Amazon Ads potential analysis: forecast, budget at risk and media performance optimization

Four levels that are often confused

The term Amazon Ads potential analysis has no fixed counterpart in a single Amazon product. It bundles four tools with four different reference times: the forecast says what will happen under the current settings. Budget at risk says how much budget will remain unspent at the end. Media performance optimization projects the expected performance at other budgets. AMC measures what actually happened. Only the last one is a measurement; the first three are models about the future.

That is why the order of checking matters, and it is also the order in which errors arise: whoever models the future first and then checks the past tunes to a forecast that the actual data does not confirm. Whoever measures first has a reliable baseline for every forecast.

The forecast: delivery confidence and its limits

The forecast in Forecast Intelligence applies, according to Amazon, to display, video and audio campaigns. It does not answer whether the campaign is profitable, but whether it can spend its budget at all. Amazon calls this delivery confidence: it combines your targeting settings, budget, frequency cap, bid strategy and viewability settings with available supply and expected competition to estimate the likelihood of delivering in full.

Two limits matter in practice. First, delivery confidence is explicitly a quick trust signal before you spend budget, not a performance forecast: accuracy varies by forecast type, campaign settings and supply source, and Amazon advises using it as a strong directional guide. Second, tightening any one of those settings narrows eligible supply, which in turn lowers the available spend amount and delivery confidence. A narrower audience therefore does not just mean less reach; it also means a worse forecast.

• **Forecast is not pacing.** A forecast predicts what will happen. Pacing measures how the campaign is actually delivering against its schedule. A campaign can pace behind while still being forecast to deliver in full, and the reverse can also happen. Conflating the two makes you consider a campaign at risk when it is not, or the other way round.
• **Some campaigns get no forecast at all.** Common reasons are unsupported ad group types or marketplaces, targeting that is too narrow or too broad, no active flight, and campaigns with more than 50 ad groups. When no forecast is available, the forecast widget explains why.
• **Very broad targeting is capped.** With very broad targeting over a long flight, spend forecasts can come out far above any campaign budget. Forecasts at that scale are not precise enough to plan against, so Amazon shows values above 1 billion USD as ">$1B".
• **Forecasts are produced per flight.** The forecast widget shows the current flight. Once a flight starts delivering, its forecast adjusts to the observed delivery.

Budget at risk: the number that looks ahead

Budget at risk answers the question of how much budget will go unspent by the end of the flight, accounting for expected changes in supply and competition. It is the counterpart to the delivery rate, which looks back: the delivery rate compares spend so far with the spend expected at this point in the flight. A campaign can be on schedule today while the forecast shows the remaining budget will not fully deliver.

The two often diverge when the campaign spent heavily early and has since slowed, when supply for its targeting is projected to decline, or when competition is expected to rise, for example at the end of the month or during seasonal peaks. Amazon also notes that budget at risk comes from a delivery forecast and not from a straight-line projection of your own daily spend: daily spend varies with the day of the week, seasonality and month-end demand, so multiplying one day's spend by the days remaining often gives a different answer. The forecast updates every day.

One practical point: budget at risk appears only after two to three days of continued under-delivery. The forecast confirms that a drop is sustained before flagging it, so a single slow day does not move the number. If you use budget at risk as an alarm, you are deliberately accepting two to three days of delay and should read the delivery rate alongside it daily.

Additional spend potential

Additional spend potential is the additional budget you could add to the current flight and still deliver in full. It appears only when the forecast is confident the campaign is on track. Amazon advises using the option when a campaign is performing well and you want to invest more, and explicitly says to check the budget vs KPI curve in Forecast Intelligence before raising the budget: it shows how added budget affects your KPI.

That is the decisive difference to the widespread assumption that more budget brings more revenue. The curve answers the question of how additional budget performs, and Amazon keeps it explicitly separate from the question of whether the campaign can spend its budget at all. Both can be true at the same time: full delivery and still declining returns per euro.

Media performance optimization: budget effect instead of delivery

Media performance optimization is the other half of the potential analysis, and it computes on the basis of the past: the playbook builds diminishing-returns curves from historical advertising data, projects the expected performance at other budgets and calculates the budget distribution that maximises a chosen KPI. Amazon stresses that this is a performance projection based on past campaigns and not a time series forecast.

In practice the two Amazon approaches complement each other but do not replace each other. The forecast answers "how much of my budget will be delivered?", media performance optimization answers "what does a given budget deliver in terms of KPI?". A campaign can spend its budget in full and still fall short of its own target KPI, and conversely a campaign with full delivery can raise its KPI without needing more budget.

The limit of both tools is the same: they do not know your actual costs. Amazon Ads delivers ad performance, not your margin. For contribution margin, purchase prices and inventory impact you additionally need the sales data from the vendor or seller path.

AMC as the actual record: what you can verify yourself

AMC measures what actually happened: clicks, purchases, revenue and the new-to-brand share, per campaign, ad group, ASIN and period. This is the level at which you can check a forecast. An example of the basic question: how many purchases came from new customers, and how many of those ran through promoted ASINs.

Two limits belong with this. First, a new-to-brand purchase is no proof of incrementality: the new_to_brand field applies to purchase events only and describes a buyer property of the brand, not the cause of the purchase. Second, AMC measures the ad and purchase events inside Amazon; Amazon Attribution measures the channels outside Amazon, and Brand Analytics in Seller Central shows the search and customer behaviour of the brand. The three tools answer three different questions and cannot be calculated against each other.

SELECT
  tracked_asin,
  SUM(conversions) AS conversions,
  SUM(CASE WHEN new_to_brand THEN 1 ELSE 0 END) AS new_to_brand_conversions,
  SUM(total_product_sales) AS product_sales
FROM
  conversions
GROUP BY
  tracked_asin

Assessment: what a potential analysis can deliver

Our conclusion, not Amazon's statement: an Amazon Ads potential analysis is a sequence of checks, not a single number. The most reliable statement comes when you start with the actual measurement, hold the forecast against it, and only then decide about raising the budget. The most common mistake is the reverse: read the forecast first, then trim your own figures until they fit.

Two limits remain. First, the forecast works at campaign level and where Amazon sees inventory; it does not know your actual contribution margins. Second, by Amazon's own statement, delivery confidence is a trust signal whose accuracy varies by forecast type, campaign settings and supply source. It is the best guidance Amazon offers, but it is not a promise.

Step by step

1. Fix the question. Is it about delivery, about budget effect on a KPI, or about the actual record? The question decides which of the four tools is the right one in the first place.
2. Measure the actual record in AMC. Evaluate clicks, purchases, revenue and the new-to-brand share per campaign and ASIN before you read a forecast. This is the baseline every forecast is checked against.
3. Check the forecast and its limits. Look at delivery confidence and budget at risk, and check whether the campaign type gets a forecast at all and whether the targeting is too narrow or too broad.
4. Project the budget effect. Use media performance optimization to compute the expected KPI at other budgets. The budget vs KPI curve in Forecast Intelligence shows how a higher budget affects your KPI.
5. Document the decision. Which level carried which decision, and which uncertainty remains, belongs in the next budget round. Otherwise the discussion repeats at the next monthly review.

Frequently asked questions