A practical guide to campaign forecasting with Amazon's media planning API suite

I'm a media strategist at a mid-size advertising agency. A wellness brand came to us with $150 million to spend on Amazon. They had one question: "Can you guarantee this will work?"

I couldn't. None of us could. We had 15 years of experience, thousands of campaigns behind us and genuine belief in the product. But we didn't have certainty. We were building a $150 million media plan the same way architects designed buildings before structural engineering existed: experience, intuition and hope. Inside, we were terrified. Three months in, the client asked: "Are we reaching our target audience like you predicted?" I had no idea. We couldn't accurately measure unique reach across channels. Our reach projections? Educated guesses.

Six months later, I attended an Amazon Ads conference and learned about Media Planning APIs that would have answered every question we couldn't answer:

• What was the modelled reach for our advertiser context across Amazon inventory last quarter?
• How many will we reach at different budget levels?
• What campaign outcomes can we predict before we commit the budget?
• Are we paying to reach the same person across multiple channels?
• Which customer behaviours predict the highest conversion rates?
• How do we measure the true impact of our media investment?

Today, that same client invests $200 million annually, not by spending more on the same channels, but by expanding into new personas and inventory identified through the API suite, and they do it with complete confidence. Not because we're smarter. Because we stopped guessing.

Think of media planning the way a film studio plans a big film release. Before spending a dollar on production, the studio asks: Who is our audience? What channels will they be on? How much should we spend in each location? What return should we expect? Media planning works the same way for advertisers: it's the strategic work you do before buying ads, so your campaign has a real shot at success.

Without a plan, you're guessing. With the right tools, specifically Amazon's media planning API suite, you're forecasting. That difference is worth tens of millions of dollars.

Amazon's media planning APIs are organised into four groups, each addressing a core planning challenge:

How many people did we reach, how many will we reach, and what is the real overlap?

What outcomes can we predict before committing budget?

Who should we target and why?

How do we measure true media impact across channels?

All APIs operate over Amazon Ads inventory and behavioural signals. Cross-publisher reach, deduplication, performance comparison and audience characterisation require third-party measurement partners.

Every campaign moves through four phases, and the Media Planning APIs plug into each one.

Before spending a pound, you analyse past results, forecast reach and performance and build your plan. Reach Curve APIs, Performance Forecasting and audience APIs give you the data to do all three.

With your plan in hand, you set up targeting, configure channels and finalise budget allocations. Audience APIs define who you're reaching, while Reach Curve APIs (Reach Forecasting, Forecast Deduplication) help you validate your channel mix before launch.

Once the campaign is running, you monitor performance, adjust budgets and optimise targeting. Reach Curve APIs (Forecast Deduplication) show whether you're reaching unique customers or paying for overlap.

After the campaign ends, you measure actual vs. forecast, extract learnings and plan the next campaign. Reach Curve APIs (Historic Reach) show what actually happened and the MMM API feeds those signals into your measurement framework.

Historic Reach gives you historical reach curves for Amazon inventory, scoped to your advertiser context. You give it a date range and audience parameters; it returns modelled unique-reach figures based on Amazon panel and inventory data. No guesswork. No assumptions.

We ran Historic Reach on the client's past campaigns using six months of data. Result: 12 million unique customers, not eight million. A 50% measurement gap. That changed everything. Their cost per reach was 37% lower than they thought, making their campaigns far more efficient than their initial metrics suggested. They weren't underperforming: they were under-measuring. With that baseline, we could confidently plan a $150M campaign.

Historic Reach looks backwards, not forwards. It won't predict next quarter. But you can't build a good forecast without understanding your actual starting point. Think of it as reading last year's financials before setting next year's budget: essential context, not the full strategy. Not useful for first-time Amazon campaigns with no history.

The client asked: "Should we spend $120M, $150M or $180M? What's the difference in reach?" Traditionally, we'd guess at rough linear scaling from last year's spend? Those question marks... that's where anxiety lives.

Reach Forecasting lets you test budget scenarios before spending anything. You put in your target audience, campaign dates and multiple budget options. It returns a reach curve showing exactly how many unique customers each budget reaches and where the curve flattens, where extra spend stops delivering proportional reach.

Illustrative example demonstrating reach forecasting API capabilities:

The curve flattened sharply after $150M. Going from $150M to $175M added only 1.7M customers at $14.70 per incremental customer versus a $7.35 average. We recommended saving that $25M and reallocating it to prospecting. The client's reaction: "You just saved us $25 million before we spent it."

These are projections, not guarantees, similar to weather forecasts. Accuracy varies by campaign conditions; the ranges shown are illustrative examples based on typical performance patterns. Always present a confidence range, e.g. "We predict 20.4M reach, with a likely range of 18.5M to 22.1M." Plan with a margin, track actuals and refine your models over time.

Halfway through the Q4 campaign, the client asked for a check-in. We told them we were reaching 21.6 million unique customers across multiple Amazon Ads channels. We felt good about that number. We were wrong.

Forecast Deduplication shows you how many unique people you're actually reaching across all your Amazon channels, not the inflated sum of each channel's reach. It maps the overlap between channels so you can see exactly where you're double-counting and paying for the same impressions twice.

Illustrative example demonstrating forecast deduplication API capabilities:

We ran forecast deduplication. Real unique reach: 13.2 million, not 21.6 million. 8.4 million 'customers' were the same people appearing across multiple channels. Connected TV channels had 68% overlap. We'd wasted $61,740 reaching people we'd already reached. We restructured the media mix, cut overlapping channel budgets, shifted spend towards Twitch (where overlap was lower) and achieved 16.8M unique reach at the same $150M budget. A 27% increase in unique reach for free.

"We were paying $61,000 to show the same people ads across five different screens. That's not multi-channel strategy: that's paying rent on the same flat five times."

Deduplication only works within Amazon's inventory. If you're also running Google, Meta, TikTok or linear TV, those overlaps won't show up here. Tell clients upfront: "This shows Amazon audience overlap. For cross-publisher deduplication, you'll need third-party measurement partners.” Set expectations clearly.

Reach is important. But the client's real question was: "How efficiently will this campaign deliver?" They needed to know, before committing $150 million, what their Cost-per-Click would look like, what video completion rates they could expect, and whether the spend would deliver at scale. The best we could say was: "Based on industry averages, trust us."

Performance Forecasting predicts campaign outcomes before you spend a pound. You select a metric (clicks, video completions or detail page views) and the API returns a forecast curve: at each spend level, you get the predicted count, cost-per-outcome (CPC, eCPM, CPDPV, CPVC) and rate metrics (CTR, VCR). Not a hopeful estimate. A data-backed prediction with confidence intervals.

Illustrative example demonstrating Performance Forecasting API capabilities:

The API returned a forecast curve showing predicted clicks and cost-per-outcome at our planned spend level, with confidence intervals that cleared the client's efficiency benchmarks even at the lower end. We walked into the planning meeting and said: "Even in the conservative scenario, your cost-per-outcome is within target. We recommend green-lighting this campaign."

Three months later, actual click-through rates and cost-per-outcome landed within 5% of the API's forecast. The campaign ultimately delivered 4.0 times ROAS as a business outcome. That combination, accurate forecasts feeding confident decisions, changed the client relationship entirely and earned us a significant renewal.

Accuracy improves with data. Variance widens for new product launches and during peak periods. Campaigns with a longer Amazon history produce tighter predictions. Communicate this clearly to clients: confidence intervals aren't a hedge, they're the honest truth about how predictions work.

The Audience API group helps you move beyond demographics to understand who your customers actually are and find new ones you didn't know existed.

The client's original target: "Adults 25–54, interested in Health and wellness, household income $60K+." That describes approximately 42 million people on Amazon. Which segments should we prioritise and how much should each get? Demographics tell you who people are. They don't tell you what people do. And behaviour predicts conversion far better than age brackets.

• Persona Builder uses Amazon's first-party data, what people buy, watch and search, to group your audience into behavioural segments. Instead of age ranges, you get actual customer profiles based on purchase patterns, streaming habits and search behaviour.
• Audience Discovery surfaces new audience segments you haven't considered: people whose behaviours signal high intent but who fall outside your current targeting. It's how you find the customers you didn't know to look for.
• Audience Overlap shows how your segments intersect, so you can understand which audiences are truly distinct and which are largely the same people. This prevents you from over-investing in segments that overlap heavily.

Together, they answer: "Who are our best customers, who else looks like them and how do we avoid targeting the same people twice?"

Illustrative example demonstrating audience API capabilities, with the remaining 10% reserved for testing new segments identified via Audience Discovery.

By shifting budget from broad demographics to behavioural personas, campaign efficiency improved significantly. The Active Wellness segment, just 8.2 million people, delivered the highest returns. They were the superfans all along. We just couldn't see them until we looked at behaviours.

Audience Discovery then surfaced a segment we'd never considered: "Fitness content streamers who buy organic snacks." Audience Overlap confirmed this group had less than 12% overlap with our existing targets: a genuinely new audience, not a repackaged version of one we already had.

Audience APIs show behaviours as seen through Amazon’s lens. If your customer primarily shops or consumes media that is not on Amazon, the picture will be incomplete. Layer in your own first-party data for a fuller view. Also, not ideal for very small niche audiences where sample sizes limit meaningful clustering.

After the campaign ended, the client asked: "What was the real impact of our Amazon investment compared to our other channels?" We had platform-level metrics, but no standardised way to feed Amazon's signals into their broader measurement models. Our analytics team spent days manually pulling reports, reformatting data and reconciling discrepancies. By the time insights reached the planning team, the next campaign was already in-flight.

MMM API gives advertisers and measurement partners programmatic access to Amazon's advertising and retail signals for faster, more scalable Marketing Mix Modelling. It eliminates manual data exchanges and automates delivery of media signals (impressions, reach, spend) and retail signals including daily shopping engagement metrics like add-to-basket events directly into existing workflows. The result: more frequent, reliable insights for cross-channel budget optimisation.

Illustrative example demonstrating MMM API capabilities:

What used to take our analytics team days of manual data wrangling became automated. Amazon's media and retail signals, including granular shopping behaviour data, flowed directly into the client's MMM on a daily cadence, not quarterly. The result: a clearer picture of how Amazon investment drove outcomes relative to other channels, updated frequently enough to actually inform the next budget decision rather than arriving too late to matter.

The MMM API provides the signals: it doesn't build the model for you. You'll need an existing MMM framework (in-house or via a measurement partner) to ingest and analyse the data. The value is in the automation and standardisation: consistent, programmatic access that reduces manual effort and enables more frequent analysis.

• Historic Reach is the foundation. You can't forecast accurately without knowing what actually happened. Start every client relationship here, even if the truth is uncomfortable.
• Forecast multiple scenarios, not one plan. Present 3-5 budget options with reach curves. Let the data recommend. Clients respect optimisation, not just maximisation.
• Confidence intervals build more trust than false certainty. Don't promise a single number. Promise a range with confidence level. When you deliver within range, you're a hero.
• Behaviours beat demographics. Stop targeting 'adults 25-54'. Target 'protein powder buyers who stream fitness content'. The conversion difference is significant.

• Supported store (confirm your locale is covered)
• Amazon campaign history (minimum six months recommended for historic reach)
• Multi-channel media mix (required for forecast deduplication to be meaningful)
• AMC instance (for first-party data layering)
• Developer registration and account-team intake

1. Month 1: Historic Reach. Run Historic Reach to establish baseline reach for your category and inventory mix over the past six months. Show your client actual vs. assumed reach. That discovery alone usually justifies the whole effort. Review our documentation for campaign-by-campaign measurement.
2. Month 2: Reach Forecasting. Test 3-5 budget scenarios for the next campaign. Present reach curves. Let the data drive the recommendation.
3. Month 3: Performance Forecasting. Add outcome predictions with confidence intervals. Set weekly tracking against forecast.
4. Month 4: Forecast Deduplication. Run mid-campaign. Find the overlap. Restructure the media mix to maximise unique reach.
5. Month 5–6: Audience APIs. Build behavioural personas with Persona Builder. Discover new segments with audience Discovery. Validate distinctiveness with audience overlap. Reallocate budget to the highest-converting personas.
6. Ongoing: MMM API. Integrate Amazon's media and retail signals into your measurement framework. Use insights to inform the next planning cycle.

New integrators should expect wider variance bands during the first 30 days while the model warms up to their traffic profile.

Explore the resources listed below to help you get started with the Media Planning API suite.

The Amazon Ads API Guide and Media Planning Guides provide technical specs, endpoints and integration guides. API documentation also provides current locale availability and API status.

Gigi, Pacvue, Perpetua, Intentwise and Skai offer pre-built integrations with no coding required, including dashboards.

The Partner Network helps you find certified Amazon Ads partners who can help with implementation.

The Media Planning Certification offers free training on media planning fundamentals and API use.

Your Amazon Ads rep provides case studies, onboarding assistance and pilot programme access.

Not a developer? That's fine. Several Amazon Ads partner platforms (Pacvue, Perpetua, Skai, Gigi and others) are progressively integrating Media Planning APIs. Zero development time required. Work with your account team to identify the right partner for your stack.

• Sales Signals: Exploring deeper integration of Amazon's retail and shopper signals directly into pre-campaign planning workflows. This will connect media planning to actual retail outcomes, not just ad metrics.
• Agentic integration layer: A new capability enabling third-party AI agents to integrate directly with media planning APIs. This means your proprietary planning tools, automated workflows and AI-powered optimisation engines can programmatically access Amazon's planning signals, moving from manual API calls to autonomous, always-on planning intelligence.

Additional resources:

• Contact your Amazon Ads account team
• Join the conversation in our GitHub community
• Complete developer registration

Sources:

¹ Amazon internal, US, 2025-2026.

² Amazon Ads API documentation, WW, 2026.

Contributors:

Vic Awokoya, Kapil Dwivedi