Most amateur games are not lost to deep strategy. They are lost to a missed fork, a piece left hanging on b5, a back-rank mate that nobody saw coming until it arrived. Which is good news, because it means the fastest available rating is also the most trainable thing in chess.
But “do more puzzles” is not a method, it is a shrug with a number attached. I went from roughly 2300 to nearly 2600, and not by grinding random positions. Here is what the training actually looked like.
Tactics are two skills, not one
When somebody says they are bad at tactics they usually mean one of two quite different things.
Recognition is seeing that a shape is even on the board: that this is a fork position, that the queen and king are lined up for a pin, that the back rank is soft. Calculation is the separate business of working out the exact move order and checking it survives the reply you would least like to see.
They improve in different ways, which is why people who train only one stall. Recognition comes from volume and repetition, from seeing the same patterns hundreds of times until they announce themselves. Calculation comes from making yourself read lines to the end before you move, which is slow, unpleasant and the only thing that works.
The method
Learn the patterns by name first. You cannot spot what you have no word for, and the naming is not pedantry: a player who knows what deflection means starts seeing deflections within a week. Sit with the core motifs and the mating patterns until each one carries a clear picture.
Then solve at the edge of your ability, which in practice means a solve rate somewhere around three in four. Comfortable puzzles teach nothing and feel wonderful, which is exactly why so many people live there. Puzzles four hundred points above you are not training either, they are just a slower way of guessing.
Calculate fully before you commit. This is the single biggest habit change available to most club players and almost nobody makes it. Pick the most forcing candidate, follow it to the end, check the best defence rather than the convenient one, and only then play the move. If you have to name the mating square out loud to prove you saw it, do that.
Review every miss, properly. Not a glance at the solution and on to the next one. Ask why: wrong pattern, or right pattern and lazy calculation, or a defensive resource you never considered? Those have different cures, and the answer is the only thing of value a failed puzzle produces.
And repeat with spacing. You will forget a pattern you met once, whatever it felt like at the time. Meeting it again four days later is what moves it from “I understood that” into “I saw that instantly”, and there is no shortcut around the calendar.
The mistakes that stall people
Solving only what you can already do, which is pleasant and useless. Rushing, which builds speed on top of a foundation you never poured. Never reviewing, so that fifty solved positions teach less than fifteen understood ones. And ignoring your weaknesses, which is the most human failure on the list: if you keep missing deflections, the answer is three sessions of nothing but deflections, not another month of general practice while quietly hoping.
A week that works
Fifteen focused positions a day, calculated to the end. One weak theme drilled deliberately two or three times a week, picked from your worst accuracy score rather than your favourite. A couple of minutes on anything you missed, the same day you missed it.
Not a marathon. A tight, honest loop, repeated until it is boring, and boring is roughly the point at which it starts working.
Most of that loop happens inside whatever tool you solve in, and the defaults are rarely set up for it, so it is worth reading how to use a chess tactics trainer properly before you settle into a routine. If you want the loop assembled for you, with the difficulty handled and the engine showing you exactly why the move failed, that is what Chess.PhD is built for.