The world’s going crazy over ChatGPT. We haven’t seen this level of excitement since Apple introduced the iPhone 15 years ago. With ChatGPT, however, there is a mix of both jubilation (from people that hate to write) and paranoia (from those who live to write and write to live). Many fear that machine-powered AI will eliminate the need for humans in numerous professions.
- How will this affect journalism – if news stories (like sports results) can be auto-generated?
- How will this affect education – if students can use AI to write answers to essay questions?
- How will this affect software development – if AI can write code?
ChatGPT and Finance
One of the job functions that there is far less buzz about is corporate finance professionals. How will generative AI technologies like ChatGPT impact accountants, finance, and treasury professions? Much of the content in annual SEC filings, quarterly earnings announcements, and internal policy and control documents is fairly standard. It would be relatively easy to have the boilerplate, common language in many of these documents generated automatically.
Another interesting area to consider is customer communications – most of which are generated by the accounts receivable team as part of the billing and collections process.
Testing ChatGPT for Accounts Receivable
Dunning Message and Customer Communications
The labs team at Ordway tested ChatGPT’s ability to generate dunning messages and other customer communications for five common use case scenarios. We found that the generative AI platform did a great job of quickly crafting messages. ChatGPT not only authored the text accurately, but it also used the correct tone of voice. Perhaps, most impressive is that the AI platform also inserted value-added content without being requested. On each message, it appended a few sentences of verbiage suggesting the customer visit the website to make a payment or contact the customer service organization with questions.
Scenario #1
Late Invoice, Payment Reminder
Getting timely payments is critical to the cash flow of every organization. A good example of how ChatGPT for accounts receivable can reduce workloads is automating communications with past-due accounts. We asked it to write an email to a customer informing them that we have not received a payment for Invoice 1234 for $10,000 which was due on January 1st. Below is the output it produced.
Scenario #2
Account 90 Days Past Due, Pending Disconnect
Second, we asked ChatGPT to write an email to a customer informing them that their account is now 90 days past due and at risk of suspension due to an outstanding balance of $50,000. Below is the output it produced.
Scenario #3
Subscription Renewal Reminder
Auto-renewals are one of the top causes of billing disputes at subscription companies. We asked ChatGPT to write an email from to a customer notifying them that their subscription will automatically renew on June 1st and they will be billed $500 for the next 12 months of service. Below is the output it produced.
Scenario #4
Card Expired Notification
Failed card transactions is one of the leading cause of payment failures. Collections teams spend. lot of time chasing customers whose card on file is invalid to update their payment method. We asked ChatGPT to write an email to a customer notifying them that the credit card on file with the last four digits 1234 has expired and that their most recent payment transaction of $150 failed. Below is the output it produced.
Scenario #5
Billing Error, Refund in Process
Another example of how to use ChatGPT for accounts receivable is refunds. We asked it to write an email to customer letting them know that they will be receiving a refund of $25,000 from their most recent payment of $50,000 due to a billing system error. Below is the output it produced.
A/B Testing and Experimentation
Most of the examples above are very common and automatically generated by billing systems based upon templates. Many collections teams use the default communications template supplied by the accounts receivable software vendor. Generative AI could create hundreds of different permutations of each message type to identify which one yields the highest response rate amongst the customer base.
Accounts Receivable Exception Scenarios
Once the optimal messages are identified from the A/B testing and experimentation, there will still be a number of exception scenarios that are not worth developing standard templates. Examples might include:
- Responding to disputes and inquiries
- Suspension and reactivation notices
- Price change notifications
- Explanations of pro-rated charges
