The cheapest way is to use the battery directly. Referring to a typical CR2032 datasheet, look at the discharge curves, and temperature dependence. If your entire circuit draws 190uA or less, you can depend on 800+ hours of life at an end-of-life voltage of 2.75V or so.
Look at the data sheet for your CPU and see if it is GUARANTEED to operate at 2.75V at the clock frequency you are using. Don't use "typical" figures. For example, here is a figure from the Microchip PIC12F1501 datasheet.
This particular part is good down to 2.5V for any clock frequency up to 20MHz, and down to 2.3V for 16MHz or less, over the entire temperature range. It will also easily accommodate the highest voltage from a new CR2032 (a bit over 3.0V at high temperatures).
Then check carefully through the MCU datasheet to make sure that all peripherals will operate over that voltage range. In the case of this particular chip we can select the brownout voltage to be 2.7V nominal or 2.4V nominal. In the first case, we can use any clock up to 20MHz, but we might only get half the battery life (for the same load current) due to the higher brownout voltage. In the second case, we're restricted to 16MHz maximum clock, but we will get 1000 hours battery life.
(Or we could pick the PIC12LF1501, which is guaranteed to operate down to 1.8V (and can accept up to 3.6V) at up to 16MHz, and go through the analysis again).
If you do find you need a boost converter, the NCP1450 is a cheap controller, but it requires an external transistor and Schottky diode. If your circuit happens to draw in the hundreds of uA, you can optimize for efficiency by picking a smaller transistor. For example, below you can see efficiency of 70% at 200uA out, so the drain from the battery should be a bit more than 300uA.
If your current requirement is in the mA, you might want to consider a complete regulator with synchronous rectification such as the AAT1217ICA-3.3-T1, which requires a smaller inductor. It's about 35 cents in quantity. If your factory is in Asia, there are even less expensive choices. The same analysis as for the MCU applies to the voltage range that the regulator is guaranteed to operate at (and can accept without damage).
The efficiency of these things is not great at low currents, however, as you can see here:
At 200uA, it's only about 35% efficient (about 660uA in for 200uA out!), so a lot depends on the actual operating current of your circuit.