Ntropy Changelog

Hi all,

Lots going on. Series A announcement, partnership with Q2, successful Money2020 where we won a ton of chess games by our booth, met new customers and partners and future team mates (you know who you are 😍).  

While our core API has been improving per usual with yours truly, Ilia and many ML engineers focused on it, we have also been quietly working on something new.

Ntropy Income Check is finally live on Product Hunt! Please upvote, share and ping for access! Super excited with the urgency on the market and cannot wait to serve you all.

In the coming months we will be launching the next generation of tools on top of Ntropy Core in order to make sure it is even easier to ship financial products using transaction intelligence.

Meanwhile last, but not least, Ntropy Core updates below:

• Added intermediaries field, which contains the payment processor or intermediary entities in a transaction. Check it out in the docs!
• 67% error reduction for merchant recognition in Portuguese and Spanish transactions
• 10% error reduction for transaction_type (business / consumer detection)
• Better and more stable merchant names for returned websites
• New SDK version 4.11.1. Lots of performance improvements and more. Check it out!

Boom and it's the middle of August. No slow down on these shores 🌊

Here is what we have shipped this sprint:

• SDK: EnrichedTransaction now also contains input fields
• SDK allows Transaction iterables as input for add_transactions() (not only lists)
• Extended supported range of MCC tags (including <1000 codes)
• Accuracy improvements in MCC tags
• MCC tags resolution no longer incurs a latency penalty
• Recurrence group of a transaction now contains own transaction_id
• Improved recurrence results for asynchronous enrichment
• Improvements in processing latency for merchants that have previously been resolved
• Improvements in merchant resolution for long-tail cases
• Recurrence field extended to resolve “repeating” transactions. It can now take of of the following values:

1. “one-off” (one time, no connection to other transactions)
2. “repeating” (regular morning coffee, friday groceries, etc.)
3. “recurring” (utilities, mortgage, etc.)
4. “subscription” (spotify, netflix, etc.)

Last but not least, Income Check via Bank data powered by Ntropy is in beta. This means you can now identify income, including gig, non-traditional sources, stipends, government payouts via transaction data only. In case you have high drop-offs with Payroll API-s or are covering a demographic that is not compensated via payroll, please do reach out to be a part of this beta.

‍We can drastically cut time to money, enabling income insights for everyone without extra authentication.

Here's what we have been up to in July!

• Lots of improvements to our documentation portal, including quickstart guides and more in-depth examples
• 5% error rate reduction in merchant recognition across all languages and regions
• Significant improvements in website and location resolution
• Released ntropy-sdk 4.7.0
• Updated account holder methods
• Increased control for retries, handling unexpected errors from the server
• New: simplified models API
• New: asynchronous enrichment method: add_transactions_async
• 12% error rate reduction for merchant identification in English transactions

Lots more coming your way soon, including our first vertical product offering on top of Ntropy API 👀 🔥

Happy Tuesday ! Here’s what team Ntropy has been up to the last two weeks:

• New field: merchant_id - a unique string that identifies each merchant
• New field: recurrence_group - a list of transaction_ids that are recurrent together
• 22.6% error reduction in merchant recognition for English transactions
• Scalability and throughput upgrades for both single transactions and batches. Ask us if you need higher limits on your API key!
• Reliable merchant confidence scores

We have a bunch of open roles. Software engineers, technical account executives, marketing! If you are someone or if you know someone, please do reach out ! We are excited to talk !

Everyone has predicted the future and settled in the present. Most of us are downloading Excel and figuring out financial planning.

Meanwhile at Ntropy, business as usual. After all we should thrive in chaos, by design. 😜

If you are focusing on building or maintaining in-house categorization for lending, merchant management, automating tax or accounting, here is why you should not: https://ntropy.com/post/the-false-promise-of-general-transaction-categorization-and-the-inadequacy-of-in-house-models

Fintech Nexus was busy and exciting, while the product team has been head down shipping. The latest updates are below:

• 43% error reduction on labels for consumer transactions
• 9% error reduction on labels for business
• 28% error reduction for merchant extraction in English transactions
• Lots of fixes and improvements to our developer portal: https://developers.ntropy.com/
• Offsite in Portugal wrapped: https://twitter.com/ntropy_dev/status/1529145822990020610?s=20&t=EWgJzahSftI6Aq_6K6o5EA

It's doomsday out there in terms of markets.

We love when our customers fund us and we are working days and nights to deliver the best transaction API experience possible.

If you are thinking of starting to integrate payments data into your workflows, do reach out.

Here's what's been cooking:

• New version of our Python SDK is out: https://github.com/ntropy-network/ntropy-sdk
• Ntropy Custom Models are now more robust to noise in the labels. Still have one or two spots for beta customers
• Significant performance improvements on specific labels on consumer transactions, particularly for “income” and “fees”
• New labels in the base hierarchy: “child support” (only for consumer transactions), “prenote” for authorizations
• Confidence scores are now more accurate and available for each API key
• 11.1% error rate reduction for English transactions
• 30.4% error reduction on Spanish transactions
• 13.8% error reduction on Portuguese transactions
• For merchants without logos or transactions without merchants, we now return beautiful category-specific icons. So you do not have to figure it out yourself
• General improvements to our developer docs: https://developers.ntropy.com
• Ntropy is sponsoring PyData London this year. 17-19th June (https://pydata.org/london2022/). We will give a 45-min deepdive on how things work under the hood. Join us!

Last, but not least, if you know amazing marketing and product people, send our way!

Seems like ages in terms of how much we have shipped, but it has only been two weeks!

First and foremost, if you ever debated whether to build something more specialized in-house or to use Ntropy, no more debating needed.

CUSTOM MODELS are here! Whether you're building in-house fraud models, underwriting, adapting to a specific vertical, Custom Models are made for you (literally), and a game changer.

Below are some more details and general updates from the team 🔥:

1. Custom models now available for consumer transactions
2. More transactions can now be used to train a custom model (50k, up from 8k)
3. Recurrence now uses the history of past transactions as additional context. This has a latency cost and can be enabled upon request.
4. 23% error reduction for merchant detection on English transactions
5. 30% error reduction for merchant detection on German transactions
6. 31% error reduction for merchant detection on French transactions
7. Significant improvement in accuracy for business / personal spend detection (transaction_type field)
8. Blog on customized transaction modeling (AKA Custom Models) by Jonny Kernes
9. Blog on how we are using ML pipelines internally by Robin Kahlow
10. Shiny new docs and dev portal at https://developers.ntropy.com/

Last but not least, for custom models private beta, drop us a line! IF you are in the real-time ACH space or have a use-case for vertical lending, we are already seeing lots of demand there.

We are moving and shipping fast. In addition to our regular improvements, we are excited to now return improved location information. And keep an eye out for more content next week...happy updating!

Product Announcements 📣

• New SDK version with Custom Models (fine-tuning) available; more information on this coming soon...

TXN Updates 📈

• 12 new categories in our hierarchy
• Improved coverage for loans, mortgages, rents
• Improved location data: ability to match transactions to merchant address, where store_ID is present

Improvements 🚀

• 8.1% error reduction for merchant extraction in english transactions
• 13% error reduction for consumer transaction categorization