Keep one record from first inquiry to outcome
A spreadsheet is enough to begin if somebody maintains it. The important part is a consistent record, not the software name.
Create one row per genuine opportunity, with a unique identifier. A person who calls after submitting a form should not become two leads. Separate test submissions and obvious spam. Keep an audit trail when merging duplicates so the original source information is not lost.
Minimum useful fields
Record inquiry date, customer or project identifier, service, location, first known source, contact route, campaign if known, fit, response owner, next action, quote date, quoted value, outcome and reason when known. Add booked value and expected delivery contribution when available.
Keep source discovery and contact route separate. “A neighbor recommended us” and “submitted the website form” can both be true. Ask: “How did you first hear about us?” and, when useful, “What prompted you to get in touch now?”
Use practical statuses
New, clarification needed, suitable, quote in progress, quoted, waiting for a dated decision, won, lost, postponed and declined can work. Keep the list short enough that people use it consistently. Define what “won” means, such as accepted scope with the required booking step complete.
Do not put every unknown into lost. Pending opportunities should stay visible with age and next action. Do not let “pending” become a hiding place for requests nobody has reviewed in months.
Protect the record
Collect only information needed for the business purpose. Limit access, keep backups and follow retention and privacy obligations. Notes should describe the commercial situation professionally. Customers are people, not labels such as “time waster.”
Your output: one record and one owner. Review it weekly before adding a more complex CRM. If you later introduce software, migrate the definitions as carefully as the data. Better tooling cannot make inconsistent stages or missing outcomes trustworthy.
The record earns its place when it answers a real question: which activity is bringing suitable work, and which live inquiry needs attention today?
Compare channels using the same definitions
Here is a fictional example of three channels after their inquiries have had enough time to reach a decision. All figures are illustrative and exclude tax. The all-in acquisition cost includes media or fees, allocated management and valued owner selling time, using the same rules for every channel.
| Channel | All in cost | Genuine inquiries | Booked jobs |
|---|---|---|---|
| Local campaign | $900 | 12 | 3 |
| Paid social | $900 | 30 | 2 |
| Trade introductions | $450 | 5 | 2 |
The cost per genuine inquiry is $75, $30 and $90 respectively. If you choose only the cheapest inquiry, paid social appears strongest. The cost per booked job is $300, $450 and $225. That tells a different story.
Suppose the booked jobs produce delivery contribution of $3,600, $1,600 and $2,400 before acquisition and fixed overhead. After acquisition, the amounts are $2,700, $700 and $1,950. These are not net profit figures, but they help compare the work created.
Do not overinterpret a small sample
One additional win could change the ranking substantially. Job types may differ. Trade introductions may be difficult to scale. A campaign may produce repeat work later. Keep the observations and uncertainty together.
Compare inquiries acquired in the same period and allow for the buying cycle. If you compare this week's advertising spend with bookings from last month's leads, the result can be misleading. Use an inquiry-date group, then update its outcomes as they mature.
Include capacity and cash timing
A channel that produces distant, complex jobs may use more delivery capacity. A large commercial booking may collect payment later than small residential work. These differences matter even when the acquisition ratios look attractive.
Your output: a monthly channel review showing cost, suitable inquiries, bookings, contribution and time. Mark unknowns explicitly. Keep improving measurement without pretending attribution is perfect. The purpose is to make the next decision better, not to create a mathematically impressive report that hides the business's uncertainty.
Go further: your quote win rate, calculated without open quotes distorting it.
Run experiments you can learn from
An experiment needs a clear question. “Does marketing work?” is too broad. “Does showing project examples beside the inquiry form increase suitable conversations from this service page?” is useful.
Write the current observation, proposed change, expected effect, measurement, budget or time limit and review date. Change as little as practical. If you alter the audience, message, price, page and response process together, you may improve results but will not know which change mattered.
Use three review levels
Technical review: immediately check that delivery, links, notifications and measurement work. Fix obvious faults without waiting for a statistical threshold.
Quality review: inspect early inquiries. Are they real, in area and relevant? Correct a clearly misleading promise or wrong geography promptly.
Commercial review: allow enough time for decisions, then compare acquisition cost and contribution. Use a predetermined loss limit so learning cannot become an open-ended expense.
Low-volume businesses may not have enough traffic for a formal split test. In that case, combine a careful before-and-after comparison with customer conversations and usability observation. Note seasonality and other changes. Do not claim statistical certainty from a handful of submissions.
Decide what happens next
Continue when the signal is promising and the risk remains acceptable. Modify when you can identify a specific obstacle. Stop when the economics or fit are clearly wrong, or when the test reaches its agreed boundary without enough reason to continue.
A stopped test is useful if you record what it taught you. “Facebook did not work” is vague. “The campaign attracted requests outside our service area and the form did not ask location” points to a correctable problem.
Your output: an experiment log with one active priority. Keep a backlog of ideas, but do not let every new suggestion interrupt the current test. Consistency is valuable only when you are consistently doing something that can produce evidence.