How to ask Flowask about sales, marketing and products

1. Know where the answer comes from.

Flowask helps you get from a question to your data in Flowpeek. The app does the business calculations through prepared reports. The model receives the results and explains them in its answer. This lets you start from the problem instead of remembering where every table sits in the dashboard.

This design has a practical consequence: the assistant works within the limits of the available reports. It doesn't run arbitrary analyses on the whole database, change campaigns or make budget decisions. You'll get the most out of it if you treat the conversation as the start of an analysis with a source you can check. A nicely worded explanation still has to match the period and metric you asked about.

2. Combine a metric, a period and one task.

Instead of "How are we doing?", type: "Summarize yesterday's sales and orders." Instead of "Which ads are best?", start with "Show costs and ROAS for the last full month." In the first step you establish the result, and only then do you ask follow-up questions about the observation. This makes it easier to check that the answer covers the right area.

Be clear about dates. "The last three months" means a rolling range, while "three full months" means completed calendar months. When comparing, specify both parts or check which periods the assistant chose. Watch out for the current day: setting a few morning hours against the whole previous day easily leads to an apparent drop. If you need a different range from the dashboard, name it in your question.

3. Start with a question a report can answer.

The examples below match Flowask's current areas. They don't require an elaborate prompt or knowing the technical name of the data. Once you get an answer, open the source and check that the scope matches what you meant. Especially with rankings, don't treat the items shown as the whole catalog.

You can also type "Prepare a summary of yesterday. What needs attention?" The summary uses the available KPIs and marketing reports. You trigger it with a question; it isn't automatic monitoring running in the background. If the result draws your attention to a channel or product, continue the work in the relevant report.

4. Read the source before you pass on a conclusion.

Check the store name, dates, currency and the definition of sales. An amount after reported returns and cancellations isn't automatically the paid amount or the value excluding VAT. If the table doesn't contain a cost, the assistant shouldn't present it as zero. Without a common currency, it shouldn't add values up into a single result.

The product ranking covers at most 20 items, and the page ranking at most 15. A list of the most frequently purchased products isn't the full total of catalog sales. Likewise, the most viewed page doesn't necessarily sell best. Check that the answer keeps these distinctions. When you prepare a note for the team, include the question, the period and the source report, not just the final paragraph.

Example: the question is about channel traffic over the last 30 days. The answer should give the dates, name the attribution model, e.g. last touch, and say if it received only part of the report, e.g. 20 of 45 items. These are three things to check: whether the dates match the question, whether you're comparing the right model and whether the conclusion only applies to the visible part of the list. Don't add the "Unattributed" row to Direct.

5. Recognize questions that need different data.

"Does mobile sell better?" requires conversion or revenue by device. The assistant's available report showing traffic shares doesn't answer that question. Likewise, a list of popular products doesn't let you calculate margin, and the current ML score isn't a revenue forecast for tomorrow.

An arbitrary filter in a sentence doesn't mean it was applied. After a product ranking, the question "And only from Google?" shouldn't lead to a seemingly filtered result if that breakdown isn't supported. You can switch topics to "Show Google and Meta sources," but you'll get an overview of channels, not product sales limited to Google. A good answer will name the missing metric and the available scope instead of substituting one for the other.

• Not supported: profit and the store's full margin.
• Not supported: arbitrary breakdowns, e.g. product ROAS or new-customer sources.
• Not supported: a forecast of future revenue.
• Device report in a conversation: traffic share, not automatically device conversion.

6. Follow up on one problem you've noticed.

Imagine a store notices lower sales. Start by comparing the results of two completed periods. If the drop is confirmed, check orders and average order value. Then ask separately about channels or the most frequently purchased products. Each step should have its own source instead of relying only on what the previous paragraph suggested.

Add context the team knows: a promotion ended, shipping changed or a popular variant ran out. These are hypotheses to test. Don't ask the assistant to confirm a cause you've already picked. The question "Why did revenue drop?" first requires checking whether, and to what extent, the drop happened. Only then is it worth working out which data can bring you closer to an explanation.

7. Record the finding and come back for a fresh reading.

The conversation history keeps the questions, answers and source context. Use it in your weekly review to remember what a decision was based on. In your note, separate the observation from the hypothesis: "average order value dropped" is different from "customers are buying less because of shipping." Add how you'll test the proposed explanation.

A saved answer describes the moment of an earlier analysis. When you want the current result, ask a new question and check the period. Don't treat an old conversation as a self-refreshing dashboard. If the answer was cut off or an error appeared, go back to the source before using it. In the consultation during your first month, you can practice the whole process on a question about your store.