That measurement problem is structural rather than a flaw in any particular platform. Attribution assigns the sale to the last advertisement seen, and retargeting is by definition shown late in the process to people already considering you. The honest way to evaluate it is a holdout: exclude a portion of your audience from seeing the advertisements and compare their conversion rate to those who saw them. Most businesses never run that test, and the ones that do frequently find the effect smaller than reported.
There is a second problem specific to first year businesses, which is audience size. Retargeting requires enough people to have visited for a platform to build a usable audience, and most platforms enforce a minimum. A site receiving a few hundred visitors a month will either fail to build one or will show advertisements to the same small group repeatedly, which is the condition that produces irritation rather than sales.
Where it genuinely works is a specific situation: a considered purchase, a defined moment of abandonment, and enough traffic to matter. Somebody who added items to a cart and left, or who spent time on a pricing page and did not enquire, is a real signal and reaching them again is reasonable. That is narrower than the general practice of following everybody who ever loaded your home page.
If you run it, cap the frequency deliberately and set a short window. Advertisements following somebody for weeks after a single visit produce resentment and are the reason people install blockers. A few days, a limited number of impressions, and an exclusion for anybody who already converted covers the legitimate use and avoids the version everybody complains about.
There is also a privacy dimension that is tightening rather than settling. Tracking across sites has become less reliable as browsers restrict it, and the regulatory definition of sharing personal information in some jurisdictions covers exactly this kind of arrangement. Building a plan that depends on it working the way it does today is taking on a risk you do not control.
One narrower use is worth separating from the general case, which is retargeting people who reached a checkout or an enquiry form and stopped. That group has a specific unfinished action, the window in which reminding them helps is short, and the message can reference the actual thing rather than the brand generally. It is the one application where the audience is small, the intent is unambiguous, and the results survive a holdout test.