Refactor for Performance and Readability

Persistent is running an account migration for a client moving off a legacy CRM. A helper function takes a batch of account records pulled from the old system and a blacklist of account IDs that should be skipped, and returns the IDs that were migrated. It works on a demo dataset of fifty accounts, but on the client's real export of around two hundred thousand accounts the migration job that used to take minutes now runs for hours and someone eventually notices the result looks wrong on a second run.

processed = []

def migrate_accounts(records, blacklist):
    global processed
    result = ""
    for r in records:
        skip = False
        for b in blacklist:
            if r["account_id"] == b:
                skip = True
        if skip:
            continue
        acc = {}
        acc["id"] = r["account_id"]
        acc["name"] = r["name"]
        acc["region"] = r["region"].upper()
        if acc not in processed:
            processed.append(acc)
        result = result + acc["id"] + ","
    return result

Refactor this function to improve both its performance and its readability. Explain what's wrong with the current version and what you would change.