Maximize Efficiency and Cut Costs with AI-Powered Vendor Bill Management
Discover how AI technology can revolutionize vendor bill management in real estate, offering solutions that simplify processes, reduce errors, and cut costs. Learn about the benefits of automation and how integrating AI tools can enhance efficiency for property managers and investors.
One of the most common complaints from investors, large and small, is the sheer amount of work required to manage and pay local vendors. As Picket grew, we began to experience these pain points firsthand. Almost every aspect of bill management is time-consuming, error-prone, and a significant cost center for the business. Some of the most labor-intensive and manual workflows include:
- Receive, open, and document vendor scopes, invoices, and receipts from emails and physical mail.
- Verify invoiced work was is complete and matches the previously agreed-upon terms.
- Translate and map vendor terms to your companies internal language.
- Pay vendors for approved invoices and tracking payment status.
- Reconcile successful payments with the approved invoices.
When you break down the tasks involved, you’ll see that none of them are particularly difficult. This is why companies often assign this work to low-cost workers. However, the large volume of tasks still adds up to a significant expense for most companies. So, how can you improve the process? Automating bill management with AI and tools like ChatGPT turns out to be a smart and effective solution.
Create Your Free Picket Account
Used properly, today's AI models can automate the vast majority of bill processing & triage - Head of engineering, Picket AI
As we gained more insights, our data science and AI teams became increasingly confident that most bill management tasks could be automated with the right technology. This idea is now being proven by Picket’s SmartBill AI, a tool we first created for our own use and are now offering to everyone through the Picket Web Services API. We also have a no-code version that easily integrates with emails and spreadsheets.
SmartBill: Revolutionizing Bill Management
The first version of SmartBill, released in September 2024, focuses on the initial stages of bill management, including data ingestion, documentation, translation, and operational validation. Our goals for SmartBill are to:
- Reduce labor costs by 75% to 100%.
- Collect, structure, and make usable all bill and vendor data from emails, files, and photos—the vast majority of which goes unprocessed today.
- Achieve an AI data entry error rate of less than 1% and an overall data entry error rate of less than 0.1%.
Data Ingestion & Documentation
We knew that for AI to process data successfully, it needed to be equipped with as much relevant context as possible. For example:
- If a bill is sent via email, the email content must be included as part of the ingested data set.
- If an email thread contains multiple files, those files need to be considered collectively rather than individually.
- If a bill references a specific work or purchase order, that order's data becomes an essential part of the context needed to process the bill.
We found that the simplest solution is to set up a dedicated bills email inbox, where each new email triggers a call to SmartBill’s API. At Picket, we have our vendors send invoices directly to this address, but the workflow also supports forwarding emails from other inboxes. Users can also add additional context in the body of forwarded emails, which is particularly useful in less typical circumstances.
By creating a single channel for inbound bill data, we greatly simplify the surface area of the bill management problem. SmartBill AI then converts this inbound data into structured data objects that are delivered either via API or as a daily emailed CSV, effectively replacing manual data entry with a single API or service. However we are still just scratching the surface. Armed with the appropriate amount of context, AI models can do much more than just data entry—they can also interpret, analyze, and translate the data.
Translation & Validation
To correctly interpret new sets of emailed data, we needed to empower our models with enough context. For the initial use case of interpreting vendor bills in real estate, we solved the vast majority of cases with three additional layers of context:
Optional Data Structure & Schema: SmartBill allows users to include a desired data structure and schema. For example, accurate accounting translations can be achieved by including your company’s chart of accounts.
Operational Events: Users can include operational events, such as work orders. This can be done by setting up a daily export/email to your SmartBill invoice account or via API (see our developer documentation for details).
Property List: Users can include a list of properties with any unique internal identifiers. While we have our own internal identifiers, our goal with SmartBill is to ensure it works seamlessly within the systems you already have in place.
While SmartBill performs well on data ingestion and documentation without these additional datasets, including them is crucial for effective AI-driven translation, analysis, and validation.
The Results
So, how does AI perform when tasked with the initial stages of bill management? Much better than expected and far superior to a fully manual approach.
Manual data entry has an expected error rate of approximately 3%, meaning 3 out of every 100 bills will include some user-caused error. Using AI to automate data entry drops the error rate to about 1%.
We are still collecting data on how well AI performs during the interpretation, analysis, and translation stages. However, early user feedback is promising:
"As an accountant, the new auto-suggested classification saves me a ton of time. It still makes some mistakes on harder judgments like expense vs. capitalization decisions, but in most cases, it gets it right."
“One really useful thing SmartBill does is automatically map to the correct property ID every single time. Even when the vendor doesn’t include a property address or property ID, SmartBill uses the work order ID to figure out the correct referenced property and adds it to the CSV export.”
Smart bill has already saved Picket a ton of headaches internally and expect others will continue to have similar experiences. We expect future versions of Smart Bill can even be used outside of real estate operations to help manage and automate general corporate expense management. It turns out managing bills is universally painful no matter company or industry and AI offers a game changing path to fix it.
In Summary
Yes, you can use AI to dramatically improve efficiency and reduce the costs associated with vendor and bill management. One option is to develop directly using the best LLM APIs like OpenAI. However, after going through this exercise at Picket, we realized there is a lot of work required to deliver an effective solution. For example, when you use Picket’s SmartBill API, you’re leveraging a combination of more than five different models. Additionally, keeping up with model updates across the rapidly evolving ecosystem of service providers is a full-time job.
Using a loosely coupled API or service like Picket Web Services (think of it as an API for a business function) is less resource-intensive to maintain over time. Picket handles keeping up with the latest and greatest AI models, and when a new model is worth using, we upgrade to those better models. But critically, as a SmartBill customer, these changes have no impact other than to improve the quality of output from the API or low-code CSV exports.
More from the Picket AI Team
Picket Team
Picket Team
Picket Team
The Open Real Estate Company
Picket is on a mission to make real estate open, efficient, and fun for all
See How It's Going
