AI for Data Analysis
Organizing data and analyzing it are closely connected.
A Phone Full of Unlabeled Files
Imagine a phone that has collected ten thousand photos and several hundred short videos over the years. Each file has a timestamp, but none of them say what is in the picture. Now multiply that by every company, every customer record, every system log, and every signed contract. The amount of data that people and organizations produce keeps rising, and most of it does not arrive in neat rows and columns.
This article answers one question: how does AI help turn that raw material into something that can be found later? The answer starts with understanding what kind of material we are dealing with.
Tracing What a Raw File Holds
Follow one file from the moment it is added to a collection. A photo of a dog on a beach arrives with a timestamp and a file name. A search tool can read those labels, but it cannot read the dog or the beach. The information that a person would want to search for lives inside the content, not in the labels attached to it.
The diagram below traces the general direction of the process. Content with no fixed structure goes in, AI produces a description or extracted fields, and the result is stored so that a later search can reach it. The source does not spell out the internal method each service uses, so the key idea to hold onto is the input and output of each step rather than the mechanism inside it.
Unstructured, Semistructured, and Structured Data
Photos, videos, logs, and PDF documents are examples of unstructured or semistructured data. Their contents are real and useful, but the information is not labeled in a form that a search tool can directly query.
| Data kind | How much organization it already has | Example from this article |
|---|---|---|
| Structured | Arranged in rows and columns with defined fields | Records that fit neatly in a table |
| Semistructured | Some labeled parts, but the content is not fully arranged into fixed fields | Logs and PDF documents, which the source places in this group |
| Unstructured | No fixed fields; meaning sits inside the content itself | Photos and videos, whose pictures are not labeled by the file |
How much organization each kind of data already carries before a search tool can use it
The practical consequence is that a file can be fully stored and still be hard to find. Storage keeps the bytes; organizing creates the labels, descriptions, or fields that search depends on.
Two Jobs for Organizing Visual Content
The source describes two related jobs that AI can perform to organize visual data. Both turn content that is hard to search directly into something that behaves like a searchable index.
Following a Query Through the Index
Finding a beach photo in a personal collection
A collection holds ten thousand photos. Each one was processed by AI so that it has a text description. A user types the phrase dog on a beach. What happens?
- 1Step 1: The phrase enters the search tool
The user's words are the query. The search tool receives text, not a picture.
- 2Step 2: The tool looks through the organized content
Because each photo already has a text description stored with it, the search does not need to re-examine every raw pixel at the moment of the query. The descriptions were created earlier, when the files were organized.
- 3Step 3: Photos that fit the phrase are returned
Images whose stored content fits dog on a beach are presented to the user, who never had to label them by hand.
The user finds the photo without any manual captioning, because the organizing step ran before the search.
Simple and Complex Extraction
Extraction is the process of pulling structured information out of unstructured sources. It moves information from free-form content into structured fields. The source groups extraction tasks by difficulty, and the Intelligent Data Processing industry, abbreviated IDP, is built around this kind of AI-driven extraction.
Simple use cases include automatically extracting information from credit cards, driver's licenses, receipts, tickets, and contact information found in email footers. These documents tend to have recognizable patterns, which makes the extraction task comparatively contained. Complex use cases include extracting data from contracts, reports, and charts. These sources carry more varied layouts and richer meaning, so the extraction task is harder.
Extracting fields from a scanned receipt
A company receives scanned receipts as images. Each receipt contains a store name, a date, items, and a total, but the fields sit in free layout with no fixed positions.
- 1Step 1: Recognize the document type
A receipt belongs to the simple group, so it has a recognizable pattern: the same kinds of information appear on most receipts.
- 2Step 2: Pull out each field
The extraction step moves the store name, date, items, and total from the image into structured fields that a spreadsheet or database can hold.
- 3Step 3: Store the structured result
Once the values are fields, they can be counted, summed, or compared across many receipts.
A pile of unstructured images becomes rows of structured fields that analysis can use.
Where AI Supports Analysis Tasks
Organizing data and analyzing it are closely connected. The source notes that AI is very good with data analysis and that it can write programs that generate data visualizations, identify outliers, and make predictions such as revenue forecasts. Each of these tasks takes data that has already been gathered and asks a question of it.
An outlier is a value that sits far from the rest of a dataset. Spotting outliers helps a team notice errors or unusual events in their records.
A revenue forecast uses past figures to estimate future ones. The source names these as examples of what AI can do and does not describe the specific methods used, so this article keeps to the tasks themselves. Organizing comes first: a team cannot spot an outlier or forecast revenue from records it has not yet made searchable and structured.
Mistakes Beginners Make with Unstructured Data
Common mistake Assuming that a file with a timestamp and a name is searchable by its contents
What it looks likeA photo named IMG_0412 is taken on a beach, but a search for beach finds nothing because the file name says nothing about the scene.
Why it failsLabels such as timestamps and file names do not describe what is inside the picture, so a search tool cannot read the scene from them.
Do this insteadGenerate a text description or extract fields from the content first, then search the stored text.
Common mistake Treating all extraction tasks as equally difficult
What it looks likeExpecting the same accuracy and effort from pulling the total off a receipt as from pulling clauses out of a contract.
Why it failsReceipts have recognizable patterns, while contracts, reports, and charts carry more varied layouts and richer meaning.
Do this insteadClassify each source as simple or complex before planning the extraction work.
Common mistake Thinking that AI describes a file in the same way a person would
Why it failsThe source says a model produces a written account of the content and does not claim that the description covers every detail a person might notice.
Do this insteadTreat generated descriptions as text to be searched and checked, not as a complete record of the original media.
Predicting Which Extraction Is Simpler
What do you think happens?
Which would be the simpler extraction task: pulling the total from a scanned receipt, or pulling the key terms from a signed contract?
Reveal answer
The scanned receipt is the simpler use case; the signed contract is a complex use case.
Receipts are on the source's list of simple use cases because they tend to have recognizable patterns. Contracts are on the list of complex use cases because they carry more varied layouts and richer meaning.
A team has a folder with three kinds of files: photos from a trip, a server log file, and a folder of signed vendor contracts. For each kind, decide whether it is unstructured or semistructured, and name one organizing step AI could take before anyone searches it. Then say which of the three extraction jobs would be simplest and which would be hardest.
Hints
- Photos are named as examples of unstructured data in the source, and logs are named as unstructured or semistructured.
- For organizing, think about text descriptions for images and videos, and about extracting fields from documents.
- Simple use cases are receipts, tickets, credit cards, driver's licenses, and contact information; complex ones are contracts, reports, and charts.
Key Takeaways
- Photos, videos, logs, and PDFs are unstructured or semistructured, so their contents need organizing before a search tool can use them effectively.
- AI can generate text descriptions for images and videos automatically, and those descriptions can be stored and searched like any other text.
- AI can match a text query, such as a scene or an object, to images that fit it, which makes visual content behave like a searchable index.
- Extraction pulls structured fields out of unstructured sources; simple use cases such as receipts and tickets differ from complex ones such as contracts, reports, and charts.
- Once data is organized, AI supports analysis tasks such as visualization, outlier identification, and revenue forecasts.