OTB #036: Why (I Think) AI-Powered Postflop Strategies Are The Future

When I first got my hands on PIOSolver, a postflop GTO solver tool, I thought asking it solve for four bet sizes would give me a great idea of what the 'optimal' strategy would look like.

I used:

• 25% pot

• 50% pot

• 75% pot

• 125% pot

The main issue with having four bet sizes is that it isn't particularly easy to work out what's going on.

In this spot, 40bb HJ vs BB single raised pot on a 7d 4s 2s flop, I can see that 25% pot and 125% pot are used most frequently here, but 51% pot does get used sometimes and 75% a little bit too.

Is 2 better than 4?

Now I've always been an advocate for and strong believer in having a two-bet size strategy post flop.

This allowed me to see when I should use a big bet or a small bet on the flop and turn and this was manageable for my mere mortal brain.

This two bet size approach allowed me to learn actionable heuristics on different flops and turns. For example, I learnt that a small bet made sense on paired boards in position against the big blind. And I learnt that I should use big bets quite a lot when I have a nut advantage and an equity advantage, like on ABB boards like AKJ and AQT.

So then I went with two bet sizes - a small bet and a big bet.

You can see that it's already easier to visualise what's going on with fewer bet sizes. Here's the same spot, but with only 25% pot and 69% pot (the geometric size) sizes:

But I always wondered if AI could work out the optimal "one-size" because that would be even easier to implement in game...

Enter GTO Wizard AI

This time I used the AI to automatically simplify and optimise my bet size for this spot.

And this is what it came up with:

Out of all three examples, which do you think is easier to visualise what's going on and learn implementable heuristics?

Grab 10% off your first purchase of GTO Wizard by clicking here.

But won't you lose EV?

According to the solver, using 1 bet size loses a little in EV. So if you are able to implement 4 bet sizes in this spot you'll gain an extra 0.01bb, which is 1bb/100.

But let's be honest, nobody is able to implement a 4 bet size approach in this spot.

By using 1 bet size, not only is it no worse than using 2 bet sizes, it's much easier to implement in game.

You are losing EV if you adopt a 1 bet size approach, but you won't be able to implement a 4 bet size approach in the first place so you're not capturing all that EV anyway.

And that's why I think AI-powered strategies are the future.

Here's a table of 11 different flop textures and the EVs possible using the different bet size configurations, including what the GTO Wizard AI believes to be the 'optimal' bet size in each spot:

Across the 11 example flops for this spot (40bb HJ vs BB single raised pot), the biggest potential EV loss was 0.02bb or 2bb/100, which is not insignificant.

---

This week's action tip: experiment with reducing the number of bet sizes in your custom post flop sims to make it easier to learn an implementable strategy. Be sure to compare the EV of the strategies to make sure you're not losing too much just to develop a strategy you can actually use in game.

---

What's your goal when you train?

Learn to play like a bot?

Or learn how to play a simplified strategy that you can actually implement?

That's all for this week.

See you next time.