Release notes

Simplifying life for you, one button at a time.

Improvements

• Autofilling your email address during sign in is even more breezy than before. We’ve made sure to only suggest email addresses, rather than IQL queries, which are subpar suggested email addresses.
• When you’ve gotten to the last field in your suggested tables, we’ve removed the Next Column button so you can press that “Finish setup” button and move on with your life. Column greediness, begone!
• Creating Smart Collections via API just got simpler and easier.

Fixes

• Downloading Views with a union got better and less buggier.

☝️ Check out this cool video about Instant Invoices.

New

Instant Invoices

• In June, we launched Instant Collections, the easiest way to extract data from popular document types. This week, we launched Instant Invoices, a type of Instant Collection that’s specifically trained on invoices. Dive right in to Instant Invoices or learn more first.

Improvements

• IQL Playground: If you modify a query that’s saved as a View, you’ll have to run that query at least once before it can be saved as a View again.
• Things get weird if you try to delete a Collection imported by another Collection. Now, we’ll let you know ahead of time with an alert message to avoid aforementioned weirdness.

Fixes

• More squashing of pesky table extraction bugs.

Don’t you just want to click this autodetected table?

Improvements

Autodetected tables

• Those purple boxes that indicate an autodetected table now have fun dancing dotted lines around them, enticing you to live a little and extract those tables.
• We’ll shrink autodetected tables if they happen to overlap with other fields. They got a well-received lecture on personal space.
• While a table is training on your page, we’ll wait to show suggested tables until the training is complete. While it isn’t a replacement for actual therapy, enjoy the purple wave of calmness and tranquility whilst tables are training.

Fixes

• Squashed a few table extraction and DocQuery bugs. Breathe easy!

Ask DocQuery your burning questions.

Improvements

DocQuery in Impira

Don’t feel like finding the answers yourself? We get it. Now in any Collection, you can type a question and create a field from the answer. This is powered by Impira’s new DocQuery.

Other

• You can now extract up to 100 pages per file instead of 50. We’re keeping it 💯, as the kids say.
• Collapse or expand the indicator next to shared fields to your heart’s content.

• We’ll prevent you from accidentally navigating to another file while you have any unsaved changes, saving you from a momentary bit of heartache.

Fixes

• Fields used to show “No value” as predictions were still processing, unfairly tricking innocent users into thinking something had gone wrong when in fact Impira was just thinking.

Ask DocQuery and you shall recieve.

New

Introducing: DocQuery

We’re super excited to introduce DocQuery, an open source tool that lets you ask questions about documents. Imagine being able to ask a machine questions like, “What’s the invoice number?”

Read more on our blog (check out our Super Nerd Translator) or watch a cool demo video.

Fixes

Table extraction:

• If automatic table extraction doesn’t go perfectly at first, we’ll direct you to set it up yourself. It’s easy to do, we promise.
• When a table’s first row isn’t yet confirmed, clicking Edit field will no longer cause Add column to open. That was weird.
• Having a ton of Impira tabs open would slow down performance, but no longer. Open up as many tabs as you want. And keep going, you multi-tasking maniac!
• Automatic table extraction sometimes picked a silly first row. We made it a bit smarter.

Using the blue box to snag the values in your table’s first row.

Improvements

Table extraction:

• Drop the blue box: The blue box you use to capture the first row of values will more reliably appear when you select table extraction.
• An assortment of tweaks, adjustments, and magic spells to make sure table extraction is running like a well-oiled machine.

Fixes

• Autocompleting IQL suggestions in your custom fields are back to being helpful — and won’t accidentally close the whole panel after you click them.

Hey, look — table prediction previews show up in Extraction view now.

New

• For Pro account holders: Annual subscription billing is going away and you’ll get friendlier, bite-sized monthly billing only.
• You can now get a sneak peek at extracted table data in the prediction preview box (check out the image above).

Improvements

• First things first: You'll get more nudges and reminders to confirm the first row of a table before moving on to any other rows below. This’ll help us help you.

Fixes

• Clicking a shared field in Collection view takes you directly to the correct row and column instead of somewhere not that.
• We’re pretty sure it’s annoying when you can’t see all the pages of a file while you’re making edits, so we fixed that bug.

Fields actively processing after upload

Improvements

• Enjoy smoother table extraction.

Fixes

• Sometimes you couldn’t tell when fields were processing after uploading files. Not anymore. Processing animations for everyone.
• Express yourself: Now you can copy and paste IQL expressions for custom fields again and again and again. (And again.)
• If you need some space, you can have it. Blank spaces in table names won’t cause any errors anymore.
• Overlapping table extraction suggestions would get in each others’ way sometimes and you couldn’t dismiss them. They’ve gotten over themselves since.

The purple wave means good things are coming

Improvements

Autodetected tables

• 🌊 Let’s do the wave: After creating your first table row, enjoy our new scanning animation that signifies we’re predicting more rows below.
• The values we automatically predicted in the first row now reactively change with any edits you might make to those values.

Other

• Creating new fields is much speedier on tables that have lots of fields.

Clicking an autodetected table

New

Autodetected tables

After your files are uploaded into a Collection, we’ll automatically detect any tables found on the first page. Open a file and click the purple box over a table to jump right into labeling your first row. Learn more.

File view is now Extraction view

The ever-useful “file view” is now called “Extraction view.” Don't worry — you can still do all the cool things you’ve always been able to do in file view, but under a new identity. Learn more about Extraction view.

Improvements

• Better, faster, stronger: Extraction for text and table values is getting even speedier

Join our beta

Want to extract information from unstructured documents like contracts, emails, and invoices? We have a beta feature that allows you to ask questions like, “What is the Total Due?” or “What is the loan term?” and Impira will extract the answers out of your documents. If that’s the type of thing that gets you excited and you’d like to join our beta, reach out to us at ankur@impira.com and mention the secret password: Question Beta.

Renaming table column headers

New

Rename table column headers

You can now rename the columns of your extracted table to your heart’s content.

Improvements

• We’ve made file uploading faster.
• We won’t re-add newly split files within a Smart Collection to that Collection unless they fit the Smart Collection’s requirements.

Fixes

• Go ahead and split files with names containing the “.” character. No more bugs!
• All bark and no byte: Now, if you upload an empty file, we’ll let you know with an error message.

Instant Collections

New

Instant Collections

Instant Collections are pre-trained on common documents and forms so you can get going faster by not having to add individual fields and train your Collection. Learn more.

Autorotation

Earlier this year, we introduced page rotation so you could fix any incorrectly oriented pages before you began extracting data, but now we’ll automatically detect those as soon as you upload your files so you can get going even faster. You can always turn this feature off by letting us know at support@impira.com.

Splitting pages with an IQL expression via the API

Want to split the pages of your file into separate files using an IQL expression? Now you can, via the API. For example, you could split your document after the word “Total” or “Page 1” appears on the page. Learn more.

Improvements

• You can now download up to 2,500 rows as a CSV file (up from 1,000 rows).
• You can now use the enumerate() builtin in the IQL playground.
• File view has been optimized to open much quicker for large files.
• If you correct a date or number field to include text, we’ll invite you to change the value type to text rather than just erroring.

Fixes

• The upload box in file view now resizes to accommodate smaller browser windows.
• Fixed processing errors that sometimes occurred while uploading some non-standard PDFs.
• Labels won’t flicker on and off your document as you set up your first row for table extraction.
• Now we prevent labels from occasionally being saved as “no value” during table extraction set up.

Uploading in file view

New

Upload files in file view

After extracting data fields on the first file of a Collection, you can conveniently upload more similar files straight into Collection while you're still in file view.

• P.S. This'll work even if the upload box isn't visible. Start dragging files in and it'll appear.

SOC 2 Type 2 Certification

Data security has always been a priority at Impira, so we've leveled up with our recent SOC 2 Type 2 certification. Learn more.

Improvements

• Hey there, IQL user! There's now a faster way to refer to all the words in a file. The old way was to write: textjoin(" ", File.text.words.word) but that seemed a tad long, so now you can just use: File.full_text.
• Dragging-and-dropping columns to reorder them is now just an overall nicer experience.
• Confirming entire rows: We made it easier to confirm entire rows and the first rows of predicted tables.
• Faster processing on document uploads.
• Picking dates: Setting date values is easier because you can now type dates yourself (in addition to choosing from a calendar).
• Improved handling of TIFF image files.

Fixes

• Joins on custom fields weren't properly rendering all the time but do now.
• We thought it'd be nice to have a gentler message if you accidentally reuse an existing table column name.

Impira's new Help Center

New

Revamped Help Center

Our new Help Center has been rebuilt from the ground up to make it easier and faster for users to find what they need. Designed to be more readable and skimmable than ever, this new Help Center will be your new best friend. (Psst... you're actually in it right now.)

New “Add field” design

Our design team has taken a fresh look at how our users choose field types. This makes it easier for new users to quickly choose between Single text, checkbox, or table values.

Import fields from another Collection

You can now import a set of extracted fields from one Collection and add them to a different one. Read more.

Create and modify Collections through API

Users can now create Collections and import fields from other Collections through the API and SKD. Read more.

Modifying files while uploading via API

Earlier this year, we launched our file modification toolkit (rotating, splitting, and removing pages), but we’ve taken it a step further to allow users to make these modifications while simultaneously uploading them through the API. This feature is currently in open beta. Read more.

Table extraction cues

Did you know that tables only refresh predictions when you confirm an entire row, and that you should always try to confirm the first few rows of each document from top to bottom? We thought that wasn’t super intuitive, so we now have new buttons for you to confirm entire rows from top to bottom. We automatically open the next predicted row with the Confirm row button ready to go.

Views

Use Views to save and manage a frequently used IQL query instead of using a bunch of bookmarks with super long URLs. Read more.

Zapier 1.3

A recent update to our Zapier integration allows users to read files from Views.

Page counter

For free trial users, you now have a “pages remaining” counter in the sidebar to help you keep track of your free page consumption.

Improvements

Rows with varied heights (table extraction)

Our table extraction feature has helped so many users extract data, but not all tables are made the same. We've developed table extraction to tackle rows of variable heights so you can extract whatever kind of table comes your way.

Tip: Confirm a few rows with different heights to train the model.

• Watch a video tutorial or check out the documentation.

Rows with shared fields (table extraction)

You can also create a shared field that's shared by multiple nested rows.

• Watch a video tutorial or check out the documentation here.

Naming files during file splitting

When a document is split, new file names are now “padded with 0's” (i.e., if there are 10 new files, we’ll have file_01, file_02, … file_10 , etc).

Account management

• You can now cancel your plan and change payment details through your billing page (found via the Account icon).
• Our user login page has been redesigned and more ready than ever to launch you into accelerating your workflow.

Other

• In financial docs, a dollar amount in parentheses indicates a negative number, so we now automatically recognize them as such.
• New experience for uploading files:
  • A new set of notifications provide more detail and transparency on your files’ uploading status and location.
  • Faster uploading of large batches of files.
• Machine learning performance improvements for large documents.
• The IQL playground now has a historical query mode.

Fixes

• A bug where deleting multiple Smart Collections only deleted one.
• Fixed PDF scaling and cropping issues.
• Multiple machine learning bug fixes that improve accuracy for text and table extraction.
• Fixed an issue where date filters used the wrong start date.

Table extraction

New

Table extraction for all

Extracting data from tables is now open to all users. Sign in to your account and start extracting table data. Read more.

Modifying files within Impira

Users can now modify their files before beginning data extraction. Rotate pages, remove pages, or split pages of a file into separate new files. Learn how.

Improvements

• Improved and more intuitive navigation for users
• Improvements and bug fixes to table extraction feature
• Improved document view for dense documents, table view for collections with many files, and file processing and downloading
• Improvements with IQL: Join performance and added support for join keys with multiple fields

Fixes

• Bug fixes for CSV and API

New

Impira x Snowflake integration

Our latest integration will allow joint users to use Impira to extract data from unstructured data files in their Snowflake warehouses and stream it back into Snowflake for further analysis. Learn more.

New billing process and pricing structure

Ready to upgrade? Get started with our new billing process by signing into Impira, clicking the Account icon in the top right of your window, then going to the Billing tab. Visit our pricing page to learn more about our new pricing structure

Table extraction: Beta

Our beta program for extracting data from tables is still underway and continually improving. We’d love for you to join our table extraction beta program, take it through the paces, and give us any feedback. Contact feedback@impira.com to access this new feature.

Improvements

• UI: Added tooltips, improved labelling and training experience
• IQL: Significantly reduced training and evaluation time for table extraction
• Created a new compact table view to make more data visibly accessible

Fixes

• The “in progress” spinners that indicate the model is training wasn’t working for some cases
• Users can now confirm “no value” predictions
• Certain valid values weren’t being properly accepted for numeric and date types
• Fixed an issue where values copied from some cells to the clipboard had double quotation marks added

New

Table extraction: Beta

You asked and we’ve answered: Currently in beta, our new table extraction capability allows you to pull data from tables within documents like purchase orders and invoices. We know that tables are rarely straightforward nor do they always follow convention. This feature allows users to extract standard grid-like tables as well as more complex tables that may not conform to a rigid grid format.

We’d love for you to join our table extraction beta program, take it through the paces, and give us any feedback. Contact feedback@impira.com to access this new feature.

Redesign: The “Add field” feature

The new “Add field” feature in our UI makes it easier for users to understand the order of steps needed to add new fields for data extraction. The redesign displays important information that allows the process to be more intuitive. This feature will be rolled out throughout the month of August.

Dynamic bounding box

The bounding boxes (used to select values to extract) will dynamically expand or contract to correspond with what a user is actively typing.

Preview box

Want to see what the field you just extracted looks like on other files? When you’re in file view, check out the lower right corner to see a preview of what Impira’s doing in real time.

Improvements

Text preprocessing improvements

We’ve updated the method of preprocessing incoming files to improve the quality of the extracted text. These changes include using a different parser for digital PDFs, improving the quality of images we use to run optical character recognition (OCR), ignoring embedded PDF text with font issues, and many more. All of these changes contribute to higher quality text extraction for your documents.

Improved performance

We’ve been hard at work improving the performance and speed of the Impira across the board. Some notable examples: Confirming and editing the values for machine learning (ML) predictions is now 10x faster, and file uploading speeds are more than 2x faster than before.

Lots of new entities

In our last release notes, we announced that we have started using Named Entity Recognition (NER) to improve the accuracy of the text extraction models. Since then, we’ve dramatically expanded the set of entities we detect, including addresses (and their constituent parts), email addresses, phone numbers, currency, and more. The ML models can use this semantic information to improve the accuracy of your extractions.

New

Review workflow

Reviewing Impira’s data predictions on your files is easier (and more fun) with the new Review workflow. Fly through the predictions that need review and actively improve Impira’s machine learning models while you’re at it through just a few clicks.

Zapier integration: Beta

The Impira x Zapier integration is now in public beta. Impira users can now use Zapier to connect Impira to thousands of other apps.

Experimental features: IQL Playground

Check out our new Experimental features button, where you can access the IQL Playground. The IQL Playground allows you to make instant IQL queries and view your data in any shape or form.

Named Entity Recognition (NER)

Impira now runs Named Entity Recognition (NER) on each new file, which identifies semantically important groups of text such addresses, currencies, email addresses, and more. Impira’s machine learning models can then use this information to learn what types of text you are trying to extract, leading to predictions with much higher accuracy.

Improvements

• More detailed API error messages.
• Product speed and responsiveness. These speed-ups are most prominent in file upload and processing, creating and editing fields, and working with many collections.

Poll API

Impira supports running complex queries against collections and datasets using Impira Query Language (IQL). You can now use /poll to just receive the changes since the last time you ran a query. This allows you to run a continuous workload, where you receive updates as they happen. Explore our Read API documentation.

Webhooks

Impira supports configuring webhooks to subscribe to changes in Collections and datasets. This is part of the new Collection automations feature, which supports continuously ingesting files and exporting data from Impira. Stay tuned as we’ll be shipping more and more of these features in the months to follow.

New computed fields for confidence

Easily distinguish between files that may need a bit more review and files with fully confident data. With newly introduced computed fields, you can quickly inspect the confidence values and states across all extracted fields for each file in a collection. The three new boolean fields are:

• File.IsPreprocessed tells you whether a file has completed processing (e.g., uploading, analyzing text, producing a thumbnail, etc).
• __system.IsProcessed tells you whether a record in a collection has completed processing across all of its ML fields.
• __system.IsConfident tells you whether all machine learning values for a record are high confidence.

You can read more about these fields in the ML confidence guide.

New text parsing improvements

Impira has significantly upgraded its process of extracting text from documents and images. This includes improved preprocessing for images and documents for optical character recognition (OCR) which will yield more accurate and complete extracted text, and more complete extraction of text embedded in PDFs.

Upload API

You can now upload files directly through Impira’s upload API, which supports both URL-based upload and direct upload over HTTP. We’ve also made several overall enhancements to the API, including simpler paths to access both collections and datasets, a simpler request/response format, and new documentation. Read more about API here.

Upload to Impira via email

You can now email attachments directly into Impira by either forwarding files straight from your inbox or add Impira’s forwarding email addresses to a mailing list so files upload as they arrive. Read more about email integration.

New text extraction

You now have access to a new version of the text extraction algorithm which is both more accurate and learns faster from each new data point. This new algorithm is the default for all users. Learn more about how to use text extraction.

Checkbox extraction

Extract checkboxes, radio buttons, and other binary indicators just as you can for numbers, dates, and other text. Find out more about checkbox extraction.

New onboarding experience

During onboarding, users will see Impira automatically detect important fields for extraction as they upload files, making it easier and faster to get started. Users will also be able to use their own files rather than rely on provided samples.