Webinar Replay: What's Your AI Expense Reporting Strategy?

Webinar Transcript:

00:00:05.000 --> 00:00:21.000
Good afternoon, everyone—or good morning, depending on where you are in the country. My name is Chris Harley with DATABASICS. Over the next 25 minutes or so, we’re going to

00:00:21.000 --> 00:00:37.000
go through a topic that’s very relevant to the conversations we’re seeing in the expense reporting world: AI and how it relates to what organizations are trying to accomplish from an expense management perspective. There are a lot of questions about AI.

00:00:37.000 --> 00:00:57.000
There are also questions about how AI operates and many areas to consider when evaluating it. Because we constantly receive these questions from customers and prospects, we wanted to give those of you who aren’t familiar with AI an introduction, then take a look at how the application operates.

00:00:57.000 --> 00:01:12.000
We’ll cover what AI is capable of doing, what to look for when developing a deployment strategy, and considerations such as cost,

00:01:12.000 --> 00:01:23.000
privacy, and more. We’ll also take a quick look at what DATABASICS provides within the application, including a live demo of the solution.

00:01:23.000 --> 00:01:40.000
As mentioned, my name is Chris Harley. I’m the Vice President of Sales at DATABASICS. At the end of the summer, I’ll reach 26 years with the organization, so I’ve certainly seen a lot change—especially the technology used for expense reporting. When I joined the organization in 2000,

00:01:40.000 --> 00:01:56.000
the web was still relatively new. Organizations wanted to automate workflows and gain visibility, and even the ability to capture receipts was still a couple of years away.

00:01:56.000 --> 00:02:12.000
Today, as we consider how technology has improved and streamlined expense reporting, AI is simply another step in that evolution. I’m sure everyone on this call thinks the ultimate goal is to eliminate expense reporting.

00:02:12.000 --> 00:02:29.000
While it remains a requirement, we’re trying to make it as simple as possible for the end user while ensuring that our core customer—typically the finance and accounting team—has all the necessary controls and compliance capabilities.

00:02:29.000 --> 00:02:45.000
We also want to ensure that you can automate and streamline day-to-day operations involving accounting, finance, revenue recognition, and more. For those who aren’t familiar with DATABASICS, we’re headquartered in the greater Washington, D.C., area. In addition to the expense reporting platform we’re showing today,

00:02:45.000 --> 00:03:00.000
we also offer a timesheet solution. The applications can be bundled together or provided independently, as with the expense reporting application I’m showing today. As I mentioned, I’ve been here for about 26 years, and the company is roughly 30 years old.

00:03:00.000 --> 00:03:17.000
Our customer base spans many industries and verticals. We do extensive work with project-based organizations—essentially, anyone allocating expenses to an internal or external activity.

00:03:17.000 --> 00:03:29.000
Examples include professional services companies, nonprofits, and oil and gas organizations—really, anyone allocating expenses to a project or activity.

00:03:29.000 --> 00:03:46.000
AI is everywhere—I don’t need to tell you that. What people are trying to understand, and what prompted today’s conversation, is what AI can and can’t do.

00:03:46.000 --> 00:03:55.000
One of the most common questions prospects ask us is a very general,

00:03:55.000 --> 00:04:14.000
broad one: “Does your application have AI?” Our team typically responds, “Yes, we have AI. What are you looking for it to do?” That often stumps people because they don’t know what AI can do or what they should expect from it. That’s one of the things we’re addressing today.

00:04:14.000 --> 00:04:29.000
When you ask a vendor that question, it’s important that the vendor understands what you’re trying to accomplish. When most people think of AI, they think, “It does my job for me.” That’s the prevailing narrative. As we all know,

00:04:29.000 --> 00:04:37.000
we don’t think AI is going to eliminate our jobs. The real question is: How can we use it to

00:04:37.000 --> 00:05:00.000
do our jobs better? There are also important questions to address when deploying AI. A poll comparing expectations with reality found that most organizations are looking to implement AI in some capacity,

00:05:00.000 --> 00:05:18.000
but actual usage remains limited, even among DATABASICS customers. We made our first generally available AI solution available around this time last year.

00:05:18.000 --> 00:05:35.000
Today, a good number of our customers have access to the AI tool, but a major part of our service is still helping them understand what it can and can’t do so they can get the most value from it.

00:05:35.000 --> 00:05:51.000
There is still a gap between expectations and adoption. Other vendors make many claims about AI handling approvals, receipt capture, and spend management—even eliminating positions.

00:05:51.000 --> 00:05:58.000
Everyone talks about how great AI is, but we want organizations to consider

00:05:58.000 --> 00:06:14.000
the key questions surrounding it. In no particular order, the first is cost. AI isn’t free.

00:06:14.000 --> 00:06:31.000
For those old enough to remember, I compare AI usage to cell phone minutes—or long-distance minutes, if you want to go even further back. Depending on how it’s used, AI usage is measured in

00:06:31.000 --> 00:06:51.000
tokens, and the vendors you work with incur costs for those tokens. Typically, some portion of that cost is passed along to the customer. One challenge with AI—returning to my cell phone minutes analogy, as someone who had a teenage daughter—is that

00:06:51.000 --> 00:07:07.000
you don’t really know how many minutes will be used until usage begins. Once it does, you can identify patterns and estimate what month-to-month usage will look like. For different organizations, depending on what they ask AI to do,

00:07:07.000 --> 00:07:24.000
costs may be very high or relatively low. We’ll discuss some of the things we’re doing to make those costs more manageable. Most importantly, AI involves a third-party cost that you should understand when calculating overall ROI.

00:07:24.000 --> 00:07:40.000
Another area to consider is privacy. Because this is a newer technology, where AI operates and what data it can access are significant concerns—whether the information is external

00:07:40.000 --> 00:07:57.000
or internal. AI can capture data from receipts and hotel bills with impressive speed and detail. That level of detail is an important strategic consideration because organizations are asking,

00:07:57.000 --> 00:08:15.000
especially in expense reporting, “How much information can I access?” That can become information overload, or expose information you probably don’t need. Another consideration is human oversight. When we start giving machines control

00:08:15.000 --> 00:08:35.000
over information review, we still need oversight to ensure that audit processes receive the right information at the right time. The last consideration is deployment. This returns to the strategic question: What are you trying to have AI do for you,

00:08:35.000 --> 00:08:46.000
and how should it be deployed? During the next 20 minutes or so, I’ll discuss some of the approaches organizations are taking.

00:08:46.000 --> 00:09:07.000
Now, what’s the difference between AI and automation? When an organization asks, “Does your application have AI?” we typically respond by asking whether it has specific challenges or use cases in mind. If the people involved aren’t familiar with AI, they may not have specific use cases yet.

00:09:07.000 --> 00:09:23.000
We start by explaining that some capabilities may not actually be AI; they may simply be automation. One major example is receipt OCR. For those who have automated expense reporting with a third-party tool like ours,

00:09:23.000 --> 00:09:38.000
OCR allows an application to read data from receipts. It has been around for years, so it isn’t really AI. AI allows us to take that data and

00:09:38.000 --> 00:09:58.000
find information based on what it captured from the receipts. Rules-based automation includes policies, receipt matching, and approval thresholds. Applications have been able to enforce policies and automate these functions for a long time.

00:09:58.000 --> 00:10:14.000
With machine learning, however, we can teach the application that whenever it sees a parking receipt from a particular airport,

00:10:14.000 --> 00:10:20.000
the user doesn’t need to identify it as a parking receipt. The application can

00:10:20.000 --> 00:10:33.000
use what it has learned to automatically create an expense type based on the receipt. Machine learning has been part of applications like ours for years, allowing the application to automatically create an expense report

00:10:33.000 --> 00:10:47.000
as long as it has seen similar data before. Generative AI helps organizations by providing a

00:10:47.000 --> 00:10:53.000
complete, interactive explanation. So the ability for the solution to

00:10:53.000 --> 00:11:08.000
It can look at a particular challenge or problem, ask an end user what happened, and support interaction between the user and the AI solution without bringing human innovation and human intervention.

00:11:08.000 --> 00:11:27.000
Another capability is natural language queries, which I’ll discuss in relation to approval processes. We’ve all seen these queries used in chat windows. One of the things we’re doing with AI is deploying it within our support organization.

00:11:27.000 --> 00:11:41.000
There are several ways to reach our support group. Instead of getting someone on the phone or submitting a support ticket, a customer can ask a question of an application containing

00:11:41.000 --> 00:12:01.000
all our internal documentation and quickly receive an answer without waiting to speak to our support team. We’re always happy to talk to people, but providing a fast answer to a simple question makes the entire process much easier.

00:12:01.000 --> 00:12:17.000
AI also supports much more granular detail, as I’ll show during the demo. By capturing OCR data, we can run very specific queries.

00:12:17.000 --> 00:12:33.000
AI can also complete queries more quickly than someone manually reviewing an expense report. For example, if you’re looking at a hotel folio and trying to decipher individual line items,

00:12:33.000 --> 00:12:47.000
the task can become very tedious. An AI approval or audit process can capture that data quickly without requiring expensive manual review.

00:12:47.000 --> 00:13:03.000
How does this reduce time and effort within an organization? The major advantage is AI’s ability to quickly query

00:13:03.000 --> 00:13:21.000
large amounts of data and evaluate individual details against your policies, internal controls, and compliance requirements. It can also match that information to other data, such as corporate card information.

00:13:21.000 --> 00:13:24.000
Then, when a report

00:13:24.000 --> 00:13:48.000
enters the approval process, AI can present the information to the approver. Today’s example shows how we use AI for expense approvals. We’ve created an AI layer within the approval workflow. When someone submits an expense report,

00:13:48.000 --> 00:14:05.000
the application reads the receipts and attempts to code each receipt to the appropriate expense type, which ultimately maps to the GL code. It also reviews the payment details, including whether a transaction is on a corporate card or is reimbursable.

00:14:05.000 --> 00:14:21.000
At that point, the application begins the AI approval layer and runs a check. This DBee reference is the DATABASICS AI tool. We call it DBee, and you’ll see it in action in a few minutes.

00:14:21.000 --> 00:14:38.000
DBee provides an initial verdict and rates how well the expense report passed the AI check. It might give the report a perfect 100% score or another percentage, such as 75%, and identify what looks good.

00:14:38.000 --> 00:14:55.000
It also shows what is wrong with the report. The user can open it, review the results, and then move it into the human workflow—the organization’s standard approval process.

00:14:55.000 --> 00:15:12.000
When I review expense reports from my team, the application may say, “These are all 100% good.” I’ll still look at them and give them my approval. If any reports raise questions, I can examine what happened. The application has already done

00:15:12.000 --> 00:15:32.000
much of the hard work by showing me what does and doesn’t need review. The queries are set up in natural language through our administrative module. The application evaluates the receipt data

00:15:32.000 --> 00:15:49.000
and brings the results into the application before sending the report to me for approval. AI can also be configured to restrict

00:15:49.000 --> 00:16:06.000
what it does and doesn’t examine. Depending on the use case and deployment, AI may not need to check everything. You can instruct it to review only certain transactions or transaction types,

00:16:06.000 --> 00:16:18.000
individuals, or types of travel. With that type of deployment, AI operates only where you want to use it,

00:16:18.000 --> 00:16:29.000
which helps control costs by limiting AI interaction to the areas where it is needed.

00:16:29.000 --> 00:16:45.000
Why and where do we still need human judgment? This is another major area we discuss with organizations. I compare it to when we first deployed expense tools 25 or 30 years ago.

00:16:45.000 --> 00:17:03.000
Organizations liked the ability to implement highly detailed business policies. However, deploying those solutions sometimes frustrated end users because the organizations could lock down expense reports so tightly. We’re not saying we’re against

00:17:03.000 --> 00:17:19.000
compliance, controls, or locking down expense reports. We’re saying that some deployments may go too far. When deploying these solutions, we advise customers to walk before they run. AI is in the same

00:17:19.000 --> 00:17:35.000
boat. AI can examine receipts in great detail. In a recent example, someone at a pizza parlor ordered a vodka pizza. While reading the receipt, AI checked for

00:17:35.000 --> 00:17:54.000
any term that referenced alcohol, so it flagged the pizza as containing alcohol. It was smart enough to identify “vodka” and “pizza,” but it still locked down the expense report. We can then tell the application that when the term is used in combination with

00:17:54.000 --> 00:18:13.000
pasta or other food items, it shouldn’t flag those line items. There is a learning curve. The important question for organizations is: Where do you really need AI? They should also understand that after deployment, they will still need

00:18:13.000 --> 00:18:28.000
some form of human intervention, unless they want the application to do everything on its own. When evaluating AI for expense reporting, there are a couple of major areas to consider.

00:18:28.000 --> 00:18:46.000
Privacy is probably the number-one consideration. Make sure the application supports zero data retention and offers a model that can be deployed specifically for your organization.

00:18:46.000 --> 00:19:02.000
You should also be able to create highly specific policies. In our deployment model, each organization has a dedicated environment. The last consideration is, again,

00:19:02.000 --> 00:19:09.000
continuing to include human intervention in the process.

00:19:09.000 --> 00:19:25.000
Now, let’s discuss the questions to ask vendors. I’ll start with value: What does the solution cost, and what are you trying to accomplish?

00:19:25.000 --> 00:19:44.000
You should also examine your strategy and the policies you’re trying to implement. I’ll show you some receipt policies in a moment. AI can be implemented at almost any level—not only during approval, but also during audits.

00:19:44.000 --> 00:20:00.000
The important point is to determine which policies you need. Once those policies are defined, vendors should be able to demonstrate them. When an organization asks us, “Can you show us how you handle X or Y?”

00:20:00.000 --> 00:20:17.000
We’re happy to set that up. Another consideration is how easy it will be for you to create and maintain individual rules. Your vendor should show you exactly how the solution will work,

00:20:17.000 --> 00:20:28.000
provide a suitable pricing model, and let you see the solution in action so there are no surprises after you

00:20:28.000 --> 00:20:43.000
deploy this. Now, from our perspective, again, very similar to the items we just talked about, you're going to look at a secure environment out there. We're always making sure that we're going to reference this right around your data, and then the other piece here is who has access to it

00:20:43.000 --> 00:21:02.000
Today, we’re focused primarily on AI for approvals. AI capabilities are also developing for end users, audits, and support. One of the other major

00:21:02.000 --> 00:21:18.000
questions is: How deployable is it, and where is the value in each of these areas? Once you’ve answered that, build a roadmap focused on your critical processes.

00:21:18.000 --> 00:21:35.000
Define the potential ROI. Examine the time and effort spent on internal processes: How long does it take to reconcile corporate card data or review receipts? Are you reviewing every receipt even though

00:21:35.000 --> 00:21:52.000
you don’t require receipts for items below a certain spending threshold? Set goals, deploy the AI, and evaluate its results. We always recommend treating AI as a supplement, not a complete replacement for your resources. We

00:21:52.000 --> 00:21:56.000
view it as an

00:21:56.000 --> 00:22:11.000
ancillary tool that supports users and simplifies the process. As I’ve mentioned, we’re still at the beginning of this technology and are determining whether it will be an expensive or

00:22:11.000 --> 00:22:32.000
a more affordable solution. Ultimately, that ties back to ROI. If AI detects costly anomalies, the return can be quick. If it’s used only to identify minor issues, it may not be as cost-effective.

00:22:32.000 --> 00:22:51.000
With that, let me show you what we’ve been discussing and then wrap up. You’re looking at a live environment of my internal DATABASICS time and expense application.

00:22:51.000 --> 00:23:09.000
I’m not going to take you through the entire expense entry process. Instead, I’ll show you how we use AI for approvals. After logging in to DATABASICS, I can create expense reports. Here are some reports I’ve created and submitted.

00:23:09.000 --> 00:23:25.000
Some have been approved, while others are awaiting approval. The pending reports came from my team over the last couple of days. I asked the team to submit them specifically for this webinar.

00:23:25.000 --> 00:23:40.000
Several expense reports are shown here. In the third or fourth column, you can see their status. In this example, Mike Duggan submitted an expense report,

00:23:40.000 --> 00:23:56.000
It has entered the approval process. Internally, the report will go through several approval steps. First, it goes through our AI process. Then I approve it, followed by someone in our accounting department.

00:23:56.000 --> 00:24:13.000
This bottom report is an actual expense report. I’ll open it so you can see what it contains. It has a single line item: a business meal for Mike. You can also see that the expense was on a corporate card.

00:24:13.000 --> 00:24:26.000
Returning to the previous screen, you can see that this report went through the AI approval process and received a 100% score, meaning the

00:24:26.000 --> 00:24:37.000
transaction—or, more accurately, the report—matched the AI’s initial query. The next report received a 77%

00:24:37.000 --> 00:24:58.000
grade on that report. And the reason for that is that these top transactions you'll see here are all approved again by the AI. But you'll notice that there's a flag down here, and what this is, is it's catching a double pay matching a company card transaction that came through on June 10th, linked to a separate expense report.

00:24:58.000 --> 00:25:14.000
Again, I asked Mike to do this, so that is what took place. And then up here we've got one from Jason Wade. And what you'll see here is that this application has actually detected alcohol being purchased within there

00:25:14.000 --> 00:25:30.000
If I enlarge Mike’s expense report, you can see where it sits in the overall approval process.

00:25:30.000 --> 00:25:48.000
The first step was AI approval, which produced the score we saw and created a note. Marcel is the AP approver, and I’m the approver for the DATABASICS sales team.

00:25:48.000 --> 00:26:09.000
The report is still awaiting my approval. After I approve it, it moves to the next approval level. Once that person approves it, the report is ready to be posted to the expense platform. This demonstrates how the application creates an AI approval level.

00:26:09.000 --> 00:26:24.000
The review can become extremely granular. For example, we could tell the application, “Review the receipts, and anytime one mentions Modelo or Tito’s Vodka,

00:26:24.000 --> 00:26:41.000
reject it—but if it mentions Absolut Vodka, allow it.” You can configure a very detailed review. Consider how long it would take someone in your organization to perform that same review manually.

00:26:41.000 --> 00:26:58.000
The application prepares the report to simplify the approval process. DBee also supports natural language queries. I could ask whether I have expense reports to approve or how much was spent with United Airlines, and it will generate reports.

00:26:58.000 --> 00:27:14.000
We still provide a full-featured reporting platform as well. Here are some of my previous expense reports. Let’s take a quick look at a report

00:27:14.000 --> 00:27:24.000
showing how AI worked within an expense report I completed a couple of weeks ago.

00:27:24.000 --> 00:27:40.000
AI is reading my receipt and showing that it wasn’t itemized. The receipt shows a sales payment with a tip, but no itemized purchase detail, so AI is telling me that an itemized receipt is required.

00:27:40.000 --> 00:27:54.000
Certain line items don’t show an AI approval because AI reviewed them and didn’t need to take any action. On this line item, alcohol may have been detected.

00:27:54.000 --> 00:27:59.000
I’ll continue reviewing the report to see whether there are any other flagged line items.

00:27:59.000 --> 00:28:18.000
This one is flagging that I didn’t produce a line item. That appears to be the last one. The main takeaway is that AI can help organizations simplify expense processes. We also see many other areas where AI can provide an advantage.

00:28:18.000 --> 00:28:36.000
Today, we wanted to introduce what you should look for, show how AI works, and explain the critical components an organization should consider when rolling it out. One last point before we wrap up:

00:28:36.000 --> 00:28:52.000
If you’re interested in seeing how this could work for your organization, we’d be happy to arrange a meeting with our team—or even with me. A URL for requesting a demo is shown here.

00:28:52.000 --> 00:29:07.000
We’ll also follow up with my contact details. If you’d like to see how we can put your data and receipts through the application and build an AI strategy and process around them, we’d be happy to demonstrate that.

00:29:07.000 --> 00:29:26.000
Thank you for joining us today. We look forward to speaking with you. We’ll send you a recording of this webinar, which you’re welcome to review or share. Thank you, and have a fantastic day.