Using AI to Find New Data

At Puritas, we have been using computers for a long time. How long? Let’s just say for decades.

Over the years, we have accepted that extracting useful information from a pile of data has always required some structure like a database, a spreadsheet, or a CSV file.

Data needs to be in rows and columns so it can be worked with. Much of the day-to-day content we work with doesn’t fit that mold; all our emails, notes, and letters are unstructured, plain-English data that leaves us with few options for analysis. You could read every item by hand or manually move specific things into a spreadsheet, but those approaches are not efficient.

AI’s like Chat and Claude have changed the game when it comes to plain English data. They bring context and understanding to unstructured text. You can hand AI a folder full of emails and ask it questions in plain English: “What were the most common complaints this month?” “Which clients haven’t I responded to yet?” The AI reads through everything, understands what it’s looking at, and gives you an answer.

What makes your email particularly well-suited to this kind of analysis is that it’s actually quasi-structured. Every email has a date, a sender, a subject line, and a body. That easily beats a litter of handwritten notes. When you export a batch of emails, you’re handing the AI material it can work with. The headers tell it who said what and when, and the bodies are exactly the kind of human-generated language LLMs were trained on.

Try it yourself!

Step 1: Collect the Emails You Want

Before you export anything, collect the emails that matter. Most email programs let you search by sender, recipient, subject, date range, or keywords. Once you’ve got a set of results, move them into a dedicated folder.

A few examples of useful filters:

• All messages from the website’s contact form
• Every email I sent last month
• Every email from or mentioning a particular person
• Every message containing the word “invoice” or “payment”

Step 2: Export the Emails to a Folder

The export step depends on which email program you use.

Outlook (desktop application). Outlook has the most straightforward export. Select the folder you want to export, then go to File → Open & Export → Import/Export → Export to a file. Choose the Comma-Separated Values format, select the folder, and save. The result is a single spreadsheet file with one row per email.

Outlook on the web. The web version of Outlook, unfortunately, does not have a built-in export feature. You have two choices. First, you can install the Outlook desktop application — it’s included in your Microsoft 365 subscription — and let it sync your mail, then export from there. Second, you can open each message individually and use the Download option, which saves each email as a separate .eml file. The second approach is tedious for hundreds of messages but workable for a few dozen.

Gmail. Gmail’s approach is different, and it’s a bit much compared to Outlook, so we won’t go into the specific steps.

In all cases, the goal is the same: end up with either a single file containing all the messages or a folder containing one file per message. If you have a folder, compress it into a .zip file — that gives you a single thing to upload.

Step 3: Upload to an AI and Ask Your Question

Open Claude or another AI assistant and upload the file. Then write a prompt that tells the AI what these messages are and what you want to learn from them.

As usual, the prompt matters. Vague questions get vague answers. Specific questions get useful ones. Here are three real examples worth trying.

Example 1 — Contact form analysis. Export every message that came through your website’s contact form in the past month. Upload the exported file. Then ask:

“These are messages submitted through my website’s contact form in April. Please tell me (1) why people contacted us — group by reason, (2) the most frequently asked questions, and (3) suggested short-answer responses I could publish on my website to share the answer with others.”

What you get back is essentially a free analysis of your client/customer support load, categorized and summarized, with draft responses.

Example 2 — Build an FAQ from your sent messages. Export every email you sent during the past month. Upload it and ask:

“These are emails I sent in April. Identify the questions I answered most often and draft a frequently-asked-questions document based on my actual responses, in my own voice.”

The AI identifies recurring themes and uses your own writing as the model for tone. The result is an FAQ that sounds like you, rather than a boilerplate help desk document.

Example 3 — Build response templates from your inbox. Export everything you received during the past month, along with your responses if possible. Then ask:

“These are messages I received last month. Identify the most common types of requests and, for each one, draft a response template I can reuse. Base the templates on my actual reply style if you can see it in the thread.”

Now you’ve got templates for the five or six situations that account for most of your inbound email built from your own responses, ready to drop into your email client.

Don’t Forget Privacy

If your email contains client-confidential information, think carefully before uploading it. Especially to free AI accounts that often use conversations for training. Always use a paid account when sharing data. For a small office reviewing its own contact-form messages, this generally isn’t an issue.

New Data Means New Information

Since the beginning of time, getting useful information from your data required organizing it first. Spreadsheets, databases, and structured forms were the price of admission. AI has lowered that price dramatically. The pile of unread email, the folder of old client letters, the years of meeting notes all of it is now, suddenly, searchable. Not just for keywords, but for meaning and valuable information.

The best part is you don’t have to buy a tool or learn to code. You just have to know how to export your data and ask a good question.