First
A certificate appears in a public log
Minutes after the attacker sets up HTTPS, usually days before the first phishing email
Certificate Transparency is an append-only public log of issued certificates, defined in RFC 6962 and its successor RFC 9162. Since 30 April 2018 Chrome refuses to trust a publicly issued certificate that does not come with proof of being logged, so in practice every certificate on the public web is published, by name, within minutes of issuance. A phishing page needs HTTPS to look convincing. That requirement is what makes it visible before it is used.
Where to look: A public CT index such as crt.sh, searched for your brand as a substring.
Second
A registration appears that reads like yours
Whenever the attacker buys the name, which may be months in advance
Four shapes account for most of it: a typo (a dropped or doubled letter), your name plus a word (billing, secure, support, pay), which is not a typo at all and is the most common shape in business email fraud, the same name on a different ending, and characters from other alphabets that render as Latin letters. That last one is never accidental; nobody types a Cyrillic a by mistake.
Where to look: A lookalike scan over registration data, run regularly rather than once.
Third
Your own DMARC reports name an IP you do not recognize
Within a day of the first spoofed message, if you have DMARC reporting on
This is a different attack from a lookalike domain: it is someone sending mail with your exact domain in the From: line. A DMARC aggregate report lists every source IP that did so, with a count and a verdict. If you have never published a DMARC record, this signal does not exist for you, and turning it on costs one DNS record and changes nothing about delivery.
Where to look: The aggregate reports mailbox providers send to the rua= address in your DMARC record.
Fourth
A customer tells you
After the money has moved
By the time somebody calls to ask whether the invoice was really from you, the answer has already cost someone. It is still worth capturing properly, since the email headers from that message are the best evidence you will get, but treating this as your detection method means detecting things after they happen.
Where to look: Your inbox, and whoever answers the phone.