The reason it beats everything else is coverage. Analytics sees what happened on your website and nothing else, which for a local business excludes phone calls made from a business profile, walk ins, referrals who searched your name, and anybody who found you through a conversation. Those paths are frequently the majority, and they are invisible to every automated tool.
Ask the question plainly and early, ideally in the first conversation rather than after the work. How did you hear about us, asked directly, produces an answer people are happy to give. Asked three months later it produces a guess, and the guess is systematically biased toward whatever they remember most recently rather than what actually brought them.
Record it as they said it rather than as a category, then group afterward. A customer saying their neighbour recommended you and one saying they searched for a plumber are different, and forcing both into a dropdown at the moment of entry loses the detail that makes the record useful. One free text column is enough.
Watch for the answer that is technically true and unhelpful, which is the internet. Somebody who searched, found your listing, checked your reviews, looked at your site, and then called has described a path with several steps, and knowing which one they would credit matters less than knowing the path existed. A follow up question about what they searched or where they saw you produces the useful version.
Six months is where this becomes valuable, which is why starting matters more than perfecting the method. Twenty entries reveal a pattern that no amount of reasoning would have produced, and the pattern is frequently surprising: a channel you were sceptical about producing steadily, or an expensive one producing nothing.
Add what they were worth alongside where they came from, since volume and value differ. A channel producing many small customers and one producing few large ones look identical in a count and completely different in revenue, and that distinction determines where the next year of effort belongs.
Keep it somewhere that survives, which means not in your head and not scattered across email. A spreadsheet with a date, a name, a source, and a value is sufficient, and a customer system with a source field is better because it stays attached to the record.
Then actually read it, which is the step that gets skipped. A quarterly ten minutes sorting by source and totalling the value answers the question every marketing decision depends on, and businesses that collect this diligently and never look at it have done the hard part and skipped the easy one.
Ask again at the end of the engagement as well as the start, since customers frequently discover their own answer during the relationship. Somebody who said they found you online will often name the specific referral or search when asked a second time, and the second answer is usually the accurate one.
Include the ones who did not buy as well, since knowing where unsuccessful enquiries came from is as informative as knowing where customers did. A channel producing volume that never converts looks successful in a lead count and costs you time.
Keep the format identical over time rather than improving it, because the value is entirely in comparison. A column that changed definition halfway through a year produces two incomparable sets of data and answers nothing.