LedgerBox
Extract Data from Financial Documents with AI
Lessons learned bootstrapping LedgerBox (AI financial doc extraction SaaS) to $1k MRR:
- Focus on solving one problem extremely well
- Get a paying customer before building the full product
- Leverage automation tools
- Prioritize features based on customer feedback, not assumptions
- Don't underestimate the power of personal outreach
Key takeaway: Validate fast, iterate faster.
What's your top bootstrapping lesson?
Looking for a solid tool for API documentation. Would be awesome if it's able to be integrated into the SaaS platform itself. OpenAPI specs (Swagger UI) is good enough for now, but I want to have detailed documentation about how to use the API and integrate it into business workflows.
-
1
Hey Sean,
Saw your post on IH about looking for API documentation tools -
you mentioned OpenAPI/Swagger is "good enough for now" but you
want something that covers workflow integration too.That's exactly what I'm building with Dokly. It's a hosted doc
platform (like Mintlify/GitBook) - you get a clean site at
ledgerbox.dokly.co, proper API reference pages, plus guides
for your business workflows.I'm offering free Pro access (6 months) to a few IH founders
in exchange for feedback. No credit card, no strings.Would take you about 20 mins to get started. Want me to set
it up for you?
Churn is a big issue that no founder likes. When I started LedgerBox, churn was high, but I have managed to decrease it with a few retention methods. One of the most important methods is quality customer service. Being responsive and providing the customer with support goes a long way. Customers are less likely to leave if they have a good, personal experience.
Getting a lot of inquiries about investing and startup accelerators. For the most part they seem predatory vs. helpful. I've been working on other projects at the same time and have some ideas to try out. I've added some new features to LedgerBox to enhance the extraction. Don't want to add new features that aren't needed. YAGNI
A customer reached out about automated monitoring of their email inbox for invoices. This is a common feature in some of the more up market/ large enterprise products. This seems like a worthwhile feature but could blow up into a time consuming venture with a service running periodically on tons of email inboxes. Another way some products achieve this is by creating a dedicated inbox for each customer where they can forward invoice/receipt attachments. This would allow for a more event driven architecture from my end which is preferable. Still contemplating the idea.
I added review status and the ability to update transactions extracted from documents via the application UI. Still working on some updates to the UI for a more interactive process. I contacted Intuit about fixing their Quickbooks Oauth client that hadn't had a release since 2021. Surprisingly, this must have lit a fire under their ass as they made a release shortly after.
Working on a Quickbooks integration for AP automation. Accounts payable is often a pain for customers to manage. In theory this feature automates a common desk job of gathering PDF invoices and manually entering the details into the AP management system (e.g. Quickbooks, Netsuite, etc.) The integration can be taken a step further with a dedicated email inbox for forwarding invoices. I see this is common among other IDP providers.
One common theme that has been coming up recently is batch processing, or as the users call it, convert a bunch of files at once. To achieve this I implemented a queue system using event driven architecture and webhooks back to my app. It wasn't that difficult, but it made me revisit the maintainability and modularity of my product. I want to ship fast, but I don't want to sacrifice code quality. As LedgerBox gets bigger and bigger, it's important for the code to be well-documented and architected with intent.
-
1
Did you implement the queue system yourself, or use something like Upstash? My old manager had a saying along the lines of "each new feature is tech debt" - curious how you weighed potential options.
-
1
I implemented the queue system myself, using cloud storage, triggers, webhooks and my application api to maintain state in my database. it sounds more difficult than it was. i can go into more specifics privately.
-
Hit $250 MRR. Finding the right LLM for certain document processing tasks has been difficult. It's hard to find the right balance for precision, cost, speed, and availability. Rolling my own LLM might make sense in the future to cut costs, but right now I'm focused on finding PMF. I can eat the cost for now, but I'm patiently waiting for improved models that run more efficiently.
-
1
Nice work and excellent progress. What was the LLM that you ended up using?
Working on a Quickbooks integration after a customer inquired about it. Seems fairly straight forward, Intuit has solid developer docs. Many small businesses/firms use QBO to manage their accounting so it should be worth the effort. Made an update to the landing page to help with organic keyword ranking. SEO is the best offense against a venture backed competitor, so and steady wins the race.
Comment