Contact

Tools · AI

Claude

The legwork goes faster. A person still makes the call.

Claude is Anthropic's AI model and assistant, used through the Claude apps, Claude Code and the Claude API. Donut Analytics uses it every day for the hands-on work: writing SQL and Python, checking data against its source and drafting analysis. jelly, our AI ad manager, runs on it too. A person checks the numbers before they reach a client and makes the call.

Let’s Chat All tools

How we use it

How we use Claude

  • The hands-on work

    Claude writes the SQL and Python, checks the data and drafts the analysis. It starts each job from a written record of the account: its notes, open questions and past decisions. Context carries over from one day to the next instead of starting cold.

  • Scheduled jobs that check themselves

    Scheduled jobs load data and refresh reports on their own. Each one checks its numbers against the source and says so when something is off, like a feed that stopped updating or a total that doesn't match the platform.

  • Reviewers that try to break the answer

    For bigger calls, separate AI reviewers rework the numbers from the raw data and try to knock the conclusion down. What survives goes to a person, and only then to you.

  • jelly, our AI ad manager

    jelly runs on Claude. It drafts campaigns and copy that fit each platform's limits, looks at your creative to pick assets, reviews its own plan before showing it, and flags spend spikes and tired creative. Everything it builds is paused.

  • Setup work, with a person alongside

    For setup inside ad and analytics tools, Claude can do the clicking in the browser while a person works alongside it.

  • Read-only until a person says go

    Our AI tools are read-only until a person says otherwise. They draft and check. Launching ads, changing budgets, publishing tags and sending client email all wait for a person's go-ahead.

In practice

Where it shows up in the work

Real work, kept anonymous.

  1. 01

    A store-visit estimate, stress-tested

    For one client, we estimated store visits that Google hadn't finished reporting. Before we recommended anything, independent AI reviewers recomputed the numbers from the raw data files, each looking for a different way the estimate could be wrong.

  2. 02

    A model refresh, checked before it ships

    For a Florida retailer's monthly media mix model refresh, Claude runs the fit and the checks once the new month's data is loaded, and separate reviews read the result before a person publishes it.

    Read more
  3. 03

    A campaign plan that reviews itself

    When jelly drafts a Meta campaign, a second pass critiques the plan and flags weak spots on the plan card before anything is built. The build is paused until a person turns it on.

    Read more

What it can’t tell you

What Claude can't do for you

  • Know your business on its own

    Claude knows what it's given. Without the account's notes, data and past decisions, its analysis is generic. That's why each account has a written record, and why it works from the source data.

  • Be right every time

    AI can write a query that runs cleanly and still answers the wrong question. Numbers get checked against the source and read by a person before they go to a client.

  • Make the call

    Whether to move budget, pause a channel or change a tag depends on things outside the data, like your margins, your calendar and your team. A person decides.

FAQ

Questions about Claude

How does Donut Analytics use Claude?
For the hands-on work: writing SQL and Python, checking data against its source, drafting analysis and running scheduled data jobs. jelly, our AI ad manager, also runs on Claude. A person checks the work and makes the call.
Does AI replace a marketing analyst?
No. At Donut Analytics, AI does the legwork faster: queries, checks and first drafts. A person decides which question matters, checks the answer and makes the call.
How do you check AI-generated analysis for errors?
Three ways. Numbers are checked against the source, like the ad platform itself. For bigger calls, separate AI reviewers rework the result from the raw data and try to break it. Then a person reads it before it goes to you.
Can Claude analyze GA4 or ad platform data?
Yes, once the data is somewhere it can read. We keep ad platform and sales data in BigQuery, and Claude writes the SQL and Python that read it. It's only as good as the data underneath, so we check that first.
Can AI launch or change my campaigns?
Not without a person. At Donut Analytics, AI tools are read-only until a person says otherwise. jelly builds campaigns paused, and nothing spends until someone turns it on in the ad manager.
Do you use Claude or ChatGPT for analysis?
Donut Analytics runs its day-to-day analysis and build work on Claude, and jelly runs on it too. The model matters less than the process around it: clean data, a written record of the account, and a person checking the result.

Pairs with

All tools

Faster answers, checked by a person.

Tell us what you’re working on. We’ll build a measurement-first plan for your brand. Email hello@donutanalytics.com or call (407) 394-9720.

Let’s Chat See our results