The "Shadow Analyst" Strategy: Turning Boring Data into Content Gold

There is something about a data point and a chart that burns a path right through your prospect’s occipital nerve and into their brain.

Not lazy, slop data.

Real, relevant data that is:

1. Credible
2. Interesting
3. Helps your prospect understand or navigate their world better

Like a chart showing exactly how much freight rates spike the week before a major holiday, pulled straight from your own transaction data — the kind of number a reporter would actually want to cite, not a stat you found on page four of a Google search.

But for this to be effective as a marketing motion, it also needs to be somehow deeply related to what the brand is selling.

Which is why this is perfect — it’s the mechanics behind the Hard Signal Protocol and the three-tier data marketing framework I’ve written about before. Those posts cover the strategy. This one covers the how — the actual files and prompts I use to get from “we probably have interesting data” to a drafted post.

But Getting This Type of Data Is Hard

There’s a bottleneck that historically made it hard to get these types of data points. Well, there’s two.

1. The first is cultural. It’s hard to give up internal, proprietery and even customer-specific information for free. Get over it. Chock it up to CAC.
2. Most marketing teams are sitting on many gigabytes or terabytes of gold insights…but its locked in a dungeon (Snowflake or BigQuery) guarded by Mike from the data science team who hasn’t answered your Slack messages since 2023.

Here’s a playbook that will help you repeatedly:

1. Brainstorm on potential insights to mind
2. Actually extract the data points and analyze them
3. Summarize
4. Write a draft post

Net time: ±10 minutes / insight

What you need:

1. Cursor/Windsurf/Claude Code
2. BigQuery access
3. A strong editorial muscle
4. Critical thinking
5. A BigQuery service key 😬
6. Permission to use the data however you’re planning to

The Shadow Analyst Strategy

This process will gives every marketer the ability to easily brainstorm potential extractable insights, run the data analysis on demand and crank out some initial storytelling tailored to your specific persona.

No code required.

The goal here isn’t to fire your data scientists.

It’s to free them from your “hey, got a second?.” By using AI as a translator between natural language and database code, you can extract unique insights from boring operational data.

The 5-Stage Execution Framework

Stage 1: Map the Territory

Before you can analyze, you need to know what exists. You don’t need to know how to query the database, but you need to know what tables are available in the first place, what it means and how to access it.

• Stalk the Dashboards: Look at existing BI tools (Tableau, Looker). Which charts does the CEO obsess over? That’s your goldmine. Try to find the underlying datasets that feed it.
• Find the Volume: Focus on datasets with millions of rows. Big data means statistical significance and the ability to spot micro-trends.
• Ask the Locals: Ask your product team, “Which table holds the ’truth’ about user behavior?”. Ask the data team which tables are being accessed the most.

Then buy your IT department a donut and get read-only access with a service key. This part may take the longest.

Stage 2: Create the “Rosetta Stone”

Remember, we’re building a replicable process so we’ll do it invest once. We’re going to create a plain-English guide that explains your complex database to the AI. Share the dataset name with Cursor, together with your service key and ask it to:

Pull dataset X from BigQuery, together with a sample of 10,000 recent rows. Then create a plain english overview of exactly what the dataset contains include the schema, business context, and general overview.

Now review the output.

Spend time making sure it’s perfect. Add in any business context or explanations you know. Spend some time explaining where this fits into your business. More time here is less time down the line, I promise.

Then rinse and repeat with 2-3 other mission-critical datasets.

You’re left with your first core file The Dataset Overview. This is a reusable “Master Guide” for that dataset. Future you will thank present you. I prefer to bundle all my datasets in one guide. I’ll show you why later.

You’re now going to create two more core files:

1. Target Audience: One file that explains who your core target audience is and what type of data insights you want to provide them. Basically, a combination of your ICP + how you want to meet them
2. Research Project Structure: This file explains what you want Cursor to do every time you brainstorm a project. For me, it includes rules like create a new folder, create one file with the analysis output, another with a neutral summary that also lists limitations and potential errors, and a third, based on the output and target audience document, with the draft post.
3. Output Goal: This is a sample and overview of what you want the final draft post sample to look like.

Okay, you’re done with the prep work. Save these files as Cursor rules and get going.

Stage 3: Ideate with Context

Now that the AI understands your data structure and what you want to get, you can start thinking.

“Read the dataset guide and the target audience file. I want to write a blog post about Seasonality in B2B buying. Give me 10 questions we could answer using this data.”*

• The Goal: Find the weird stuff.
  • “When is the ‘booking dead zone’?”
  • “Which regions are price-sensitive during the holidays?”
  • “What is the optimal window to buy before prices surge?”

Go back and forth. Push the LLM to come up with good ideas. Bring your own. Make sure the questions are reasonable. I strongly recommend pairing this with a document about your user persons and what you like to educate them about — the same target-audience thinking I walk through in the Expertise Amplification Method.

Stage 4: Trust but Verify (The “Don’t Get Fired” Phase)

This stage turns ideas into hard numbers.

1. Generate SQL: Ask the AI: “Write the SQL query to extract the data for idea #3.”
2. Run the Query: Copy the code into your data warehouse and download the results.
3. The Feedback Loop: Upload the results back to the AI.
   • Step A: “Analyze these raw numbers.”
   • Step B (Crucial): “Write two alternative queries to double-check this math. Did we accidentally count cancelled orders?”

Warning: AI is like a confident toddler. It will hallucinate logic. Always ask for the raw numbers and sanity check them manually. More importantly, if you are using Cursor and have a service key for API access, you can literally just ask Cursor to run this query for you.

Stage 5: Craft the Narrative

With verified data in hand, you can now write content that no competitor can rival.

Don’t just dump a chart and call it a day. Find the story. Is the market volatile? Are users becoming impatient? Use AI to draft the structure, but apply your human editorial oversight to ensure it sounds like you, not a robot.

To accelerate this process, I like to pair this with two other system documents that I have saved in Cursor:

1. A general playbook for how to approach each research process. Things like:
   1. Start a new folder for each research
   2. Keep one file for the analysis, one for takeaways, and another for the written post
   3. What type of insights to focus on
   4. What to not ignore
   5. What calculations were made along the way
2. An overview of how I like the output file drafts to be written

The Human Element (Yes, You’re Still Needed)

This framework fails without three specific human skills:

1. The Editorial Eye 🧐

AI can find patterns, but it cannot determine value. A correlation between “Monday mornings” and “login failures” might be statistically significant, but is it interesting? You decide.

2. Domain Knowledge

• Bad AI Inference: “Sales dropped to zero on December 25th.”
• Human Context: “That’s because we’re closed on Christmas, you glorified toaster.”

3. Critical Skepticism

You must push back against the AI. If a number looks too good to be true, it probably is. You are the Quality Assurance lead for your data.

The Strategic Advantage

By operationalizing your own data, you move from “sharing opinions” to “proving facts.” You aren’t just curating other people’s research; you’re publishing primary research.

And the best part? You can react to market shifts in hours, not weeks.

This transforms your data warehouse from a dusty storage locker into a high-octane publishing engine. Start your engines. 🏎️

This is the tactical companion to two other posts: the three-tier framework for turning data into an owned audience, and the Hard Signal Protocol for picking data worth publishing in the first place. If you want to see this playbook applied to a real dataset, the Ramp data marketing teardown is a good next read.

Frequently Asked Questions