Service Updates

2025-02-20

Advised Stakes feature added

• Before sending an alert, our model returns an expected value figure that correlates strongly with ROI
• Therefore to optimise profit and ROI, we will now be introducing an 'advised stake' of 1, 2 or 3 units
• The higher the advised stake, the stronger the expected value is in the model and the higher profits and ROI should be
• For tracking purposes on the website, all alerts sent prior to these feature addition will be unchanged and kept at 1 unit per alert
• From 2025-02-20 17:30:00, the advised stake and profit will be live for alerts and tracking
• For example if on 2025-02-21 there's 5 x 1 unit alerts, 4 x 2 unit alerts and 1 x 3 unit alert, this will show up as 16 units staked (rather than 10 units previously). Profit will be calculated from the advised stake also
• Once the figures build up, we will also add a breakdown by stake size on the website to track the different 1, 2, 3 unit results

New Cumulative profit charts added

• Improved visuals for cumulative profit over time on Live Tracking, SBC Bucket Trials (static), SBC Bucket Trials 2 (ongoing) tabs

New columns added to tracking

• £20 and £50 per unit staked columns added to all of the live tables to illustrate potential returns with different stake sizes

Minor website bug fix

• Fixed minor bug whereby 'SBC Bucket Trials' tab was continuing to accumulate results beyond the cut off date

2025-02-13

Buckets reformat to 24/7

• After analysing member feedback in the recent survey, we have now changed the buckets to 4 x 24/7 buckets to replace the 4 x AM1, AM2, PM1, PM2 format
• Instead of being in an 'AM' or 'PM' bucket, everyone will now be in a bucket that receives alerts 24/7
• Hopefully this will provide customers with more flexibility of when they can follow the channel
• The new buckets will automatically replace the old buckets in the same telegram channel (now renamed)
• The volume of alerts will still be the same as your previous bucket (25% of all alerts), however they will now be spread more evenly over the 24 hour window rather than receiving a more concentrated sample in one half of the day

Tracking results on the website

• A new tab 'SBC Bucket Trials 2' has been added to the website that will track individual A,B,C,D results as well as 'All'
• 'SBC Bucket Trials' tab will remain there to view results from the original AM1, AM2, PM1, PM2 and 'All' buckets
• 'Live Tracking' contains results from all alerts since going live / SBC management have been tracking that is a good summary pre and during all bucket trials
• 'Backtest' contains the backtest results to reference for expected results in the long run

Goals spike update

• The goals increase has remained consistent since it began on 17th January
• This suggests that whatever caused it is a permanent change, rather than a regular game patch update
• The model was altered a few weeks back to reflect this and volume, over/under distribution and results have since returned to expected
• We are keeping a close eye on this, but we don't expect any of the above to change or any further disruption to the service

2025-01-31

• Model monthly retraining complete
• Active as of 2025-01-31 17:09:00

2025-01-28

• Minor bug fixed where website tracking results were resulting on the opening line and odds rather than alerted line and odds
• Only a small sample of games affected where odds moved between opening and alert time

2025-01-27

Update on goals spike

• The spike in goals hasn't come down at all over the last 9 days since it started
• Bet365's lines are slowly creeping up but their mean goal lines are still lower than the mean actual goals every day in the last 9
• Last week the team has extensively been looking into previous occasions since the data started in 2020 when we've seen these spikes and this scenario (a spike in goals that hasn't returned down within 1/2 days that is consistent over every group of players). The conclusion was that this is unchartered territory in terms of a sustained significant change in goals since the data started
• Whatever has happened, the change is even more extreme (the average goals across thousands of games since 17th Jan has gone up by around 1.5 goals) than anything in historical data, including when they switch over the whole FIFA game each year (e.g switching from fifa 24 to fifa 25). Possible explanations could be changing console that they are playing on, an extreme change in game settings, or an extreme patch update - but there's no real way of knowing
• Whatever it is also doesn't matter too much as we have been creating tracking metrics and alternative strategies this past week to keep on top of any change back or other significant changes in the future

Under the hood

• For those of you who are interested in understanding a bit more the following screenshot is a good summary of what's been going on:

Glossary:

• ft_goals = full time goals
• goal_line = bet365 asian goal line
• diff = mean(ft_goals) - mean(goal_line): Positive values = actual goals > b365 lines
• over_roi / under_roi = roi on betting on all overs / unders selections for every match (nothing to do with alerts)
• AVG(preds) = The average difference between model predicted goal expectancy and b365 goal line (as you can see past 9 days huge skew towards model predicting unders as it's still using some data from before this change to make a prediction)

What we've been anticipating and waiting to see (which happens in the regular cycles) is:

• mean_ft_goals coming down over time
• mean_goal_line going up over time
• AVG(preds) remaining negative but less extreme

This is usually the perfect trifecta that leads to a big period of unders profit that we described last week. This may still happen, but for now as none of those 3 things are moving in that direction, and the change is something that the data hasn't seen before, we've deemed it sub-optimal to continue predicting and making tips that are using any data pre 17th Jan.

Temporary Model / Strategy Alterations

From Jan 26th:

• all of the data from 17th Jan has been changed to effectively a 'new FIFA', meaning no data from before this is currently being included in the model input
• EV threshold has been temporarily increased for now, whilst there isn't much data on this 'new FIFA'
• This will result in fewer bets (temporarily) but a more even combination of over and under bets
• This should reduce volatilty seen over the last 10 days
• We will be keeping a close eye on all of the metrics above and others that we are now tracking and have tweaks and strategies ready for each scenario
• We expect profit to be more stable and bet count will increase again over coming weeks
• Live Tracking tab and SBC Bucket Trials tab on the website follows real time model, strategy and alerts that were sent. This has now all been updated to reflect any changes over the last week.

Message to customers

• We appreciate this has been unfortunate timing to start the trial. January looks like it will be a blot on our perfect record (live and backtest) of never having a losing month.
• More importantly, we certainly hope that customers haven't made any significant loss during this time as this is exactly the opposite of the consistent service we are looking to provide.
• The high turnover of unders alerts for a few days that were losing is definitely not a regular occurrence and has never happened before in the data without an even bigger profitable patch following it. We are also now well placed to ensure this doesn't happen again in the future.
• Following this abnormal period, the service is running at +2.2% ROI at the time of writing for all buckets combined. This falls below the ROI level and consistency that we expect to deliver going forward.
• We expect to return to 4-7% ROI every month from Feb onwards.
• Once the turnover also increases this should lead to high levels of 'unit profit' that can be very lucrative to individual customers.
• We will be keeping the service free until the figures return to where we expect them to be.

2025-01-20

Some of you may have noticed that the bet count (mainly on unders) has been through the roof over the weekend. This is due to a huge spike in average goals across the Battle tournament from 17th Jan onwards. Please see goals table:

We just wanted to reassure everyone that whenever there is a sharp increase or decrease in turnover it gets looked into in detail at our end to check that it's all part of an expected cycle. This instance has been looked into over the weekend and the model is performing as expected, with the explanation below.

Due to the significant spike in goals from the 17th Jan, Bet365 have increased their goal lines accordingly and the model is generating high value on unders across the board. This is because machine learning has taught the model that 'reverting back to the mean' outperforms 'following the spike' historically.

This particular spike is extreme and is still currently on the rise. It's likely that a patch update on the FIFA game they are playing on is the cause for this. In this scenario even minor updates to gameplay can cause these players (who play everyday as their job) to need an adjustment period to improve their defending post patch update. This is fairly normal and the backtest results factor in these sorts of changes and cycles.

What we anticipate to happen next (based on what's happened in the past) is that B365 will continue to increase their lines and the goals will start to revert back down to a new mean (once defending improves) which should fall under b365's new lines.

If you check the website, you will see that the model lost heavily on the 18th Jan. This is because B365 adjust more heavily to short term results than our model does (which is the foundation of how it makes profit - using a more effective balanced lens than just short term results). Despite goals still being on the rise since then, the model broke even on the 19th but is so far making loss again on the 20th (due to another big spike in goals - see attachment). The losing period correlates with a very extreme short term goals spike. It's unclear exactly where we are in this cycle - but it's rarely longer than a couple of days. The good news is that whilst goals continue to spike, B365 lines will continue to rise. The more extreme the spike. the higher the lines rise. The longer the spike and losing period, the longer the lines will stay inflated when the goals revert back to a new mean and the longer winning period should follow. Hence why the model always wins even in these volatile periods.

This follows historical cycles of a heavy losing day or two, followed by a period of breakeven (stabilising - our model adjusting to a big change moving in one direction vs b365 overreacting in the other direction) and this is often then followed by periods of heavy winning on the same selection (in this case unders).

This is because we anticipate: a) b365 lines will continue to rise too far as goals are still increasing currently, b) actual goals will revert back down to a new mean and c) our model will be perfectly placed to take advantage of both a and b happening simulataneously.

We thought it was important to explain this now, because we understand that new customers may be seeing lots of unders bets on games that are ending 7-4 and 5-5 etc causing a loss in confidence. The reality is though that history suggests that there could be a period of much heavier profit (on unders) to follow.

We would suggest keeping an eye on the website and the profit closely if you are betting at the moment. You won't go wrong continuing to follow every alert, but if you want to reduce volatility you could adjust stakes or volume based on the understanding of this losing->stabilising->winning cycle, of which we're in an extreme one right now and we believe somewhere between losing and stablising based on each player grouping (E.g the players who have already played a few days since the spike, lines are already getting too high. Other groups of players who played less over the weekend are spiking now).