Marketing Analytics Blog | Adverity

How To Give AI Context to Accurately Interpret Marketing Performance

Written by Sophie Barwood | Sep 10, 2026, 7:30:00 AM

This is the second article in our series about making AI reliable, looking at what AI needs at each stage of a marketing analytics investigation.

To make AI trustworthy, marketing teams must first provide three things:

  1. Knowledge of your data: AI needs a dynamic map of what your data means and how it should be used.
  2. Knowledge of your business: AI needs a comprehensive context on what’s happening in your business and industry to accurately interpret the data.
  3. The ability to act: AI needs the tools to execute the query and prove how it got there.

The first article covered the marketing data knowledge AI needs before it touches the data: metric definitions, canonical concepts, field mappings etc. Building on that, this blog will explore the live, situational understanding AI needs during the investigation.

Why is context crucial for accurate analysis?

Marketing data lives within a web of external variables that inform it, and AI needs to understand that in order to accurately deduce meaning. ROAS might fall because performance declined. It might also fall because the budget shifted, a promotion ended, a market entered a test phase, or a campaign objective changed. A spike, drop, or anomaly usually has a reason behind it, and a metric can be calculated correctly while still being interpreted poorly.

The data shows us that something has changed, but it’s the context that tells us how to explain what changed.

AI needs access to that business context during the analysis - things like campaign objectives, budget changes, active promotions, live tests, market-specific rules, reporting frameworks, and historical performance patterns. Without these, people lose trust in the answers provided, because they don’t reflect reality.

 

What business context does AI need access to?

To make AI reliable during analysis, marketing teams need a consistent way to provide context such as:

  • Campaign objectives and success criteria
    What each campaign is meant to achieve, and how success should be judged.
  • Budget changes and media plan shifts
    Where spend has moved, when it moved, and whether changes in performance reflect a planned shift in strategy.
  • Promotion calendars and launch dates
    Which offers, product launches, seasonal campaigns, or market events were active during the period being analysed.
  • Active tests and experiment structures
    Which markets, audiences, channels, creatives, or campaign groups were part of a test, and how results should be read.
  • Relevant files for plans, and briefs
    Media plans, campaign briefs, budget sheets, test plans, PDFs, and planning documents that explain what was supposed to happen.
  • Brand profiles
    Knowledge about brands, product lines, markets, competitors, positioning, strategic priorities, and reporting conventions.
  • Audience segments and ICP
    Priority audiences, customer groups, lifecycle stages, targeting segments, ideal customer profiles, and how performance should be interpreted for each.
  • Messaging strategy
    Core positioning, value propositions, product messaging, approved claims, category language, and competitor differentiation.
  • Channel roles
    How different channels are expected to contribute, such as awareness, acquisition, retargeting, retention, demand capture, or pipeline creation.
  • Market-specific rules and exceptions
    Local market conditions, exclusions, regulations, seasonality, competitor pressure, or reporting differences that affect interpretation.
  • Data quality caveats
    Known issues such as missing days, duplicate rows, tracking outages, platform sync failures, incomplete markets, or mixed daily and monthly rows.
  • Product, pricing, and availability changes
    New products, price changes, stock issues, offer changes, or availability constraints that may affect demand or conversion.

 

The important shift is moving this context out of individual memory, scattered documents, and one-off prompts, and into a governed layer AI can use during every investigation. Every user starts from the same business understanding, and AI gets a fuller picture of the situation behind the numbers.

 

How does Adverity Atlas assemble context without asking users to explain everything?

The goal is to avoid inconsistent context being given by multiple people through multiple prompts for every question. Context needs to be consistent and scalable.

In enterprise marketing, the context needed to interpret performance is spread across media plans, campaign briefs, budget spreadsheets, promotion calendars, test plans, performance benchmarks, and team knowledge. Atlas is designed to assemble that context as part of the investigation.

It does this in several ways.

1. It brings the relevant business context into the investigation

Atlas can incorporate campaign objectives, media plans, budget shifts, promotions, tests, briefs, benchmarks, and prior team decisions while the analysis is happening.

So instead of asking the user to restate what happened around the numbers, Atlas can apply the context that explains how performance should be interpreted.

2. It applies reusable business knowledge where it already exists

Some context should persist across investigations: brand profiles, audience definitions, channel roles, messaging strategy, benchmarks, market exceptions, and approved team decisions.

If the organisation has already confirmed how a market should be treated, what a channel is expected to do, or which benchmark applies, Atlas can carry that into the next investigation.

3. It incorporates files as context

A lot of marketing context lives outside the warehouse. Atlas can use uploaded files such as media plans, campaign briefs, spreadsheets, PDFs, test plans, and promotion calendars alongside structured data.

That matters because the explanation for a performance change often sits in the plan, not the table.

4. It carries context through the conversation

Once a user sets a brand, region, market, campaign group, date range, or comparison period, Atlas can keep that context active across follow-up questions.

The investigation does not reset every time the user asks the next question.

5. It captures reusable business knowledge over time

When a team confirms a benchmark, market exception, channel role, campaign convention, or recurring caveat, that knowledge can be captured and reused in future investigations.

Context is collected automatically, and admin users can review, approve, reject, or archive context as it’s captured.

That is how the system becomes more aligned with the business over time.

 

Why do generic AI tools struggle with context?

Generic AI tools are good at working with the information placed in front of them. The problem is that marketing context is rarely placed neatly in front of them.

You can wire your LLM straight into your data but it won’t automatically know:

  • Which campaign objective applies
  • Which promotion was live
  • Which budget changes happened recently
  • Which media plan shifts were intentional
  • Which tests are active, and how results should be read
  • Which markets have local rules, exceptions, or seasonal effects
  • Which audience or ICP the campaign is meant to reach
  • Which message, offer, or creative strategy was in market
  • Which channel role should shape interpretation
  • Which benchmark is relevant for this brand, market, channel, or objective
  • Which decisions or context from previous investigations should persist

It may make a plausible guess, but enterprise marketing analytics needs more than plausibility. The hard part is getting the right context into the investigation at the right time.

Generic AI tools often depend on one of three fragile routes:

The user explains the context in the prompt → The answer depends on whether the user knows what matters, remembers the relevant details, and describes them clearly every time.

Someone uploads or pastes the relevant material → The process becomes manual and inconsistent. The right media plan, campaign brief, benchmark, caveat, or prior decision may be missed, and the AI still has to work out what is relevant.

A team builds a custom layer around the model → This shifts the problem into maintenance and governance. The layer has to retrieve the right context, apply it to the right investigation, respect permissions, stay up to date, handle changing business rules, and connect the answer back to source across brands, markets, teams, campaigns, and reporting cycles.

This kind of custom build is difficult because the layer needs to do more than retrieving information, it needs to create rules explaining which context matters for marketing performance, when to apply it, how to carry it through a conversation, and how to make the result auditable.

How does Atlas simplify making AI reliable?

Atlas massively simplifies this process because business context is discovered, captured, governed, and made reusable as part of the Atlas architecture, rather than becoming a big, fragile build around your LLM. Atlas can pull context from files, conversations, prior investigations, and team inputs, then turn confirmed information into reusable knowledge.

This knowledge is automatically collected, but also centrally governed, meaning admin users can review, approve, reject, or archive context as it’s captured. Atlas checks for duplicates automatically, and overlapping concepts can be merged rather than left to conflict with each other, so the business context AI relies on stays accurate and under control as it grows, instead of decaying into a pile of inconsistent, half-remembered definitions.

Because Atlas is composable, that context can be applied across agents, workflows, and AI tools, so teams work from the same business understanding wherever analysis happens. So, data teams don’t have to build retrieval, permissioning, governance, and audit workflows from scratch every time the organisation wants to use a new model or AI tool - and marketers don’t have to restate the business picture in every prompt.

The result is faster decisions with fewer validation loops, because AI outputs are grounded in consistent business context, governed metric meaning, and are fully traceable.

 

Context is what makes the answer usable

The promise of AI in marketing analytics is faster decision-making. But speed creates value only when teams trust the interpretation.

Metric definitions make AI consistent before the query starts. Context makes AI reliable while the investigation is happening. It gives the system the information it needs to interpret performance as marketers understand it: as the result of campaigns, budgets, tests, channels, markets, and business decisions changing over time.

That is what turns AI from a fast reporting interface into a reliable way to understand marketing performance.