Merchant Center Predictive Insights: Check the Economics Before You Act

Which kind of insight are you looking at?

Under Analytics → Pricing, Merchant Center offers price benchmarks and sale-price suggestions. Google describes the latter as predictions based on recent performance, price sensitivity and comparable businesses; predicted click or conversion uplift is not a guaranteed outcome. A benchmark answers how your price compares. It does not tell you whether your business can afford that price.

Recommended promotions are another feature. Google's current Help limits eligibility to active US merchants with a Merchant Center promotion created in the preceding 365 days. Recommendations can suggest products, discounts and seasonal timing. You can edit or dismiss them; accepting the idea still requires the offer to work on your website and pass Google's promotion review.

Automated discounts are different again: once configured, they can optimize discounts within the limits you supplied. A recommendation awaiting a decision and an enabled price-changing feature are not the same operational state. Check what is already active before discussing whether to adopt a suggestion.

AI performance insights are a reporting feature, not a sales forecast. The current dedicated Help page describes English-language availability in Australia, Canada, India, New Zealand and the US, under Analytics → Products → AI performance. Its current scope is organic AI traffic, not paid ads. Do not use that share-of-voice number as a Performance Max ROAS report.

If a card uses another label, read its description and scope. A broad announcement about predictive capabilities is not enough to invent a stockout alert, confidence score or expected demand increase that the account never showed.

More orders can mean less contribution

Consider a hypothetical product priced at $100. After $60 in product and variable fulfilment costs, it contributes $40 per order before advertising. Suppose a suggested sale price is $85 and the business models 30% more orders. These are illustrative assumptions, not a result from Merchant Center.

At the original price, 100 orders contribute $4,000 before ad costs. At the sale price, 130 orders contribute $3,250: 130 × ($85 − $60). Revenue rises from $10,000 to $11,050 while contribution falls by $750. A higher conversion count would not rescue those economics.

To match the original $4,000 contribution at $25 per order, the business needs 160 orders, a 60% increase, assuming the same variable cost and ad spend. If acquiring the extra orders costs more, the required result rises again. Returns, payment fees and tax treatment can change the calculation, so use the business's own consistent definitions.

That does not make the discount wrong. Clearing old inventory, acquiring genuinely valuable new customers or avoiding storage costs may justify a different objective. Name that objective and its cost instead of calling a revenue increase a profit increase.

Inventory is an independent check. Use actual sellable units, reservations, inbound stock and supplier timing. Merchant Center's product availability is not a live warehouse ledger, and a prediction of conversion uplift is not proof you can fulfil the extra orders.

Check the proposed change, not just the headline number

A missing recommendation is not proof that the account is broken. Google's pricing guidance says eligibility does not guarantee suggestions when it sees no substantial opportunity; opted-in automated discounts also change whether sale-price suggestions appear. Check the stated conditions, including that purchase conversions are being reported, before treating an empty card as a data problem.

Visibility reports need different care. Merchant Center's AI report can display zero share of voice with insufficient impressions, a dash with no impression data, and 100% when competitor data is unavailable. Those are not interchangeable business outcomes. Do not celebrate a market-leading score until you know what its comparison includes.

• Match the exact product, market and currency. A suggestion for one variant should not silently become a catalogue-wide discount.
• Identify the predicted metric and its assumptions. More clicks, more conversions and more profit are different outcomes.
• Calculate contribution with the proposed price and the business's actual costs. Include the extra advertising cost if the plan requires more spend.
• Check stock and the store's offer implementation. The displayed promise must work through checkout, not just in a recommendation card.
• Inspect any existing automation and price limits. Avoid manually assessing a suggestion as though nothing can change while an automated feature is active.

Make the smallest change that answers the question

Correct a clear store or product-data error because it is wrong, not because a prediction promises a return. For a discretionary discount, decide whether the possible contribution is worth testing and whether you can fulfil the offer. If not, leave the suggestion unused.

When a test is worthwhile, avoid changing price, creative and budget together unless that combined package is what you intend to evaluate. Define the commercial outcome you need, allow for conversion and return delays, and compare like-for-like periods or an appropriate controlled test. An inconclusive result remains inconclusive.

Keep the decision understandable to the person who owns the store: the product, the proposed price, the cost assumption and what would make the test worth continuing. That may fit in the existing promotion workflow. It does not require a new weekly reporting ritual or a predetermined rollback date for every insight.