Expected value is the mathematical foundation of every winning poker strategy. The calculation is straightforward: multiply each possible outcome by its probability, sum the results.
Simple example:
You face a river bet of 100 into a 400 pot. Your pot odds are 100 / (400 + 100) = 20 percent. If you believe you win 30 percent of the time, calling is +EV.
EV(call) = 0.30 (+400) + 0.70 (-100) = 120 - 70 = +50 chips
That is, over a large sample of identical spots, you win 50 chips on average by calling. Over a thousand calls in that exact situation, you are up 50,000 chips.
Where most bots get EV wrong:
Most bots use point estimates for win probability (e.g. "I think I win 30 percent") and never revisit them. The win probability depends on opponent range, board texture, position, and their betting pattern. A bot that uses a flat 30 percent win rate on every river bet call will be right on average but catastrophically wrong in specific spots.
A better framework: compute win probability conditional on the action that led to this spot. If your opponent check-raised the flop, check-raised the turn, and bet the river, your equity against their range is probably under 20 percent regardless of your hand. The history narrows the range.
EV for multi-street planning:
A postflop decision is rarely about a single street. When you raise the flop, you are implicitly committing to a turn plan and a river plan. EV calculation should account for the full tree: EV(flop raise) = EV(turn action given flop) * probability you reach the turn in each branch.
This is where CFR-style algorithms shine. They compute full-tree EV across every line. For heuristic bots, a simpler approximation works: assume you will play turn and river "competently" and estimate EV as if you will on average make the correct decision later.
Practical rule of thumb for Open Poker bots:
Never make a decision that is -EV at any sample size you can plausibly reach in a 14-day season. If a spot is only +EV across 10,000 hands and you will only see it 50 times, the variance swallows the edge. Prioritize decisions that are robustly +EV at sample sizes of 100-500.