Which customers came from where? Measurement for growing businesses.

Why attribution is genuinely hard when the sale is a conversation

Ecommerce measurement is easy because the click and the sale happen in the same machine. Measurement for anything sold through a conversation is hard because they don't: the enquiry is a call or a form, the decision takes a proposal and a meeting or demo, the close happens weeks later, and the revenue is recorded in accounting or CRM software with no idea any marketing ever occurred. Platform dashboards paper over this by reporting what they can see: clicks, impressions, "conversions" they define themselves, each platform claiming credit for the same customer, and increasingly some of it estimated rather than observed. The gap between "the ad was clicked" and "the customer paid" is exactly where marketing budgets get argued about, and no dashboard closes it for you.

Tracking calls and forms without breaching trust

The mechanics are unglamorous: call tracking numbers that record which channel made the phone ring, forms that carry their source with them, and, where a call number can't be used, the front desk habit of asking "how did you find us?" and writing the answer down. Add "an AI assistant suggested you" to the options; it's an answer people are starting to give, and it's otherwise invisible. Do all of it inside the privacy rules as they actually are: tell people what's collected in a privacy notice written in sentences, respect consent for anything beyond what the service needs, and collect the minimum. Australian Privacy Act reform has been tightening these expectations, and the direction of travel is one-way. As third-party tracking narrows, the first-party records you hold with consent become the measurement that lasts. A measurement setup your customers would object to if they saw it described plainly is a liability with a dashboard. It's also unnecessary, because none of the numbers that matter require surveillance.

The join that makes reporting mean something

The fix is unglamorous too: connect the enquiry record to the customer record. When a call or form arrives, log its source in the same place the proposal and the invoice will live. Most CRMs and business systems have a source field sitting empty right now. That single habit turns "the ads got clicks" into "the ads produced these enquiries, which became these proposals, of which these were won, worth this much": the sentence every marketing argument is actually about. Not every record will match; calls arrive with no history, referrals lie about themselves, names get typed wrong, and research done inside an assistant leaves almost no trail. Count the unmatched honestly instead of forcing them into a story. A report that admits what it doesn't know is the only kind worth reading.

The handful of numbers worth watching

Once the join exists, most dashboards can be deleted. Enquiries by source. Qualified rate: what fraction of enquiries were worth a proposal, which is your quality filter per channel. Win rate on proposals. Won value by source, against what each source costs. And, where repeat business matters, whether customers from each source stay, renew or come back. Watched monthly, those few numbers answer the real questions: which channel brings customers worth having, whether the paid spend is buying enquiries or just clicks, and whether the website is converting the visits it gets. Everything else (impressions, sessions, followers) is weather on the way to those few.

The agent-native lens: research you can't see, joins you can audit

When an assistant did the research. More buyers are starting to ask an assistant to compare options and draw up a shortlist before they ever reach a website. The trail that leaves is thin: an answer engine may cite your page or may not, the person may arrive by typing your name, or the assistant may send the enquiry itself. Expect more enquiries to land as direct or unknown, and plan for that rather than pretend otherwise. Look for referrals from answer engines in your analytics, give assistant-made enquiries a source value of their own in the CRM, and every so often put your buyers' usual questions to an assistant yourself, noting whether you're named and what it says. That's a rough signal, not a metric, so log it as one.

When the join runs on agents. The weekly join decides whether measurement survives, and an agent can do most of it: match new enquiries to CRM records on phone, email and timing, propose a source for the ones that don't match cleanly, and draft the monthly table. Two rules keep it honest. Every match carries a reason and a confidence, and doubtful ones stay unmatched until a person decides; an agent that forces records into channels is writing the story the report exists to test. And the agent reads only what the join needs, under the privacy notice your customers were shown. Moving budget remains a named person's call, with the working attached.

When the data convicts the expensive channel

Eventually the join will say something uncomfortable: a channel you like, or pay a lot for, isn't producing customers. Before acting, check the cycle. Work sold through conversation closes slowly, so judge a channel over quarters, not a bad fortnight, and remember that some channels assist without closing. But when the verdict holds, act on it: renegotiate, restructure, or stop, and put the budget where the won-value column says it earns. This is the entire point of measurement: not prettier reporting, but the standing to make that call on evidence instead of on whoever argued loudest. The businesses that get this right aren't the ones with the most data. They're the ones whose few numbers connect to money. As platforms report more estimates and more research happens inside assistants, that connection to your own records is the part that will still hold.