Most EAs are bought on a screenshot and abandoned within three months. The alternative is to treat the purchase as what it is, an assessment of evidence, and to ask the questions in an order that rules out the weakest candidates first.
First, the record
Is there a live record, verified by a third party, from a stated start date, with the equity curve and every trade visible? How long is it, and does it include a losing period? What is the maximum equity drawdown, at what risk setting, and how long did recovery take? If there is no such record, stop here. Everything else is a description of something that has not been shown to exist.
Second, the method
Does every trade carry a stop loss placed at the broker? Is position size calculated from risk rather than entered as a fixed lot? Does the strategy use grid, martingale or averaging into losing positions, under any name? Are the entry and exit rules described in plain language, even if the parameters are not? A "no" to any of the first three, or a description that says only that trades are "managed by advanced algorithms", rules the EA out.
Third, the fit
Which markets does it trade, and does your broker offer them? Which account type does it need? What minimum balance does it recommend at the risk setting you would use, and why? Does it need a specific VPS location, which would mean it is execution-sensitive? Does it trade through news, and does that suit your account and any prop firm rules?
Fourth, the seller
Is the developer identifiable? Do they answer questions about losing periods directly? Do they publish backtests with their assumptions stated, and label them as backtests? Do they issue new "optimised settings" after bad months, which is a sign of fitting? What are the licence terms, and is there a refund policy?
Three candidates. The first has a beautiful backtest, no live record, and 40 inputs. Out at the first question. The second has a verified live record of fourteen months with a 22 per cent equity drawdown, a 91 per cent win rate, and an average loser eight times the average winner. Out at the second, because it is holding losses. The third has a verified record of nineteen months, an 11 per cent maximum drawdown, a 43 per cent win rate with winners twice the size of losers, stops on every trade, and a plain description of three strategies. It also needs two indices your broker does not offer. Change broker, or accept a partial book. That is the decision left.
The order matters because each stage costs time and the earliest questions eliminate the most candidates. A vendor who cannot get past the first two has nothing to sell; the remaining questions are about whether what they have suits you. Most of the work of choosing an EA is done in the first ten minutes, and it is done by refusing to look at anything until the record has been seen.