End-to-End Autonomous Data Analysis

Above is an illustration of an end-to-end analytical workflow. I've been putting puzzle pieces together in VerbaGPT to be able to run such a workflow autonomously. This post is about the first time I was able to test it end to end.

For the example, I wanted to use 2023–2024 Affordable Care Act (ACA) Minimum Loss Ratio (MLR) data to extract interesting insights and patterns. This is not a simple dataset. It's a set of a dozen different files for each year where you need complex joins, column meanings that depend on row context, industry jargon, messy self-reported data, and subtly misleading naming conventions. Not simple.

I asked VerbaGPT in local mode to:

• Download 2023–2024 MLR data from the CMS website
• Understand the documentation, extract and organize the files locally
• Securely load it into a Snowflake database
• Inspect the loaded data, identify and remove unreasonable values
• Analyze the data to answer a complex question
• Do a deep review of all code and narrative

All in one go. That took about 20 minutes. A lot of instruction, a lot of steps — but in VerbaGPT it's wrapped in a reusable "project." Once created, steps are trivial to edit, and future runs inherit all that context automatically.

The big idea: for any repeatable, complex analysis or deliverable, you can set up a workflow that produces high-quality output reliably. Due to the probabilistic nature of LLM generations, you don't get the same output twice. This is a feature and can be a bug. The right LLM harness increases the odds of the former.

A Note on Context

LLM results deteriorate when too much context is added at once — much like giving a person a 20-point instruction. They forget 30% of it instantly. This is a key reason the project framework in VerbaGPT matters. Steps run sequentially without biting off more context than the LLM can handle, building on prior steps — with optional human or AI review after each step for accuracy and reasonableness.

I needed to give it an analysis "skill" — relatively simple notes that are progressively loaded by the LLM at runtime — so that it defined profitability the way I wanted, and avoided a couple of data issues it tripped on the first time. For example: it summed rows including "grand total" rows, effectively doubling member months. Once that was corrected in the skill, it didn't make the same mistake again.

Surprisingly, I didn't need to fix anything for data ETL. It encountered hiccups, recovered automatically, and correctly loaded the data into Snowflake.

Results

Here is a sampling of the visuals and insights the analysis produced. First, the question I asked:

In case that's hard to read, the question was:

Project: MLR Analysis — Step 3/3. Analysis of 2024 data. I have 2024 MLR data. Can you analyze and give me 5–10 interesting insights for small group and individual ACA markets? Include a variety of visuals. Use mlr skill. Ideas: waterfall from premium to profit (pmpm basis), level of competition in a state, trends in pmpm profitability for risk adjustment payers vs. receivers, variance in admin as % of premium, relationship between plan size and profitability.

Here are a few of the resulting charts:

Chart 1: Risk adjustment transfer receivers are more likely to be profitable — a counterintuitive finding, since risk adjustment methodologies typically undercompensate riskier members. This has more to do with scale, expertise in documenting diagnoses, and other operational factors.

Chart 2: Relationship of size to profitability. The individual market shows a stronger pattern.

Chart 3: Dispersion of admin costs and their relationship to profitability — more variation in the individual market.

The Last Word

Projects are not just about producing deliverables. The way you get to a good-quality deliverable is through people with deep domain expertise working collaboratively. It's crucial to have easy ways to share prompts and projects that can be iterated on and run by different people in an organization, with auditable workflows that can be checked and verified. That's a core philosophy guiding what I build.

I don't think we're in a timeline where LLMs will replace knowledge workers. The reason is simple: we need professionals wielding these tools. An accountant who can serve 10x the clients is in greater demand, not less. The same goes for actuaries, lawyers, management consultants, and data analysts.

Workers and companies that harness AI with the right tooling will have far greater optionality to increase earnings and market share — a classic Jevons' paradox. Organizations that shun these tools, or hobble them with hamstrung implementations, risk falling behind.

Originally posted on LinkedIn · January 13, 2026