Describe the signal without confusing it with the cause
A hypothesis begins with an observation specific enough for another person to verify. Record the metric, period, group analysed and relevant conditions before explaining what may have happened. Separating the signal from its interpretation helps prevent a suspicion from being treated as a confirmed diagnosis before the evidence supports it.
“The campaign is tired” is an interpretation that is difficult to assess without context. “Cost per enquiry increased during the period, with fewer completed forms” describes the starting point more clearly. You still need to check whether traffic sources, the page, measurement or the way enquiries were counted changed.
First confirm that the data exists and is comparable. A recording failure can resemble a sales decline, while a change in the definition of a conversion may produce an improvement that is only apparent. Fixing a known fault takes priority over designing a test to explain data that does not represent actual operation.
Keep a reference to the report or dataset used, together with the date you read it. You do not need to copy every number into the record, but someone should be able to reconstruct the observation. That detail becomes useful after further changes accumulate and the report displays a different period.
To prepare your next step, explore Artificial intelligence in digital marketing.
Write the hypothesis as a testable explanation
A useful hypothesis connects an observed problem to a plausible mechanism and a specific change. Explain why that change might improve the result that matters without presenting it as certain. Include an alternative explanation as well, so the team knows which possibilities have not yet been ruled out by the available evidence.
In a hypothetical example, many enquiries come from people who need the service immediately, but the business cannot start new work for several weeks. The hypothesis might be that limited visibility of the timeframe contributes to unsuitable enquiries. The proposed change would make that timeframe clear before the form.
The expected result would be a higher proportion of suitable enquiries, even if total volume fell. An alternative explanation would be that the team communicates different timeframes to interested people. In that case, changing only the page would not resolve the entire difficulty, and the record helps preserve that possibility.
This formulation makes you connect the action to the business. “Let's try another image” remains incomplete until you explain what information or perception you intend to change. If you cannot explain the mechanism, you may need to listen to more sales conversations or review the offer before producing another version.
Define the change and preserve the previous context
Record exactly what will change and preserve a reference to the previous version, including its message, destination and relevant conditions. Identify simultaneous changes that might affect interpretation. Clear documentation makes it easier to understand the result without assuming that every difference arose from the particular intervention you originally proposed to make.
In the timeframe example, you can save the previous and new wording, where the information appears and when it went live. Keep the versions available during each period identifiable. A generic note such as “page optimised” does not explain the experience of the people included in the report.
When you need to change several things for an operational reason, record them as a combined intervention. Do not later attribute the result to one sentence if the offer, budget and service process also changed. The intervention may be necessary and useful, but conclusions should match what you can actually distinguish.
If automation is involved, save the rule authorising the action and what it actually executed. The requirement to explain why an automation changed a campaign applies before and after the change. A description generated by the tool itself must also match the actual record.
Choose the primary outcome and acceptable limits
Define in advance the indicator that answers the hypothesis and the outcomes you cannot acceptably compromise. Specify the observation period, available investment and conditions for stopping because of an operational problem. Your criteria should reflect the sales process and data quality instead of relying on universal numbers detached from the context.
To assess suitable enquiries, you need a stable definition of suitability and enough time for the team to classify them. An improved percentage accompanied by a steep fall in opportunities may not serve the objective. Also monitor opportunity volume and cost, distinguishing primary indicators from supporting information.
Plan the comparison before starting. A test with comparable groups supports a different interpretation from observing one period before and another after, both exposed to changes in demand and operation. If you use the second approach, acknowledge that limit and avoid treating the observed difference as isolated proof of causality.
The budget required to test an advertisement depends on the question and the frequency of relevant outcomes. No number of days, on its own, guarantees a sound conclusion. Decide beforehand how to handle insufficient data instead of extending or ending evaluation simply because you like a temporary result.
Use a short record someone can execute
The record should bring together observation, hypothesis, change, comparison, primary indicator, limits, owner and review timing. Each field needs enough information to guide execution and later interpretation. Keep it short, but remove ambiguities that could lead two people to run different experiments while believing they followed precisely the same instruction.
In the hypothetical scenario, the record could specify clarifying the actual start timeframe on the page and corresponding advertisements. The sales team would apply the same suitability definition throughout evaluation. The review would examine compatible enquiries, cost per compatible enquiry and any effects on the ability to fill the calendar.
Record who owns execution and who will assess the result. In a small team, one person may perform both roles, provided decisions remain explicit. If enquiry classification depends on someone else, confirm beforehand that the task fits their routine and that the necessary information will be entered.
A review date does not require you to declare a winner on that date. It can be used to check execution, measurement and whether enough information has been collected. Distinguishing this monitoring from a final conclusion reduces pressure to invent an answer before the process can support one.
Close the evaluation without rewriting the original hypothesis
When review time arrives, compare what happened with the recorded criteria and preserve the original hypothesis. Classify the conclusion according to the evidence available, including inconclusive results. Then decide whether to keep the change, restore the previous version or prepare a new question based on what the evaluation has taught you.
A favourable result deserves the same attention to context as an unfavourable one. Check that the change was implemented as intended and whether relevant external changes occurred. If the team also started responding faster, that information should accompany the interpretation of conversions rather than disappearing from the account.
Add the result to the record and connect it with the learning recorded at the end of a campaign. Keep observation, interpretation and the next decision separate. Anyone consulting the history can then find a usable explanation, including the limits of what that particular experiment allowed you to conclude.