Record rebuilt nightly from the bots' own databases

Tips priced by maths. Results you can audit.

BetSmart runs Poisson, Dixon-Coles and Elo models across 48 football leagues and every UK & Irish race — then posts only the bets where the bookmaker's price is genuinely wrong. Every tip is written to a database before the off and published afterwards. Winners, losers, and the experiments that failed.

Start free on Telegram → Show me the losing bets

No screenshots. No deleted tips. No "guaranteed winners". 18+ — please gamble responsibly.

2,263 Settled football tips

70.1% Strike rate

+89.33u Profit at level stakes

+3.9% Return on investment

Football core record, 11 Jun – 29 Jul 2026. Level 1-unit stakes at advised prices.

Two bots, one method

Pick your sport. Or take both.

Each bot is a separate model with its own database, its own published record, and its own Telegram channel. They share one principle: only bet when the number says the price is wrong.

BetSmart Football

48 leagues, priced every day

Goals, corners, cards and match markets modelled with Poisson / Dixon-Coles scorelines and Elo team strength, de-vigged against the market to find the genuine mispricings.

2,263Settled

70.1%Strike rate

+3.9%ROI

Daily tips split into Safe, Value and Longshot — each with its own published record
In-play alerts on live shots & shots-on-target lines
Best-price shopping across bookmakers before every post
Strategy Lab: backtest your own rules over stored games
Research tools — team form, referee profiles, shot maps, lineups

Get football tips

BetSmart Racing

Every UK & Irish meeting

Form, speed ratings, going, draw bias, trainer and jockey strike rates weighted into a single price, then compared with the board. Only the genuine overlays get posted.

72Settled

30.6%Strike rate

+13.4%ROI

Value bets with a stated fair price and edge on every selection
Plain-English reasoning — the case for the horse, in full
Each-way priced on the best available terms, not the modal ones
Closing-line tracking against Betfair SP
Trainer, jockey, course and pace research on demand

Get racing tips

The part other tipsters delete

Here is everything, including what lost.

These tables are generated by a script that opens each bot's database read-only and counts every settled tip. Nothing is selected by hand. If a category is losing money, it appears here losing money.

Football — by category

Category	Settled	Strike	P/L	ROI
Safe	2,063	70.6%	+10.50u	+0.5%
Value	195	66.7%	+83.83u	+43.0%
Longshot	5	0.0%	−5.00u	−100.0%

Read it honestly: Value is the engine. Safe wins most of its bets but at short prices, so it barely clears the vig. Longshot is 5 bets — far too few to mean anything, and all five lost.

Racing — value bets

Segment	Settled	Strike	P/L	ROI
Value	72	30.6%	+9.68u	+13.4%

72 settled bets is a small sample. A +13.4% ROI over that many selections is encouraging, not proof. Treat it as an early signal and judge it again in a few hundred bets.

The experiment we're not selling. A high-volume player-props model ran alongside the core product: 7,343 settled bets, 34.5% strike, −21.1% ROI. It loses money. It is excluded from every paid tier and shown on the results page purely so the record stays complete.

Open the full live record →

How a tip gets made

Five steps, no opinions.

There is no tipster sitting in a room having a feeling about Brentford. There is a pipeline, and it runs the same way every day.

Model the event, not the market

Football fixtures get a Poisson / Dixon-Coles scoreline distribution built on Elo team strength, recent form and opponent-adjusted shot data. Races get a weighted blend of form, speed figures, going, draw, trainer and jockey strike rates. The output is a probability for every outcome — produced without looking at the bookmaker's price.

Strip the bookmaker's margin out

Bookmaker odds are converted to implied probabilities and the overround is removed with a power de-vig, so the market's genuine opinion can be compared like-for-like with the model's. Without this step every price looks like value.

Calibrate against what actually happened

Raw model output is over-confident on rare outcomes. Each market carries its own calibration built from settled history, so a stated 60% chance means 60% — not "60% according to a formula nobody checked".

Only post a genuine edge

A selection has to clear a minimum expected-value bar before it goes out, and stake size is set by fractional Kelly rather than a flat gut number. Most fixtures produce no tip at all. Quiet days are the system working.

Log it, then live with it

Every tip is written to the database with its price, its fair price and its edge before the event starts, then settled automatically from results. Nothing can be edited after the fact — which is precisely why the losing rows above still exist.

What you actually get

A research desk in your pocket.

The tips are the headline. The tools are why people stay — everything below runs inside Telegram, on demand, no spreadsheet required.

Value scanner

Model-free scan comparing every bookmaker against the field to surface prices that are out of line with the market itself. /value

In-play alerts

Live reads on player shots and shots-on-target as a match develops, priced off the next line above the current count. /inplay

Strategy Lab

Backtest your own rules over stored games and see the hit rate and break-even price before you risk anything. /strategylab

Team & player research

Form, goal-timing profiles, shot maps, home/away splits and head-to-head records pulled on demand. /team /player

Referee & lineup reads

Bounded card expectations by official, plus lineup-aware pricing that adjusts when the teamsheet lands. /ref /lineup

Racing form tools

Trainer, jockey, course and pace analysis, plus a followed-horse tracker that pings you when your horse runs. /horse /tracker

Closing-line proof

Every tip is compared with the closing price — the single hardest test of whether a model has real edge. Published openly. /proof

Your filters, your feed

Set minimum odds, pick leagues, mute markets. Every delivery path respects your preferences instead of spamming everything. /filter

Bet builder & accas

Correlation-aware multiples using a Gaussian copula, so a same-game combination is priced properly instead of naively multiplied. /builder

Why believe any of it

Four things most tipsters won't do.

The record is generated, not written

A script opens each bot's SQLite database read-only and counts every settled tip at level stakes. There is no content-management step where a human chooses which results to include.

Losing segments stay published

The longshot category is at −100% ROI and the player-prop experiment is at −21.1%. Both are on this page. A record that only contains winners is a marketing asset, not a record.

Judged on the closing line

Beating the closing price is the test that separates a real edge from a hot streak. Where the closing line was never captured it is stored as unknown — never quietly counted as a zero.

Sample sizes are stated, not hidden

Every figure on this site carries the number of bets behind it. A 30% strike rate over 72 bets and one over 7,000 are very different claims, and we label which is which.

What this is not. It is not a guaranteed income, and the record above covers roughly seven weeks — a short window in betting terms. Variance is real, losing months happen, and a model that worked last month can stop working. Bet only what you can afford to lose.

Pricing

Start free. Upgrade when the record convinces you.

You can read the entire public record and use the research tools without paying anything. That is deliberate — the numbers should do the selling.

Free

£0

No card. No trial timer.

Full public track record and closing-line proof
Value scanner and research tools
A sample of each day's selections
Results and settlement notifications

Open the bot

Both sports

£34 /month

or £279/year — about 2 months free

Every football and racing selection, in full
In-play alerts and steam moves
Strategy Lab, bet builder and multiples
Personalised filters across both feeds
Priority support

Subscribe to both

Single sport

£24 /month

Racing only: £19/month

Every selection for your chosen sport
Full reasoning on every tip
Research tools and filters
Results and P/L notifications

Football — subscribe Racing — subscribe

Cancel any time from Telegram. Subscriptions are billed by MyMembers. Prices in GBP and include VAT where applicable.

Questions

The things worth asking.

No, and it's worth being precise about that. There is no neural network and no language model picking bets. BetSmart uses established statistical methods — Poisson and Dixon-Coles scoreline models, Elo ratings, market de-vigging, per-market calibration and fractional Kelly staking. That is a narrower claim than "AI", and a more honest one: every number has a method behind it you could look up and check.

Because losing categories are still on the page. The results file is produced by a script that reads the bots' databases read-only and aggregates every settled tip — there is no step where anyone chooses what to include. The longshot category sits at −100% ROI and the player-prop experiment at −21.1%. Those would be the first two things a dishonest operator deleted.

Because most of that volume is short-priced "Safe" selections. Winning 70.6% of bets at around even money barely beats the bookmaker's margin — that segment returned +0.5%. The profit is concentrated in the Value segment: 195 bets at a 66.7% strike and +43.0% ROI. A high strike rate is not the same thing as a profitable one, and anyone selling you the first number without the second is misleading you.

It varies with the fixture list and how the prices land. Some days produce a handful, some produce none. The model only posts when a selection clears its expected-value bar, so a quiet day means no edge was found rather than that something broke. You can set minimum odds and league filters so your feed stays the size you want.

The closing price is the market's final, sharpest opinion. If you consistently take a bigger price than the one available at the off, you were probably right for a reason rather than lucky. It's the standard test professional bettors use, it shows up long before profit does, and it's much harder to fake than a win-rate screenshot.

Yes. Straightforwardly, yes. A positive expected value over thousands of bets still contains long losing runs, and the published record covers about seven weeks — not long enough to promise anything. Bet only with money you can afford to lose, stake consistently, and stop if it stops being enjoyable.

From the subscription link in the bot, in a couple of taps. There's no retention call, no cancellation form and no minimum term. Access continues to the end of the period you've already paid for.

Ready when you are

Read the record first. Subscribe second.

That's the right order, and it's the one we'd want if we were you.

Open the bot — free Inspect the full record