner before the referee has its say.
It retrains itself every night on the day's results, so the version that picks tomorrow's bets is never quite the same as the one that picked today's.
It's easier to picture all this with an example, so let me walk you through the kind of race the software sees every day.
Imagine a 7-furlong handicap at a Midlands track on a Tuesday afternoon, with 11 runners and good to soft ground.
Most punters would glance at the favourite, notice it won last time, and either back it or go hunting for something at a bigger price.
The system starts somewhere else entirely.
First the pace specialist maps the race and finds only one horse that likes to lead, which means the early gallop is likely to be steady and the hold-up horses will struggle to get involved.
Next the fitness specialist notices that the last-time-out winner was pushed hard on heavy ground 9 days ago, and it knocks that horse's estimate down a peg.
The trainer specialist then spots that one of the mid-market runners comes from a yard whose horses often improve on their first run in a handicap.
Meanwhile the market specialist has seen a steady trickle of money for that same horse since the overnight prices went up, with none of the frantic surges that usually come from casual punters.
The referee pulls all of that together, and the stress test replays the race 20,000 times with the pace and the traffic shuffled around each time.
If that horse keeps finishing in the first 2 in the large majority of those replays, it makes the shortlist, and if it only wins when everything goes perfectly, it's dropped.
Even after all that, it still has to beat the other shortlisted races of the day for a place in the morning email.
That's the journey every single selection takes, and it's why so few of them make it through.
When people ask me what Senator Bet Bot analyses, the straightforward answer is: far more than I can explain in a short space.
But let me give you a sense of the key layers.
The most obvious inputs are the ones every serious punter considers.
Recent form, trainer stats, jockey booking, going preference, distance suitability, course record.
The model handles all of those as a starting point.
But where Senator Bet Bot pulls clear of anything a human analyst could copy is in the cross-referencing.
The model builds a detailed pace map for every race.
It works out the likely early pace and how that pace will spread through the field, then picks out the horses that should benefit from the way the race is run.
Then it lays that over furlong-by-furlong speed data from the horse's previous runs, looking past whether it won or lost to how it spent its energy and where it made its move.
A horse might have a dull form line that puts punters off, yet its speed data might show it finishing faster than the winner in each of its last 2 starts, after being stuck behind traffic it couldn't get round.
That's the kind of horse the market tends to overlook, and it's exactly the kind the model finds time after time.
The software writes it into what we call the hard-luck register, and it keeps an eye on every horse in there until the circumstances line up for a proper run.
At the moment there are just over 1,100 horses in that register, and the list changes every night as new races are run.
A horse only leaves the register when it gets the clear run it was denied, or when Senator Bet Bot decides the bad luck was really a lack of ability dressed up as misfortune.
That second judgement matters more than you'd think, because plenty of horses are called unlucky by their connections when the truth is they simply weren't good enough.
Senator Bet Bot watches how prices move from the overnight markets through to the morning exchanges, and it reads those movements against its own estimate of each horse's chances.
When a price drifts out, most people take it as a bad sign, but that isn't always what's going on.
Sometimes a horse is left to drift by people who want a bigger price before they get involved late on.
The model knows what normal drift looks like, and it flags the unusual patterns that suggest informed money is waiting in the wings.
It works the other way round as well.
When a horse is backed hard first thing, the model checks whether that support makes sense given everything else it knows, or whether it's just a popular yard's runner attracting the usual weight of hopeful cash.
Big names in the training ranks draw money from casual punters whether their horses deserve it or not, and the software has learned to tell the difference.
Most punters look at how many days it's been since a horse last ran.
Senator Bet Bot looks deeper, at the total physical toll across a run of races, taking in distance, pace, ground and how hard the horse was asked to race.
A horse that had an easy spin 14 days ago is in a very different state to one that ran flat out on heavy ground 17 days ago.
The form book makes those 2 horses look identical, but the model can tell them apart.
We built this layer after a winter in 2024 when the model kept backing horses that looked fresh on paper and then faded tamely in the last furlong.
Mark spent a fortnight rebuilding how it measured effort, and those tame finishes all but vanished from our list of losers.
And then there's the trainer pattern recognition.
This is one of the areas where the AI operates at a level no human could match.
The model has catalogued the habits of several hundred British trainers across thousands of race entries.
It knows which trainers tend to run horses for experience before targeting them at a race they're primed for.
It knows which ones use particular jockey bookings as a signal of intent.
Some trainers' horses run best fresh, while others need a couple of outings to find their rhythm, and the model keeps track of which is which.
It even notices when a yard's runners have gone quiet for a fortnight, which often means a bug is doing the rounds of the stable, and it steps away from that yard until its horses start running to form again.
A human analyst might develop an intuition for one or two trainers they follow closely.
Senator Bet Bot holds that level of knowledge across the entire British training landscape simultaneously.
There aren't enough hours in the day.
If you watched the model at work for a morning, the thing that would strike you most isn't how many bets it finds.
On an average day there are somewhere between 30 and 45 races on the British card, and every one of them gets the full treatment.
By the time the referee and the stress test have finished, more than 90% of those races have been dropped.
Some are dropped because the field is too big for the software to read with confidence.
Some go because too many of the runners are unraced or lightly raced, which leaves the bot guessing, and it isn't allowed to guess.
Others fall away because the ground has changed since the overnight declarations and the numbers no longer stand up.
And a few are dropped because 2 of the specialist models disagree so strongly that the referee refuses to take sides.
In my spreadsheet days I'd have found a reason to bet in 20 of those races, and I'd have been wrong in most of them.
It waits for the specific races where its reading of the contest is clear and the numbers back it up.
Those are the bets we take, and because we only take them when the case is clear and measurable, the profits keep building month after month.
A member once told me that following Senator Bet Bot felt like having a friend who knows more about racing than anyone alive, but only opens his mouth when he's sure.
I've never heard it put better.
We went live cautiously in the spring of 2024, betting small stakes with our own money and nobody else's.
Controlled testing is one thing, but real money in live markets is something else entirely.
The first few weeks were modest, because the model was finding winners but not enough of them, and at lower prices than I'd hoped.
We were making money, but not at a pace that would excite anyone.
I remember Mark asking me in June whether we should shelve the whole thing, and me saying no a little too quickly, because I didn't want to admit I'd been wondering the same.
We agreed to give it until the end of the summer holidays, which for me meant the week before the new timetable went live.
Then Mark made a significant change to how the model handled its own loss data.
Instead of treating each losing bet as just a negative outcome to learn from in aggregate, we restructured the feedback loop so the model drilled down into the specific characteristics of each losing selection.
Each loser is pulled apart and compared with every similar bet the program has ever placed, winners and losers alike, to find out what made this one different.
It began picking out categories of race where its reading looked right on paper but kept being undone by factors it wasn't weighting properly.
Certain race conditions and certain field sizes kept cropping up, along with combinations of pace and ground it had been treating as similar when they weren't.
One example was 2-mile handicap hurdles on soft ground with more than 14 runners, which had quietly been costing us money all spring.
Once those categories were identified, the model started either avoiding them entirely or applying much tighter criteria before flagging a bet.
The immediate effect was a drop in the number of daily selections.
Some days, the model would find nothing that met its criteria and send nothing.
But then I looked at the results over the following 6 weeks, and far more of the selections were winning than before.
The average price of those winners had held up as well, and that was the part that made me put my mug down and read the figures a second time.
The model wasn't placing fewer bets because it was struggling.
It was placing fewer bets because it had become more selective.
And that selectivity was the thing that transformed the results.
By the end of 2024 we'd tightened the model 11 more times, and each change had to pass the sealed envelope test before it was allowed anywhere near real money.
In December, Mark and I sat in the same pub where it had all started and agreed it was ready for other people.
I should tell you that opening it up to members wasn't my idea.
For most of 2024 I was quite happy for Senator Bet Bot to stay a private project between me and Mark.
Then one evening in November, Alison came into the back room while I was reading the morning's selections for the fourth time and asked me who else knew about it.
She said it seemed a shame, because I'd spent 20 years moaning about tipsters who didn't know what they were doing, and now I'd got something that did.
She reminded me of the Saturday afternoons my grandad spent helping the other stallholders with their bets, scribbling names on the back of paper bags for anyone who asked.
He never charged anybody a penny, but he loved being the man with the notebook.
I lay awake thinking about that for most of the night.
By the morning I'd decided to let a small group in and see what happened.
The first members came in during January 2025, a group of 30 people, mostly friends of friends and a few of the regulars from five-a-side.
January finished at £3,190.40, and that figure pleased me more than any since, because it was the first month where other people's money was riding on the software alongside mine.
February brought £3,640.75, and by April the monthly figure had passed £4,800 for the first time.
Every single month of 2025 finished in profit, and the year as a whole came to £57,800.30.
July and August were the strongest months of that year, at £6,140.90 and £6,495.25, and they taught us that the system thrives on summer racing, when the ground is predictable and the form is fresh.
The quieter months came in the depths of winter, when heavy ground and abandoned meetings thin out the information the software relies on.
December 2025 was the lowest month of the year at £3,920.10, and I was still delighted with it, because a lean month that still pays is the best sign a system has been built properly.
But I won't dwell on 2025, because 2026 is the year Senator Bet Bot really came into its own.
In the summer of 2025, Mark did something that none of our members noticed, which is exactly how he wanted it.
The software had been running on a single rented server, and the nightly retrain was taking just over 4 hours to finish.
That was fine most nights, but on busy days it left very little time to react to late non-runners and changes in the going.
Over 3 weekends in July, Mark moved everything onto a new setup that splits the work across several servers at once.
The nightly retrain now takes about 40 minutes, and that freed up time for a final check at 6:30 each morning, just before the email goes out.
That check looks for overnight non-runners and any change to the going, and if something important has moved, the model reruns the affected race from scratch.
It sounds like a small thing, but a non-runner can alter the whole shape of a race, especially if it was the only horse likely to set the pace.
Since then, that simply hasn't happened.
January began with a change we'd been working on over Christmas.
The all-weather tracks each have their own character, and until then the software had been treating them a little too much like one another.
Mark built a separate pace model for each of the 6 all-weather courses in Britain, because a horse that loves the long straight at one track can find the tight bends of another completely against it.
January finished at £4,480.00, the best January the bot has ever had.
February brought £5,130.50, and it was the month the new going-reader earned its keep.
Official going descriptions can lag behind what the ground is really doing by several hours, especially after rain overnight.
So we taught the software to estimate the real ground for itself, using overnight rainfall readings alongside each course's drainage history.
On more than one morning in February it quietly stepped away from a race because it had worked out the ground was softer than the official description said, and it was proved right by the time the horses went out.
It was Cheltenham month, when a lot of punters go completely overboard, and the program placed fewer bets during festival week than in any normal week of the year.
It simply refused to wade into huge handicaps where half the field had been laid out for that one afternoon and nobody could say which of them had been left the most to find.
While plenty of people were licking their wounds by the Friday night, our members had come through the week in good shape.
April jumped to £6,730.40 as the flat turf season got going, and it gave me one of my favourite days in the whole history of the software, which I'll tell you about in a moment.
May was the month everything clicked, finishing at £7,085.10, which is still our best month on record.
We'd raised the stress test from 5,000 simulations to 20,000 at the end of April, and the selections that came through in May were the most robust I'd ever seen.
Royal Ascot fields are enormous, and the system sat out most of that week because it couldn't read the big handicaps with the confidence it demands.
I'll admit I found that hard to watch, because it's my favourite week of the racing year, but the software was right to keep its powder dry.
July finished at £6,412.60, helped by the long run of evening meetings and the summer festival at Goodwood.
That was the month we added a new filter for firm ground, because a handful of horses had been running below form on quick surfaces in a way the fatigue layer hadn't fully explained.
In August, Senator Bet Bot also started taking travel into account more carefully, so a horse that had been boxed up for 5 hours on a hot day is now treated differently from one that lives 20 minutes from the course.
September added £5,874.90, as the ground started to ease and the big autumn meetings at Doncaster and Ayr came round.
It's a month when the form can turn quickly, and the going-reader we built back in February did a great deal of the work.
Across the first 9 months of 2026, that's £52,445.10, and not one of those months finished in the red.
I want to tell you about a specific moment from earlier this year that stuck with me.
It was a Wednesday afternoon in April.
The model flagged a selection at 14/1, a price that immediately made me look twice.
At 14/1, the market is telling you it's an unlikely winner, and most experienced punters would assume the algorithm had wandered off somewhere.
I pulled up the race card, and the horse hadn't won in 11 months and was dropping in trip after 2 poor runs.
On the surface, there was nothing obvious about it at all.
But I've learned to look past my own reaction and look at what the data is actually saying.
The speed data from its last 3 starts showed a horse finishing with energy it couldn't use, because each time it had been caught in traffic approaching the final furlong.
It had been sitting in the hard-luck register since February, waiting for its chance.
The drop in trip, from a mile and a quarter to a mile, matched a pattern the model had picked out as a strong positive for this horse's stride length.
Then there was the jockey change, with a rider who'd partnered the horse twice 2 seasons earlier, finishing second both times, booked to ride again.
But the model treats it as a meaningful signal when combined with the other factors.
It won by 2 lengths, and it was returned at 14/1.
That's the kind of bet you'd never find on a tipster's list, because no human analyst was putting that horse up.
But the model found the convergence of factors that made it the highest-probability selection in the race.
Most of our selections go off at much shorter prices, but that day reminded me why I trust the system even when my own instincts pull the other way.
Grandad Walter would have loved it, and I'll admit I raised my mug of tea to him as it crossed the line.
There's another day I think about just as often, and it's the opposite of that one.
It was a big Saturday in the early summer, with 6 meetings on the go and one of them live on terrestrial television.
I woke up expecting a busy email, and instead the software had sent a single line to members saying there were no bets that met its standards.
I couldn't believe it, because there were more than 40 races that day and some of them looked, to my eye, like gifts.
So I did something I'm not proud of, and I had a few bets of my own, the old-fashioned way, with the form guide and a strong coffee.
That evening I asked Mark to show me why the model had passed on the races I'd fancied.
The going had changed overnight at 2 of the courses, and the one race I'd been certain about had 4 horses the specialist models couldn't agree on.
The software had seen all of that before breakfast, while I'd seen a sunny day and a busy card and got carried away.
I haven't placed a bet of my own since, and I don't miss it one bit.
This isn't a comfortable thing to say, but the traditional tipster model is becoming obsolete.
Not because tipsters aren't knowledgeable, many of them are deeply knowledgeable, but because the volume and speed of the information environment they're working in has changed beyond what any individual can keep up with.
A human tipster can cover a handful of meetings properly and keep a few dozen horses they know well in their head, but that's about the limit.
But they can't hold the entire British racing card in their head simultaneously.
They can't follow overnight market moves while also building pace maps, cross-referencing trainer habits, checking fitness and digging through speed data, and they certainly can't do it every day in real time.
Put the two side by side and it's no contest, because the bot scans every UK meeting in seconds while a tipster might manage 3 or 4 in a whole morning.
The speed alone is a huge advantage, because by the time a human analyst has reviewed one meeting, the model has scanned the whole day's racing and ranked every runner.
And unlike a person, the model doesn't have bad days or carry frustration from yesterday's losers into today's selections.
It doesn't get cocky after a good run, and it never talks itself into a bet because it fancies some action.
Every decision rests on the same criteria, applied just as strictly on a wet Monday in January as on the biggest Saturday of the summer.
That steadiness is something no human tipster can offer, and it's the foundation of everything Senator Bet Bot delivers.
I say that as someone who spent 20 years trying to be that tipster for myself, and never quite managed it.
We opened Senator Bet Bot to members gradually rather than all at once, because I wanted to see how it performed for different people with different accounts and staking habits before I let it grow.
What struck me most in those early months wasn't the profit figures, it was how different following AI-generated selections felt compared with following a human tipster.
That wasn't because they were winning every bet, because they weren't, but because they could see the system had a steady process behind it.
When a loser came along, it didn't feel random or chaotic, just a calculated risk that hadn't come off this time and would come off more often than not.
That psychological shift, from gambling to systematic investing, is something I didn't fully anticipate when we built Senator Bet Bot, but it's become one of the things members mention most.
One member, a retired signalman from Cumbria, told me he'd stopped watching the races altogether and now just checks the results with his evening cup of cocoa.
Another, a mum of 3 from Lincolnshire, places her bets while the kids eat their cereal and then forgets about them until the school run is over.
I love hearing that, because it's the opposite of how betting used to feel for me.
By limiting selections to 2 or 3 per day, your betting pattern looks natural and measured rather than like someone running through a long list of tips.
Bookmakers keep an eye on accounts that show erratic, high-volume patterns, and Senator Bet Bot keeps you well clear of that.
The low number of selections also keeps your accounts in good standing, so you'll rarely have trouble getting on at the advertised price.
There's a practical wisdom to it that goes beyond just finding winners.
"I've been betting since before decimal currency came in, and this is the first summer I've finished with more in the account than I started with. My only complaint is that the lads at the bowls club keep asking why I'm in such a good mood."
Keith Rowbotham, Grantham
"Before this I had 4 tipster subscriptions and a kitchen drawer stuffed with losing slips. Now I get one email a day, and my husband has gone from rolling his eyes to asking me what the software fancies."
Lesley Farnsworth, Barnstaple
"As a retired electrical engineer, what won me over was reading how the software is tested on races it has never seen, because that's exactly how I'd have built it myself. I've followed every selection since June, and the only surprise has been how pleasantly dull the whole thing is."
Martin Coles, Ludlow
"My son signed me up for my birthday in July and I told him it was a waste of £20. I've since had to buy him a very expensive dinner and admit, in front of his mother, that I was wrong."
Derek Ollerton, Whitby
Every morning, the model runs its overnight calibration cycle.
It pulls in the previous day's full results data and runs it back through the system.
Any selections it got wrong are dissected.
The model identifies which variables it had weighted incorrectly and adjusts accordingly.
It scans every race on the British card and passes on most of them entirely, either because there's nothing clear enough to bet on or because the conditions don't match what it's been trained to target.
The races where it finds a clear, measurable case go on a shortlist, and then the shortlist gets filtered again.
By the time a selection makes it into my morning email, it's passed through multiple layers of scrutiny, not just a single probability calculation.
Put simply, the overnight data comes in, the model recalibrates, the full card is scanned, the best bets are picked out and the email goes to members.
The whole cycle starts at 1 o'clock in the morning and is usually finished well before 6.
I read through every selection over breakfast before anything is sent, not because I overrule the software, but because I like to understand what it has seen.
In nearly 2 years of doing that, I've never once changed a selection, and on the few occasions I've been tempted, I've been glad afterwards that I didn't.
I place the bets at the best available odds, and then I send one email to all members first thing in the morning, with everything they need to get their own bets on.
You don't need to interpret anything or do any research, you just read the email and place the bets.
There's no dashboard to log into and no software for you to run.
Once you've acted on the email, the program's work is done for the day.
Each selection comes with the course, the race time, the horse's name and the price I took, so you know exactly what you're aiming for.
If a price has shortened a little by the time you get on, that's normal, and the staking guidance inside the member area explains how to handle it.
To get started you'll need a modest betting bank, and somewhere between £50 and £100 is plenty, using simple level stakes to keep the risk low while you find your rhythm.
As your bank grows, you can raise your stakes gradually as a fixed percentage of your total funds, and that's the approach that turns steady monthly profits into a meaningful second income over time.
I want to be straightforward with you on this.
Senator Bet Bot isn't a get-rich scheme and it's not designed to be.
The people who get the most from it are the ones who treat it as a steady, systematic way to build a reliable second income.
The system picks between 2 and 3 races a day, and some days it picks nothing at all.
In a typical month, you're looking at around 40 to 48 bets in total.
Not every bet wins, and there will be days when every selection gets beaten.
What matters is that the winners outweigh the losers by a big enough margin to produce a profit every month, and since January 2025 that's exactly what's happened.
You don't need to understand how the model works or be able to interpret form data yourself.
All you need is a betting account and a couple of minutes each morning to place a simple bet.
Everything else is handled by the system.
I'd like to be clear about the kind of person this isn't meant for, because it saves us both time.
If you want 10 bets a day and a thrill every half an hour, you'll find Senator Bet Bot painfully slow, and you'll probably get bored and start adding bets of your own.
If you're hoping to turn £50 into a house deposit by Christmas, this isn't it, and nothing else is either.
If you can't resist adding your own fancies to the list, you'll undo most of the good work the software does.
And if you're betting with money you can't afford to lose, please don't join, because even the best-built system has losing days and I don't want anybody lying awake over one.
The people who do well with the system are patient, and they're happy to let it do its job without interfering.
If that sounds like you, you'll fit right in.
People sometimes ask why I'm still teaching when the software does so well.
The answer is that I love it, and I've no intention of giving it up while the school will have me.
Running the system takes about 20 minutes of my time each morning, and that fits neatly before registration.
It also explains the price of joining.
I'm not trying to make a living from membership fees, because I already have a job I enjoy and a system that pays me far more than any fee would.
The £20 is there to make sure the people who join have made a real decision to follow the selections, rather than signing up on a whim and forgetting all about it.
The cap on numbers is there because the software is worth far more to me working properly for 50 extra people than working badly for 5,000.
My Year 11 group have no idea any of this exists, and I'd like to keep it that way, so if you ever find yourself in Chesterfield, please keep it under your hat.
Right now, we're taking on 50 new members before closing the doors for a while.
The reason we limit numbers isn't just marketing talk, it's practical.
When too many people follow the same selections into the same markets, prices shift.
The price that made a bet worth taking gets squeezed by the weight of money chasing it.
We keep the membership tightly controlled to protect the quality of the selections for everyone inside.
The price to join is £20, a one-time fee with nothing recurring and no monthly charges.
That £20 gets you full access to the Senator Bet Bot member area and the daily email with every selection.
You'll also get full details of the staking plan, plus direct email access to me if you have any questions about managing your betting bank.
I want people inside who are committed enough to act but not deterred by a price that's out of reach.
The first winning bet will cover the fee and then some.
Join today and follow the selections for the next 30 days, and if you're not happy with what you see, email me and I'll refund your £20 straight away, without a single awkward question.
You're not risking anything, because you're stepping into a system that has made a profit every month since January 2025, with a full safety net underneath you.
When 50 memberships are gone, this page closes.
The next opening won't be at this price.
If you've read this far, you already know this is different.
The question is whether you're going to act on that or let it go.
The selections for today are loaded and ready.
If you join before midday, you'll receive today's tips in time to place your bets.
I've been doing this long enough to know that most people who arrive on a page like this have been let down before.
They've followed tipsters who had a good run and then fell apart, or paid for services that went missing the moment anyone asked a hard question.
I understand that completely, because the betting tips market has a patchy reputation, and not without reason.
What I'd say is that Senator Bet Bot isn't asking you to take a leap of faith.
It's asking you to risk £20, with a full guarantee behind it, on a system that has made a profit every month since January 2025 and can show you exactly why.
That's a very small ask for what's on the other side of it.
The model is running better in the autumn of 2026 than at any point in its history, and the selections have never been more carefully thought through.
Grandad Walter spent 40 years filling a notebook by hand on a market stall in Chesterfield, and this is what that notebook grew into.
I'd like to think he'd approve.
No, and it still makes me laugh.
Mark has never placed a bet on a horse in his life, and he tells me he has no intention of starting now.
What he does do is read the overnight logs every morning, looking for anything the software did that he can't explain.
If he finds something, he won't rest until he understands it, and a fair few of the improvements I've described on this page began with Mark frowning at his laptop over a slice of toast.
I'm afraid not, because the software doesn't work as a question-and-answer service.
It doesn't look at horses one at a time, it looks at whole races and only forms an opinion when the whole picture is clear.
If a horse you like isn't in the morning email, that's your answer, and the program's silence usually means it saw a reason to stay away.
It's happened twice, once in 2024 and once last winter, when one of our data suppliers had a technical fault in the small hours.
The software is built to notice when a feed is missing or out of date, and when that happens it refuses to pick anything at all.
On both of those mornings, members got an email explaining there would be no bets that day.
I'd much rather send that than a selection built on half the facts.
Partly because the list would run to several pages and make for very dull reading.
Mostly, though, it's because a handful of those inputs are things Mark and I worked out ourselves over many months, and I'd rather not hand them to anyone with a spreadsheet and a free afternoon.
I've described the main layers on this page, and members get a fuller explanation inside.
Show them the guarantee first, because a £20 fee you can get back in full isn't much of a con.
Then show them the email itself, which is usually no more than a few lines about horses and races.
A good number of members tell me their partners went from rolling their eyes to asking what's running today within a month or so.
Chasing the price is the big one.
When a selection shortens before they've got on, some people panic and back something else in the same race, or double their stake on the next bet to make up for it.
Neither of those helps, and both undo the discipline the software was built to provide.
If you miss the price by a long way, the best move is to let that bet go and wait for the next morning's email.
Yes, and a surprising number of members do exactly that.
The email lands early in the morning and most races don't start until the afternoon, so there's usually plenty of time to place your bets after a shift before you go to bed.
Plenty of nurses in the membership do just that, along with a couple of bakers who start work at 3 in the morning.
Racing throws off a huge amount of detailed information that most people never look at properly, from pace and fitness to the way trainers behave through the seasons.
Football markets are watched by millions of people and priced with enormous care, whereas a Tuesday afternoon handicap at a small track gets a fraction of that attention.
That gap in attention is where Senator Bet Bot does its best work.
Nobody can promise that any method will work forever, and I'd be wary of anyone who tells you otherwise.
What I can tell you is that the model retrains every night and runs a post-mortem on every loser, so it adapts as racing changes rather than standing still.
It's a very different animal today from the version that went live in 2024, and I expect it'll be different again by this time next year.
Yes, when the evidence points that way, although it happens less often than you might think.
It has no bias for or against short prices, because it doesn't know or care which horse the newspapers have tipped.
What it won't do is back a favourite simply because it's the favourite, and a good number of the races it passes on are ones where the market leader looks weaker than its price suggests.
That's fine, because there's nothing recurring to cancel and nothing to pause.
Your membership stays open, and when you're ready you just pick up the next morning's email and carry on.
We're into the first days of October, and the racing calendar is at one of its most interesting turning points.
The flat turf season is heading towards its finale, while the jumps season gathers pace week by week.
It's a period that tests any betting method, because the ground changes quickly and plenty of horses are running on tired legs after a long summer.
That's exactly where the fatigue layer and the going-reader come into their own, and October 2025 finished at £4,930.30 even as the form was turning upside down around it.
May 2026, at £7,085.10, is still our best single month on record, and April wasn't far behind at £6,730.40.
I won't make predictions about October, because the bot doesn't make predictions about months, only about races.
If you're going to join Senator Bet Bot, the start of a new season, with the software in the best shape it has ever been in, is the time to do it.
If you join in the next few hours, you'll receive today's bets alongside every other member who's already inside.
P.S. The 50-place cap is firm.
I checked the member list this morning and spaces are going faster than I expected at this stage.
If you come back tomorrow and the page is gone, you'll know why.
Today's selections are already queued and ready to go out with the morning email.
For £20, with a full money-back guarantee behind you, there's very little reason to wait.
I'll see you on the inside,
Tom Cavendish