Skip to content
AI stack, job

AI for marketing analytics and reporting

Last verified

In marketing analytics, data access matters more than which model you pick. Give the agent read-only access to ad platforms, analytics and payment data, make it write the query or script it used, and reconcile totals against the source dashboards before anyone acts on a chart.

The workflow for this job
The workflow for this jobFour steps in a loop: the model proposes, the output is checked, a person approves, then it ships and is measured, and what you learn goes back into the next brief.01 MODEL PROPOSESQuery and firstread02 CHECKTotals reconcile03 PERSON APPROVESReads the query04 SHIP AND MEASUREReport, thendecideNEXT BRIEFThe workflow for this jobFour steps in a loop: the model proposes, the output is checked, a person approves, then it ships and is measured, and what you learn goes back into the next brief.01 MODEL PROPOSESQuery and first read02 CHECKTotals reconcile03 PERSON APPROVESReads the query04 SHIP AND MEASUREReport, then decide
Illustrative flow, simplified.
The model proposes, a check catches what it got wrong, a person approves, and the result ships and is measured. What you learn feeds the next brief.

Why data access decides the result

Marketing analytics is a data problem before it is a model problem. A model without a connection to your numbers answers from a pasted screenshot or last week's export, and it cannot tell you the export was stale. A model with read access can write the query itself, run it against the source and show you the code.

Three sources cover most web funnel and app reporting: ad platform spend and clicks, analytics events, and actual payments. If the platforms disagree with each other, payments settle the question, because a payment either cleared or it did not.

Setup: read-only connectors and a payments export

Give the agent the narrowest access that answers the question. Reporting needs reads only, so keep write scopes off the credentials you hand over.

  1. 1.Create read-only credentials per source

    Use a viewer role for analytics, a reporting-only token for each ad platform and a restricted key with read permissions for payments. Store them in an env file and load them at runtime, never in the prompt.

    node --env-file=.env scripts/pull-report.js
  2. 2.Connect analytics and ad platforms through MCP servers

    The Google Analytics MCP server is published by the platform vendor and is described as read-only reporting. For Google Ads, the google/skills repo includes Ads API MCP setup guidance. The Search Console connector iannuttall/seo was not verified in this check, so read its licence and scopes first. Some Meta Ads servers can launch campaigns and upload creatives, so strip or disable those tools for a reporting agent.

  3. 3.Export payments on a schedule

    Pull successful payments, refunds and timestamps into a sheet or database with a fixed date range and a fixed timezone. Most reconciliation gaps start with a timezone or window mismatch.

  4. 4.Log every query

    Save the SQL or script next to each chart, with the date range and the source it ran against. A number without its query is an opinion.

The table lists MCP servers for this job. Read the access, licence and safety notes before connecting anything, because third-party servers are unreviewed code until you check them.

MCP servers for this job
NameJobPublisherAccessLicenceChecked
googleanalytics/google-analytics-mcpAnalyticsPlatform vendorNot reviewedApache-2.0Class A
surendranb/google-analytics-mcpAnalyticsThird partyNot reviewedMITClass A
ysntony/tiktok-ads-mcpPaid UA agents, AnalyticsThird partyNot reviewedMITClass A
pipeboard-co/meta-ads-mcpPaid UA agents, Analytics, Static creativesThird partyNot reviewedBUSL-1.1Class B

Sanity checks before anyone acts on a chart

An agent produces a confident chart whether or not the data underneath is right. Run the checks below on every report, and have the agent print the query so you can read it.

Sanity checks for AI-built reports
CheckHow to run itStop if
Spend totalCompare the agent's spend to the ad platform dashboard for the same dates and timezone.The totals differ and the gap has no explanation.
Conversions vs paymentsCompare platform-reported purchases to successful payments in the payments export.Platform count is far above payments with no known cause.
Query shownAsk for the SQL or script used, then read the filters and joins.Filters are missing, or a join can multiply rows.
Date windowConfirm start, end, timezone and attribution window on every source.Sources use different windows.
Cohort denominatorCheck that each rate uses the size of its own cohort or subgroup as the base.A percentage is applied to the original total instead.
Refunds and feesConfirm whether revenue is gross or net of refunds and processor fees.The report does not say which.
Spot checkRecompute one row by hand from the raw export.The hand figure and the agent figure disagree.

Skill packs can encode these checks as written rules. Treat each one as unreviewed code: skills may bundle scripts, and licences and write scopes differ by repository.

Skills for this job
NameJobLicenceChecked
coreyhaines31/marketingskillsAd copy, Landing pages, Analytics, SEO and GEOMITClass A
google/skillsPaid UA agents, AnalyticsApache-2.0Class A
AgriciDaniel/claude-adsPaid UA agents, AnalyticsMITClass A
autonnel/autonnel-skillsLanding pages, AnalyticsApache-2.0Class A
ua5er/user-acquisition-skillsPaid UA agents, Analytics, Landing pagesMITClass A
manojbajaj95/claude-gtm-pluginAnalyticsMITClass A
amplitude/mcp-marketplaceAnalyticsMITClass A
wshobson/agentsAnalyticsMITClass B

Worked check with illustrative numbers

These figures are illustrative, not real data. Say you spend $10,000 on a web funnel with a $50 first payment, and the ad platform reports 500 purchases. In this illustrative case, platform revenue is 500 x $50 = $25,000, so platform ROAS is $25,000 / $10,000 = 2.5.

Say the payments export in this illustrative example shows 400 successful payments. In the illustrative case, payment-based revenue is then 400 x $50 = $20,000, so ROAS is $20,000 / $10,000 = 2.0. In the same illustrative case, the gap is 500 - 400 = 100 purchases, or 100 / 500 = 20% of the platform count. Until you find the cause (duplicate events, a longer attribution window, failed charges), report 2.0.

Now a cohort question, still illustrative. Say 25% of those 400 payers renew in month two. Apply 25% to 400, not to 500: 400 x 0.25 = 100 renewals in this illustrative cohort. At $50 each that adds 100 x $50 = $5,000, so illustrative cumulative revenue is $20,000 + $5,000 = $25,000 and cumulative ROAS is $25,000 / $10,000 = 2.5. This is gross revenue before refunds, fees and other costs.

Open-source BI board for web funnels

The Web-to-Web BI Board (ua5er/web2web-bi-board) is a static BI dashboard for subscription web funnels. It shows cohort ROAS, LTV and payback, is MIT licensed, and is published by the site owner. It is linked from the open-source page.

Use it as a reference view. If the agent's cohort ROAS differs from the board's on the same export, one of the two has a filter or denominator error, and the diff tells you where to look. The safety scan found that analytics.js calls fetch for its own analytics, so check that before hosting it.

Model options from the data

No clear leader exists for data analysis and SQL in this check, and tool use matters more than the raw model. Split by job: a cheap tier for routine pulls, tagging and formatting, and a top tier for open-ended analysis. Prices below were read on 15 November 2026, and several have dated changes, so check the vendor page before budgeting.

  • Cheap tier: Claude Haiku 5.5 lists from $0.10 input and from $0.50 output per MTok, and GPT-6 Luna lists $0.10 / $0.50. Gemini 3.8 Flash lists $0.75 / $3.75 until 2026-12-31, then $1.50 / $7.50.
  • Middle tier: Claude Sonnet 5.5 lists $2 / $10 and is described for SQL and spreadsheet analysis. GPT-6.1 Sol lists $2 / $10.
  • Top tier: Claude Fable 5.2 lists $8 / $40 and GPT-6 Astra lists $10 / $50. Gemini 4 Carbon lists $2 / $10 until 2026-11-30, then $4 / $20.
  • Test any tier on your own export with the checks above. No accuracy figure for SQL is claimed here.

The table shows every model listed for this job, with access, limits and the date each was checked.

Models for marketing analytics and reporting
NameVendorJobStatusPrice as the vendor states itCheckedSource
Claude Fable 5.2claude-fable-5-2.AnthropicAd copy, Market research, Analytics, Paid UA agentsCurrent$8 input / $40 output per MTok.AnthropicAnthropic
Claude Opus 5.5claude-opus-5-5.AnthropicAd copy, Market research, Analytics, Paid UA agentsCurrent$4 input / $20 output per MTok.AnthropicAnthropic
Claude Sonnet 5.5claude-sonnet-5-5.AnthropicAd copy, Market research, Analytics, Paid UA agentsCurrent$2 input / $10 output per MTok.AnthropicAnthropic
Claude Haiku 5.5claude-haiku-5-5.AnthropicAd copy, Market research, Analytics, Paid UA agentsCurrentFrom $0.10 input / from $0.50 output per MTok.AnthropicAnthropic
GPT-6 Astragpt-6-astraOpenAIAd copy, Market research, Analytics, Paid UA agentsCurrent$10 input / $50 output per MTok (standard, short context).OpenAIOpenAI
GPT-6.1 Solgpt-6.1-solOpenAIAd copy, Market research, Analytics, Paid UA agentsCurrent$2 input / $10 output per MTok.OpenAIOpenAI
GPT-6.1 Terragpt-6.1-terraOpenAIAd copy, Market research, AnalyticsCurrent$1 input / $5 output per MTok.OpenAIOpenAI
GPT-6 Lunagpt-6-lunaOpenAIAd copy, Market research, AnalyticsCurrent$0.10 input / $0.50 output per MTok.OpenAIOpenAI
OpenAI audio modelsOpenAIVideo and UGC, AnalyticsCurrentSee vendorOpenAI
Gemini 4 Carbongemini-4-carbon.GoogleAd copy, Market research, Analytics, Paid UA agentsCurrent$2 in / $10 out per MTok until 2026-11-30, then $4 / $20.Google, stated until 30 November 2026Google
Gemini 4 Argongemini-4-argon.GoogleAd copy, Market research, Analytics, Paid UA agentsCurrent$2 in / $10 out per MTok, rising to $4 / $20 later per page.GoogleGoogle
Gemini 4 Hydrogengemini-4-hydrogen.GoogleAd copy, Market research, Analytics, Paid UA agentsCurrent$1.50 in / $7.50 out per MTok.GoogleGoogle
Gemini 3.8 Flashgemini-3-8-flash.GoogleAd copy, Market research, Analytics, Paid UA agentsCurrent$0.75 in / $3.75 out per MTok until 2026-12-31, then $1.50 / $7.50.Google, stated until 31 December 2026Google
Grok 4.7grok-4.7xAIAd copy, Market research, Analytics, Paid UA agentsCurrent$2 in / $0.50 cached / $6 out per MTok (up to 200k prompt), $4 / $12 above. 500k context, knowledge cutoff May 2026.xAIxAI
Muse Spark 1.1muse-spark-1.1MetaAd copy, Market research, AnalyticsPreviewSee vendordev.meta.ai
Qwen models (Model Studio)AlibabaAd copy, Market research, Static creatives, AnalyticsCurrentSee vendorAlibaba
JevTypeSafe AIAd copy, Market research, AnalyticsLimited access$42 per billion input tokens ($0.042 per million), input tokens only; output tokens free (currently).TypeSafe AITypeSafe AI

Questions

Which AI model should you use for marketing analytics?
The facts name no clear leader for data analysis and SQL. Data access, a visible query and reconciliation against source dashboards matter more than the model name. Use a cheap tier for routine pulls, a stronger tier for open analysis, and test both on your own export.
Can AI analyse GA4 data directly?
Yes, through a GA4 MCP server or the Analytics Data API with read-only credentials. The agent writes the report request, and you check totals against the GA4 interface. Sampling, thresholds and date windows can differ, so confirm them before trusting a trend.
Should the agent be allowed to change campaigns?
For reporting, no. Use read-only tokens and disable write tools, since some ad MCP servers can launch campaigns or upload creatives. Add write access later only for a defined task, with preview before write and a human approving each change.
How can you check an AI-built marketing report?
Ask for the exact query or script, compare spend and conversions to the source dashboards, then compare platform purchases to actual payments. Recompute one row by hand. If cohort percentages are off, check the denominator each one uses.

Sources

Each source shows the date it was last checked.

Show all 35 sources
  1. Meta ai.meta.com/blog/muse-spark-1-1/
  2. Google aistudio.google.com/docs/models
  3. Google aistudio.google.com/docs/pricing
  4. Alibaba alibabacloud.com/help/en/model-studio/models
  5. Anthropic claude.com/pricing
  6. dev.meta.ai dev.meta.ai/resources/blog/build-with-muse-spark/
  7. OpenAI developers.openai.com/api/docs/models
  8. OpenAI developers.openai.com/api/docs/pricing
  9. TypeSafe AI docs.typesafe.ai/api/system-one.md
  10. TypeSafe AI docs.typesafe.ai/introduction/quickstart.md
  11. TypeSafe AI docs.typesafe.ai/models.md
  12. xAI docs.x.ai/docs/models
  13. GitHub github.com/amplitude/mcp-marketplace
  14. GitHub github.com/google/skills
  15. GitHub github.com/googleanalytics/google-analytics-mcp
  16. GitHub github.com/iannuttall/seo
  17. GitHub github.com/manojbajaj95/claude-gtm-plugin
  18. GitHub github.com/pipeboard-co/meta-ads-mcp
  19. GitHub github.com/surendranb/google-analytics-mcp
  20. GitHub github.com/ua5er/web2web-bi-board
  21. GitHub github.com/wshobson/agents
  22. GitHub github.com/ysntony/tiktok-ads-mcp
  23. learn.chatgpt.com learn.chatgpt.com/docs/pricing
  24. GitHub AgriciDaniel/claude-ads: LICENSE
  25. GitHub autonnel/autonnel-skills: LICENSE
  26. GitHub coreyhaines31/marketingskills: LICENSE
  27. GitHub ua5er/user-acquisition-skills: LICENSE
  28. Anthropic platform.claude.com/docs/en/about-claude/pricing
  29. Anthropic platform.claude.com/docs/en/about-claude/models/overview
  30. GitHub AgriciDaniel/claude-ads: README.md
  31. GitHub AgriciDaniel/claude-ads
  32. GitHub autonnel/autonnel-skills
  33. GitHub coreyhaines31/marketingskills
  34. GitHub ua5er/user-acquisition-skills
  35. TypeSafe AI typesafe.ai

Need this wired into a web-to-app funnel? Start with a short audit.

See the service