A potential customer hears about a business from a neighbour, later sees a social post, searches for its name and finally makes an enquiry. Which contact created the order? A report may assign the result to the last measurable visit, or distribute credit across several observed interactions. The conversation with the neighbour may remain invisible. No single view necessarily contains the whole story.
Attribution assigns credit for an observed outcome according to rules or a model. Google explains this idea for Analytics. For a small business, the essential task is to understand the view being used and retain its limitations when comparing channels.
Define the outcome before discussing its source
A click on a telephone number, a submitted form, a suitable enquiry and a paid order are different events. If one system reports contact clicks and another reports completed orders, their totals cannot be compared directly. A disagreement may arise from definitions before any tracking problem is involved.
Write down the outcome definition for each report. Also establish whether repeated contact from one person can be counted more than once. Businesses with lengthy quotation stages need to preserve pending outcomes: an open quotation is neither a won order nor a lost one. Otherwise, the latest month may appear weaker simply because decisions remain unresolved.
Recognise different reporting windows
Systems can consider different intervals between an interaction and an outcome. They may also place an outcome under the date of the earlier interaction or the date of the later action. Monthly totals can therefore differ even when both systems are behaving as designed. Read the settings and definitions before assuming that a technical fault exists.
In a simplified example, a person clicks an advertisement near the end of September and enquires in early October. A report based on enquiry arrival counts the request in October. Another report may associate it with the September interaction. The actual arrival date matters for staffing, while the earlier connection may help assess September activity. Both questions require clear labels.
Even within one tool, the selected perspective affects a channel comparison. First known source answers a different question from last observed interaction. A model distributing credit provides another view again. Document any change of approach rather than comparing unlike reports without explanation.
Do not invent missing interactions
People may decline measurement, change devices, use different browsers or contact the company directly by telephone. Offline recommendations and previous experience can also matter. A report therefore describes what was captured or modelled under its particular conditions. Missing entries should not be allocated to a preferred channel by intuition.
A voluntary question such as “How did you first hear about us?” can provide another perspective. It is not a flawless control measurement. People remember differently, identify several contacts or answer a different question from the one represented by the technical report. Treat these responses as supplementary information and allow “not sure”.
There is no need to request a detailed personal history of media use. A short, purpose-specific question may be enough. Collect it only when the answers support a real decision; otherwise, the business gains another field to maintain without gaining insight.
Avoid adding overlapping claims
If several advertising systems each report a contribution to the same enquiry, adding their totals does not automatically produce the number of distinct enquiries. Maintain an appropriate business record of actual requests and orders. Platform reports then offer different perspectives on those outcomes instead of being treated as separate sales.
This does not require exhaustive tracking of individuals. For a small operation, the number of genuine enquiries and orders during a period can already provide an important cross-check. Keep known attribution separate from unknown origin. An honest unassigned category is preferable to a complete but invented distribution.
Ask a different question about additional impact
An attributed order does not by itself establish that the order would have disappeared without that activity. Additional impact is a separate question. A simple before-and-after comparison may be affected by seasonality, prices, capacity and simultaneous changes elsewhere in the business.
For a sufficiently important decision, a carefully designed experiment may help. It needs an appropriate comparison group, a defined outcome, sufficient observations and a method suited to the business. Very small volumes may prevent a reliable answer. In that situation, acknowledge uncertainty and make staged decisions instead of claiming precision the evidence cannot support.
A monthly report can retain two clear layers: what the attribution reports show, and what the business actually knows about enquiries and orders. Add known limitations and the next question that can be investigated. This is more useful than declaring a single channel responsible for every result.
