The market, and the format.
Decide where the product will be sold before anything else. Permitted ingredients, maximum doses, label wording and the documents a customs officer will ask for all follow from the market, and a product built for one region is not automatically sellable in another.
Then the format. Capsules, softgels and tablets each suit different fills, doses and volumes, and the choice decides the equipment, the shell or tooling, the packaging and the price. It is the decision every other one depends on, and it should be made on the formula, not on a preference.
The materials, and their paperwork.
Specify every raw material and excipient before you price the product: grade, origin, particle size where it matters, and the certificates you will need for the market you chose. Two materials with the same name can behave differently on the line and cost differently at the supplier.
Decide who holds the documentation. Certificates of analysis, specifications, safety data sheets and allergen statements arrive from different suppliers at different times, and a product without a complete file is a product that cannot be listed, shipped or audited.
Ask for the list at the start. Given the product and the markets, what documents will exist, who produces each one and when. Written down once, it stops being a scramble at the point of a first order.
The batch size, and the line.
Decide the first batch size on what you can sell, not on what the machine can make. A smaller first run costs more per unit and less in total, and it buys the right to change the formula, the shell or the pack while being wrong is still cheap.
Then decide the line to match: one machine, a few stages or a complete setup, sized to the volume you expect in a year rather than in a month. Capacity that sits idle costs as much as capacity you do not have.
Every one of these decisions can be made with samples in hand: shells, materials, a trial batch on the intended equipment. That is what samples are for, and it is the cheapest week a supplement project ever spends.
