We're delighted to announce Heron Data's first funding round, a $1.2M pre-seed round completed in September. We're excited to be backed by a slate of great investors, including Y Combinator, BoxGroup (Plaid, Stripe), Flex Capital (Chime, Flexport), Musha Ventures (Streak, Lever) and a range of angels, including Fintech unicorn founders Matt Robinson (GoCardless), Shivaas Gulati (Remitly) and Jonathan Levin (Chainalysis). We have raised this round to help us move faster towards our immediate goal, which is solving the problem of understanding and enriching bank transaction data.
You may sometimes look at your bank statement and ask yourself what a transaction description like the below could possibly mean:
SP * PINEAPPLE CO 05/06 PURCHASE 098314763
As infrastructure allowing third-parties to access bank transaction data is growing around the world, more and more companies have to know the answer to that question, too. A new class of start-ups is building anything from budgeting apps to tax preparation software on this data. Banks that do not enrich and act on this data risk being left behind by the start-ups that do.
In building an earlier company that was leveraging such data, we realised that the problem of making sense of this data by labelling and categorising it is unsolved. Most companies rely on in-house solutions that are constrained by small data lakes, mapping existing category labels to the custom categories that their business logic requires.
We are solving this problem today for multiple paying customers, and you can get a sense of our approach here. We're offering a categorisation service with bespoke labels, in which our customers benefit from a data lake much larger than their own. The impact of a full suite of world-class products on transaction enrichment are vast: Think of the hours spent by book keepers reconciling bank transactions to accounting entries; or the resources needed to reply to consumers disputing legitimate charges on their bank statement that they cannot identify.
As for the above, it turns out that the transaction comes from the Pineapple Collaborative, a company with the slogan “pantry staples, made by women”, making Olive Oil locally in California. Depending on your use case, this may be tagged as “Groceries” or "Supplies", but it could also be "Personal expense", "Independent retailer", or "Low-carbon retail".
Solving the problem of labelling such an object using data science models (as opposed to a tagging farm) poses interesting technical challenges. This is because our object of analysis is unique: A transaction description string, an amount and a timestamp. This means that:
- We can use elements of the advances in NLP (like GPT-3), but we are not quite analysing natural language, so these do not just work off-the-shelf.
- We operate in an environment without a ground-truth (e.g., multiple underwriters at a Fintech lender may classify the same transaction differently), so we need to build technology that weighs the judgements of different annotators to come to a consensus label for a given transaction
- We can calculate features of the data that are powerful in predicting category labels, for example the recurrence of a transaction
More broadly, we believe that the "Finance edition" of software-is-eating-the-world is barely underway. By the time it is finished, every financial institution will be a software company. We will enable that change by building technology that makes it easier to create intelligent, personalised digital financial products with minimal overhead. With advances in NLP and machine learning, a future where every consumer and SME can profit from the kind of bespoke financial advice and services currently reserved for the rich and large corporations is within reach. We want to power that future.
Talk to Us
If raw transaction data is causing your business pain, email us or book a demo on our website - we'd love to work with you.
If you want to join us, have a look at our Careers page - even if no position is open, we always enjoy talking to people interested in what we do.
If you want to be kept up to date on our progress, drop us a line at firstname.lastname@example.org, and we'll send you semi-regular updates.
Onwards and upwards,
Dom, Jamie, and Johannes