Where Revenue Comes From

One year of sales, campaign, product, and customer data from a household and lifestyle retail business. This included daily transactions, campaign results, product movement, and customer buying patterns, showing how people shop, which campaigns work, and how different products perform over time.

Analysis focused on revenue, refunds, product demand, campaign performance, and customer loyalty. We added simple indicators such as month, week, discount level, product mix, and purchase frequency to build models that predict revenue, campaign outcomes, product demand, and long-term customer value.

Revenue is driven mainly by the products customers choose and how many items they buy. Some products naturally bring in more money, either because they cost more or sell more often, and larger basket sizes quickly raise revenue. Campaigns and timing have some influence, but their impact is much smaller. Discounts matter the least and do not meaningfully increase revenue per transaction.

Revenue stays steady for most of the year, with consistent monthly averages. February is the only clear dip, likely due to shorter shopping periods or the post-holiday slowdown. The biggest jump happens at the end of the year, where November and December contribute the highest revenue and show strong seasonal demand linked to holidays and bonuses.

Net revenue is highest when no discount is offered. Full-price purchases bring in more valuable customers and larger transaction amounts compared with discounted sales. Low and medium discounts do not outperform full price, and they often attract customers who spend less or are more likely to request refunds.

The forecast captures the general average level of revenue but cannot follow the daily ups and downs seen in real sales. The forecasts are still useful for understanding long-term direction, such as whether revenue is growing or stabilising. Accuracy would improve greatly by adding more detailed information about campaigns, product demand, and seasonal effects.

Campaigns perform best when they offer a wide range of products. Customers respond well when they have more options, and campaigns with stronger product variety consistently generate higher sales. Longer campaigns also help, but their impact is smaller compared with product mix. Discounts and refund rates show almost no meaningful effect on overall performance.

Customer spending stays almost the same across all campaigns, with average transaction values sitting between RM88 and RM96. This shows that campaign design does not meaningfully change how much customers spend. Refund rates also vary only slightly, and there is no link between high spending and higher refund risk.

Only a few campaigns contribute most of the net revenue. One campaign in particular clearly outperforms the rest, while several others offer moderate results, and many contribute only a small amount. The best way forward is to learn from the top performers and repeat what works.

The model can identify which campaigns are likely to succeed or fail, often giving very clear yes-or-no signals. Basic features such as product mix and duration already reveal simple patterns. The model is useful for screening new campaign ideas. If a planned campaign receives a low predicted score, it likely needs to be redesigned.

Most products sell at almost the same pace, with daily sales sitting between one and two units. There is very little difference in how products behave—no major best sellers and no items with extreme ups and downs. This creates a stable and predictable sales pattern across the entire range.

Demand stays low and steady for most of the year, with the occasional spike when certain items sell slightly more than usual. These increases do not follow a repeated seasonal pattern, and there are no consistent peaks tied to particular months or holidays. Sales movements are more likely influenced by promotions or short-term campaigns.

Sales and revenue are spread evenly across many products. No single item dominates performance, but no item consistently fails either. Products move slowly but steadily, lowering the risk of both stockouts and excess stock piling up. This creates a balanced inventory environment where lean stock levels are practical.

The forecast captures the general daily average but cannot predict sudden increases in sales. Most products sell one or two units per day, so the model learns a flat pattern and misses spikes that are often caused by promotions or price changes. The forecast is reliable for routine stocking decisions but not for anticipating promotional peaks.

Customers who stay active for a longer time are the ones who bring in the most value. They keep coming back, make more purchases over time, and rarely refund items. Refund behaviour also matters—customers who frequently return products tend to be low-value, while those who keep their purchases contribute more steadily. Repeat buying is another strong sign of future value.

The data shows three clear groups. A large portion of customers buy once or twice and then stop, which means they disengage early unless the business encourages them to return. A second group makes a few more purchases but does not reach high-value status. The third group is the most important—they stay longer, buy more often, and become your most profitable customers.

Different campaigns attract different types of customers. Some bring in shoppers who later become long-term buyers, while others bring in mainly one-time purchasers. Campaigns with a broader mix of products tend to attract customers who stay longer and spend more over time. Marketing resources should go toward campaigns that consistently pull in higher-value customers.

The model separates customers into three groups: those unlikely to return, those with moderate potential, and those with a high chance of becoming long-term buyers. These predictions allow the business to focus its retention efforts where they will have the greatest impact. Medium-potential customers benefit from early engagement offers, while high-potential customers should receive consistent loyalty attention.