Contributors Dropdown icon

AI marketing reporting uses AI to summarize performance data across paid channels, identify notable changes, check data quality and suggest follow-up investigations. Digital marketing teams and marketing agencies can use it to prioritize weekly reviews across Google Ads, Facebook Ads and X Ads. This guide shows how to create a reusable Funnel AI skill for that job, so you can spot changes that need investigation before you act on them or change ad spend.

Imagine this hypothetical example: Google Ads conversions doubled last week. Most of the increase came from one UK Google Ads account, with 37 conversions attributed to a single day. What should you check before calling that a performance improvement?

That is the kind of question a useful weekly review should surface. It should show what changed, identify where to look, flag possible tracking or data quality issues, and suggest what to investigate next, so you can turn platform-level changes in marketing performance into timely action instead of missed anomalies or bad budget decisions. Many marketing agencies deliver monthly reports to help prove ROI, but a weekly review catches problems like this sooner.

You can create a Funnel AI skill, much like a Claude skill, for this recurring task. Define the reporting setup, add data checks, write the skill's steps, test the answer against your data, and customize the analysis for each platform before saving the skill for reuse.

In this guide, we'll build a skill around one question:

"What changed last week on each paid advertising platform, and what should we investigate?"

We'll use a fashion retailer demo account to illustrate what a useful answer looks like.

Where AI fits in marketing performance analytics

What AI can do in a weekly review

AI in digital marketing can scan performance metrics, summarize what moved and flag unexpected changes in marketing performance. That reduces manual work when marketing data flows into reporting processes. Think of this skill as a first pass you can run every week: it compares two weeks across marketing channels, checks the data and proposes what to investigate.

It is not an executive summary report, which condenses headline performance into one page for stakeholders, and it does not replace dashboard templates, which show the numbers. The skill explains what changed and what to check, so the findings can inform your marketing strategy and focus your marketing efforts.

What AI needs to know

AI does not automatically know your business goals, your target audience, how you define a conversion or whether someone changed a tracking tag during the comparison period. But you can tell it. Put that business information in your workspace Instructions, so it applies to every chat, or in the skill's reporting setup, such as conversion definitions, goals, markets and known tracking changes. The less the AI has to guess, the more the review can fit your marketing strategy.

What this skill does not do

AI in marketing covers more than weekly reviews. It can identify audience segments from behavior, improve customer targeting and personalization, and use historical signals for predictive analytics. Some tools also flag underperforming ads or optimize budget allocation dynamically. This skill does none of that. It reviews two complete weeks, flags what changed and suggests what to check. Because one week is thin evidence, it leaves decisions about pausing ads or moving budget to you. It also stays separate from SEO reports, which track organic visibility and user behavior over time.

Is marketing analytics going to be replaced by AI?

AI already handles much of the routine work: comparing weeks, running data checks and flagging what to investigate, which is what this skill does. As AI becomes more capable, expect it to take on more of the analysis, including more of the investigation itself. How far that goes is still uncertain, but marketers who learn to give AI the right information and steps will be best placed to turn performance data into actionable insights and data driven decisions.

The practical question is how to give AI those steps consistently, week after week. In Funnel, you do it with a skill.

Give the skill one clear job

Funnel AI skills are step-by-step rules that guide Funnel AI to perform a specific task. Funnel AI uses a skill when it is relevant to your request, or you can run it directly: type / in chat, select the skill and send your request. Learn about skills in Funnel AI.

Skills and Instructions: what is the difference?

Funnel AI has two ways to guide the agent. Instructions are rules it follows in every chat, so they suit information that always applies, such as how your team defines a conversion. A skill teaches it how to do one specific task, and it is used only when the task matches. Each skill has a name, a description that says when to use it and steps that describe how to do the task.

For this skill, the job is to compare two complete weeks, identify the main changes on each advertising platform and suggest up to three investigations.

Keep the task focused. A weekly review can identify a change worth examining and feed your marketing strategy, but establishing its cause may require more history or additional information. This review covers paid results, so search engine optimization and organic traffic are out of scope.

Define your reporting setup

Effective marketing reports need clear goals and KPIs, so start with the performance metrics and outcomes you want to understand on each platform. In performance marketing, where advertisers pay for results and common KPIs include CPC and CPL, accurate outcome counts matter most, which is why the steps ask you to confirm definitions and sources first. Confirm the exact metric fields, included accounts, campaign identifiers, reporting timezone and currency, including how accounts that report in other currencies are converted. This setup helps the skill read your marketing data consistently and ask useful questions when something essential is missing. Clicks and impressions can turn into vanity metrics when disconnected from outcomes, so supporting metrics should stay secondary unless they tie directly to the review's business objectives.

Keep each platform's outcomes separate

In digital advertising, each platform counts outcomes in its own way. Keep Google Ads, Facebook Ads and X Ads outcomes separate. Conversion definitions and attribution settings may differ, and multiple channels may claim credit for the same purchase. Grouping Facebook and X under Paid Social does not make their outcomes interchangeable, even though both are social media platforms. Google Ads covers search engine marketing as well as other campaign types, so it works differently again. That is why the skill reviews one platform at a time.

Marketing channels also differ in the target audience they reach, so a blended number across paid channels can hide more than it shows.

Check which sources feed each metric

To keep those reviews separate, check which data sources contribute to each metric. In our demo workspace, the shared revenue field contained values from Facebook Ads and Google Analytics, the web analytics source for website traffic. Filtering by source ensured the Facebook Ads review used the revenue generated according to Facebook Ads' own reporting and excluded Google Analytics outcomes. The review should also state which source supplied each outcome, so you can verify it.

Within a platform, combine accounts only when their reporting definitions and scope are compatible.

Copy and customize the skill steps

In Funnel AI, a skill has three fields to fill in: a skill name, a description that says when the AI should use it, and the steps.

Skill name: Weekly paid marketing review by platform

Description (when to use it): Use when the user asks what changed last week on each paid advertising platform, or asks for a weekly paid marketing review.

Steps: Replace the bracketed fields with your workspace details, then test the steps in chat. The next section shows how to fill in the outcome sources, the part most worth getting right.

Review weekly paid marketing performance in this Funnel workspace.

Answer:

What changed last week on each paid advertising platform,

and what should we investigate?

Reporting setup

- Included advertising platforms: [insert]

- Included accounts, markets and filters: [insert]

- Spend field: [exact field]

- Platform-specific outcome fields and definitions: [insert]

- Source that populates each outcome field: [insert]

- Supporting metrics: [clicks and impressions; others optional]

- Platform/source dimension: [exact field]

- Account dimension: [exact field]

- Channel dimension: [exact field]

- Platform-specific campaign name and ID fields: [insert]

- Reporting timezone: [confirm]

- Reporting currency: [currency]

 

Scope and definitions

Create a separate review for each advertising platform.

Keep outcomes separate even when platforms share a channel label.

Do not add together conversions or attributed revenue across

platforms. Do not calculate blended CPA, ROAS or AOV.

Use documented workspace definitions.

 

Confirm which sources populate each metric.

Do not infer a metric's source or meaning from its name alone.

If essential definitions or mappings are ambiguous, ask one

concise clarification message. Continue with unaffected metrics,

but withhold conclusions that depend on unresolved information.

Exclude Google Analytics outcomes and budget pacing unless

requested. If included, report them separately using their

own definitions. Do not count budget figures as actual spend.

 

Dates

Compare the last complete Monday-Sunday week with the preceding

Monday-Sunday week in the reporting timezone.

State both date ranges and exclude the current partial week.

Explain unresolved differences in account timezones.

Use the date basis (ad-interaction date or conversion date)

and conversion-delay notes from the outcome field's definition

and the workspace documentation. If the basis cannot be

confirmed, mark it as unconfirmed, treat the most recent days

as provisional and withhold conclusions that depend on when

conversions occurred. Use the same date basis and conversion

settings for both weeks.

 

Data checks

Check for missing dates, missing values and reporting gaps.

Do not treat missing data as zero.

If outcome fields return only zeros, check whether they represent

measured outcomes or unavailable tracking before interpreting them.

 

Check for overlapping data sources for the same advertising

account. Compare account identity, report type, level of detail,

filters, breakdowns and attribution settings. Where sources

overlap, use one source per metric and scope so values are not

counted twice, and keep sources that add different detail.

Treat matching names or daily spend as clues, not proof. State

which sources you used and ask the user to confirm when unsure.

Use a consistent currency and state any conversion method.

Show supplementary costs separately and check for overlap

with media spend.

 

Mention freshness or completeness limitations when they affect

the interpretation. Lower activity alone does not establish

that data is missing.

 

Analysis within each platform

1. Compare spend and verified outcome metrics.

2. Show both weeks, absolute change and percentage change.

If the previous value is zero, show percentage change as N/A.

3. Combine accounts only when definitions, attribution settings,

currency and reporting scope are compatible.

Otherwise, report those accounts separately.

4. Calculate documented ratios from matching totals.

Do not average campaign or account ratios.

If a denominator is zero or missing, show the ratio as N/A.

5. Use AOV only when its formula is verified and the denominator

counts purchase orders compatible with the revenue.

Otherwise, omit it.

6. Break down changes by market, account or channel where useful.

7. Identify up to three campaigns contributing most to the

largest absolute changes in spend or the chosen outcome.

Use stable campaign IDs with platform and account labels.

Do not match campaigns across platforms by name.

If IDs are unavailable, identify campaigns by name with

platform and account labels.

8. Check campaign fields in the relevant accounts before

concluding that campaign detail is unavailable.

Use account-level detail where necessary.

9. Inspect unusual daily concentrations. Include the previous-week

comparison before describing a contributor to weekly change.

 

Output

Format

Use a table for each platform's weekly comparison and short text

for everything else. Do not use charts unless asked.

 

Scope

Briefly state dates, platforms, accounts, currency and exclusions.

State which source supplied each outcome metric.

 

Separate platform reviews

For each platform, provide:

- Up to two observations supported by values

- A weekly comparison table

- Main contributors

 

Prioritized follow-up

Give up to three specific checks across the review.

Label each with its platform.

Keep data validation separate from performance investigation.

Tie each investigation to an observed change.

 

Checks performed

List the data checks you ran and any you could not run.

 

Interpretation

Mention unavailable detail only when it limits a finding

or follow-up.

Do not rank platforms by CPA or ROAS when their definitions

or attribution settings differ.

Separate observations from possible explanations.

Do not claim causation or statistical significance.

Do not invent targets or benchmarks.

Do not recommend budget changes solely from one week of results.

Keep the narrative concise.

 

 

Example: filling in the outcome sources

Use exact workspace metric and dimension names, and include field IDs where available to remove ambiguity. Name the source that feeds each outcome. Your field names will differ. In the demo workspace, the outcome lines looked like this:

- Platform-specific outcome fields:

Google Ads: conversions:platform

Facebook Ads: conversions:platform and revenue:platform

X Ads: confirm available outcome fields and tracking

- Sources that populate outcome fields:

revenue:platform receives values from Facebook Ads and

Google Analytics. Use it only with Facebook Ads sources

selected. Keep conversions:platform filtered to the

platform being reviewed.

 

 

In the demo workspace, the Conversions field pulls from a different source for each platform: Google Ads Conversions for Google Ads and Purchases for Facebook Ads. That is why the review keeps it filtered to one platform at a time.

If you leave a definition as "confirm", the steps tell the skill to ask for it and to withhold conclusions that depend on it, such as CPA and ROAS, until you answer. Keep the steps about how to analyze the data separate from previous findings, so last week's answer doesn't bias this week's review.

What a useful review looks like

We tested these steps against a fashion retailer demo workspace containing synthetic data, comparing Oct. 19-25 with Oct. 12-18, 2026. This excerpt of the Google Ads review shows one finding:

Metric

Oct. 12-18

Oct. 19-25

Change

% change

Spend in USD

$52,266.80

$53,466.25

+$1,199.45

+2.3%

Reported conversions

34

70

+36

+105.9%

UK account conversions

11

44

+33

+300%

Synthetic demo data, excerpt of the review. Monetary metrics are shown in USD using the workspace's currency settings.

Of the UK account's 44 reported conversions in the latest week, 37 were attributed to Oct. 23. Its other days that week recorded between zero and three, and its daily range the previous week was one to three, so the concentration stands out. The UK account also accounts for 33 of the 36 additional reported conversions.

That gives the marketer a focused starting point. The finding identifies where the reported increase occurred. Establishing whether it represents additional purchases or online sales requires validating the conversion definition and settings first, with revenue and conversion rate among the key metrics to review alongside the anomaly. Recent conversion totals can also change as delayed conversions arrive, so the review treats them as provisional. It helps to document once, in a field description or in your Instructions, which date each platform counts conversions on.

The same review a week later shows why this matters. Comparing Oct. 26-Nov. 1 with Oct. 19-25, reported Google Ads conversions fall 34%, from 70 to 46, mostly because the 37-conversion day sits in the earlier week. Without that day, the earlier week would show 33 conversions. A week-over-week drop can reflect a one-day spike in the earlier week as much as a real decline, so the steps ask the skill to inspect daily concentrations first. In that run, the 37 conversions were one each across 37 Search campaigns, which looks more like an import or tracking pattern than 37 separate customer decisions, so the review suggested checking them by conversion action.

Scope and Google Ads (Oct. 26–Nov. 1)Funnel AI review of Facebook Ads and X Ads for Oct. 26 to Nov. 1 with comparison tables and main contributors, and no conversion data for X Ads

Oct. 26-Nov. 1 run in Funnel AI chat.

How the skill calculates week-over-week change

The skill compares each week with the one before it. Week-over-week change is this week's value minus last week's value, divided by last week's value. Conversions rising from 34 to 70 is a change of 36, or about 106%. When last week's value is zero, the percentage is shown as N/A. Ratios such as cost per click and cost per conversion are calculated from weekly totals, not by averaging campaign ratios.

What the review did with Facebook Ads and X Ads

Facebook Ads and X Ads each received a separate review, keeping their outcomes distinct from Google Ads. The workspace has no conversion data for X Ads, so the review assessed delivery only and did not report X Ads as having zero conversions, because missing data is not the same as zero. Campaign IDs were not available in the demo workspace, so the review identified campaigns by name plus account. That is reliable within one account but would merge campaigns that share a name and split one that was renamed, so use campaign IDs where you have them.

A prioritized follow-up

The review closed with up to three prioritized checks, each labeled by platform and tied to a finding, with data validation kept separate from performance investigation, so the report ends with next steps you can act on. In a fuller executive summary report, these checks would sit beneath the headline numbers. Two examples:

Google Ads, data validation: UK reported conversions rose from 11 to 44, with 37 attributed to Oct. 23. Confirm which conversion actions were included and whether tracking or conversion settings changed.

Google Ads, performance investigation: Once the metric is validated, identify which paid campaigns contributed to the increase.

Two prioritized checks from the Funnel AI review: a Google Ads data check on the UK conversions on 23 Oct and a Facebook Ads performance check on DE spend

Oct. 19-25 run in Funnel AI chat.

Check the answer before saving the skill

Compare the review's dates, totals and definitions, including cost per conversion, with a report you trust. Then check three things.

Sources

Confirm that the review states which source supplied each outcome, and that it matches your field definitions, so a field's name is not mistaken for its origin.

Largest single day

Look at the biggest daily value behind any large weekly change, and compare it with the previous week.

Follow-ups

Check that each suggested investigation follows from a finding. The review points to what to check. Decisions about ad spend, budget allocation or marketing strategy need more than one week of results.

The review also lists the data checks it ran and any it could not run. Read that list before relying on its conclusions. Refine the steps if the answer uses the wrong metrics, misses useful detail or draws conclusions the data cannot support. Test any changes before sharing the skill with your team.

Checks performed list from the Funnel AI review covering outcome sources, ratios, missing dates, duplicate connections, supplementary costs, freshness and campaign detail

Oct. 19-25 run in Funnel AI chat.

Save the skill for reuse

Once the review produces a useful, checked result:

  1. Select AI in the left navigation, then Skills under AI Context.
  2. Select Add skill, then Write a skill. The dialog is titled "Create a skill".
  3. Enter the skill name.
  4. Choose who can use it: everyone in the workspace, or just you.
  5. In Description, say when the AI should use the skill.
  6. In Steps, paste the tested steps. Steps can be up to 10,000 characters.
  7. Save the skill.

The Add skill menu also lets you create a skill with Funnel AI or upload one. You can also ask Funnel AI in chat to save a task as a skill, and approve it on the Save skill card. You need workspace edit permission to create a workspace skill. To run a saved skill in chat, type /, select the skill and send the prompt. See the setup guide.

Funnel AI Skills page with the Add skill menu open, showing Create with Funnel AI, Write a skill and Upload a skillCreate a skill dialog in Funnel AI with fields for skill name, who can use the skill, description and steps

Run the skill every week with a scheduled task

Saving a skill does not schedule it. To automate the weekly review, create a scheduled task that uses the skill.

  1. Open Scheduled tasks in the AI menu and select Create manually.
  2. Name it, for example "Weekly paid marketing review".
  3. Select Add skill and choose your saved skill. It is added to the prompt, where you can add anything specific to this run.
  4. Choose when you want to receive the updates: weekly, a day and time, and a timezone that matches your reporting timezone.
  5. Choose how to receive the results. They are saved to a Funnel AI chat, and you can also send them by email or, once Funnel AI is connected to Slack, to Slack.
  6. Select Test run to preview the result, then create the task.

A scheduled run starts a new chat, so keep definitions in the skill or your Instructions, not only in your test chat. Leave the option that lets the task make changes switched off, because the review only reads data. Someone on your team should still review the findings. You can then ask Funnel AI to dig into the changes using your Funnel data, for example by breaking the Oct. 23 conversions down by conversion name. Budget edits, bid changes, paused ads and change history live in each ad platform, so check those there.

Funnel AI Scheduled tasks page with the create menu open, showing Create with Funnel AI and Create manually

New scheduled task form in Funnel AI with name, prompt, Add skill option, schedule, email delivery and recipients

Frequently asked questions

What is the difference between Instructions and skills in Funnel AI?

Instructions are rules the AI follows in every chat. A skill teaches the AI how to do one specific task, and it is used only when a request matches the skill's description or when you ask for it.

What is attribution in digital marketing?

Attribution is how a platform decides which of your marketing channels or ads gets credit for a conversion along the customer journey. Each of the three platforms applies its own attribution settings, so two platforms can claim the same sale. That is why this skill never adds conversions across platforms.

What is a conversion in a Google Ads report?

A conversion is the desired action you have defined as valuable, such as a purchase or sign-up, that Google Ads counts after someone interacts with an ad. Which actions count depends on your account's conversion settings, so confirm the definition before comparing weeks.

What should you check when Google Ads conversions spike?

Start by checking which conversion actions are included in the conversions column, whether their counting settings or tags changed and whether the same event is firing twice. Remember that recent totals can shift as delayed conversions arrive. Then compare the Google Ads report with orders or Google Analytics website traffic and sales data as a cross-check, not an expectation of identical totals, because reporting and attribution differ. A one-day spike, like the 37 attributed conversions in our example, or a cost per conversion that halves overnight, is worth checking before you treat it as real. In search campaigns, keyword performance can also show which terms were behind a spike.

How do you write AI prompts for marketing?

Be specific about the fields, dates and definitions to use, describe your target audience, business goals and marketing strategy, state what the answer should contain and tell the AI what not to do. This skill, for example, says not to guess the outcome metric and to ask one clarification question when a definition is missing.

What are common data quality issues in marketing reports?

Typical issues in digital advertising reporting include missing dates, overlapping data sources for the same account, mixed currencies, zero values that mean missing tracking and performance metrics defined differently across platforms. The steps include checks for these issues, such as comparing overlapping data sources for the same account and asking you to confirm when unsure, before interpreting changes in marketing performance. Read the review's list of checks performed before relying on its conclusions.

Start with your next weekly review. Check the findings and choose the first question worth investigating.

Contributors Dropdown icon
Want to work smarter with your marketing data? Discover Funnel