AI Document Processing: Boost Speed, Reduce Errors, and Save Time

Shifting from Manual to Accelerated AI Document Processing

Companies across every industry—from healthcare and insurance to finance and manufacturing—process hundreds of thousands (if not more) of documents daily. The sheer volume creates a massive operational cost. And, processing manually becomes a burden that slows down workflows, drains resources, and introduces risk at scale.

Organizations still relying on humans to extract data from PDFs, faxes, or handwritten forms are burning time, introducing errors, and bottlenecking their workflows. AI document processing changes the equation. Introducing AI to automate ingesting documents, interpreting the information with human-like understanding, and scaling instantly, makes operations faster, more accurate, and dramatically more efficient.

In this podcast, I sat down with Ryan Francis, President and Partner, to discuss what AI document processing is, how it works, and why it matters—covering real-world use cases, how AI handles unstructured data like PDFs and handwritten forms, and the clear business value in speed, accuracy, and operational efficiency. Anyone listening will walk away with a solid understanding of how to identify use cases and implement AI to replace manual document workflows.

Read insights from our experts about how to leverage AI document processing for your business!

Read the Transcript:

STEFANIE KULBERG: Hi, everyone. I’m Stefanie Kulberg, and I’m here today with Ryan Francis, the President and Partner here at Launchpad Lab. And we are going to be discussing document processing and how that is relevant for those of you who have applications that you’ve built. For those unfamiliar with the term, I’d love for you to just start by walking through an explanation detailing what exactly document processing is and how it works.

RYAN FRANCIS: Right. Right. Like, basically, up until the last, like, year or two, dealing with unstructured data has been really hard. So, as programmers, we love structured data. We like, like, SQL databases with tables and columns where each column we know that first name column is a string field, and we can expect some type of a short name. Or, like, a date is, like, gonna be a date field and kinda know what to expect with the data.

RYAN FRANCIS: With documents, it’s been really difficult over the years to find ways to automate the intake of documents. Documents come in all sorts of different formats. Sometimes they’re just images, like scans of a document. And so processing a document, pulling data out of a document.

RYAN FRANCIS: A good example would just be like an invoice. You get an invoice and automatically finding and pulling out the amount due and the date that’s due. Historically, it’s been a hard problem to solve in programming. But since the recent advancements in large language models, these large language models can very accurately and not just, like, pull data out of documents, but, like, truly understand the document that it’s looking at even if the document is an image.

RYAN FRANCIS: Right? And truly, like, kinda understand the information being portrayed. So even if, let’s say, for example, the invoice, instead of saying due amount, it just says, like, AMT due or something like that. Traditionally, in programming, you’d have a problem because we’re like, oh, we were looking for the word amount, and it’s AMT instead.

RYAN FRANCIS: So then the program breaks. But with the large language model, it knows AMT is an abbreviation for amount, and it, like, does that human sort of reasoning about the information on the page. So long story short, what this basically means is that any organizations who have unstructured data like documents where they have people that have to analyze those documents and do something with those documents, like pulling data out and putting it in a database or in a CRM, they can now use AI to do that instead. So that’s what we mean by, like, document processing.

RYAN FRANCIS: So, basically, a document comes in and an AI analyzes the document and does something with it, whether that be to write data into a database, make some type of initial determination about the documents, so think about, like, a claim coming in for insurance. The AI can analyze that claim and understand whether or not it should just be immediately approved or if it needs to go through some type of human review.

STEFANIE KULBERG: It’s definitely interesting to think about the different use cases. What I’d love to do though, is ask you about the business value.

RYAN FRANCIS: In terms of business value, a lot of our clients that we’ve built these AI document processing engines with they have, in some cases, ten or more people dedicated to just handling these documents. Ten or more people that just got all they do every day is they take these documents and they pull data out of them and move the process along. So the business value can be pretty high in these situations where you have large teams that are just dealing with pulling data out of unstructured documents. From a business value perspective, it’s like, let’s repurpose those hours to more value add activities in the firm, like actually servicing the client or actually engaging with the client as opposed to pulling data out of a PDF.

STEFANIE KULBERG: I’d also imagine there’s a level of risk with human versus automated data polls where it’s like, I can, as a human, like, type in where it says twenty-five dollars. I might, like, fumble and hit, like, twenty-six. There’s that risk.

RYAN FRANCIS: You’re right. There’s more to it One would be the human error rate. You’re gonna have fat-fingering happening when you’re pulling data out. You do have an error rate with the AI as well, but depending on the type of document, it can go either way. You can have some situations where the AI is just so much more accurate than the humans, or vice versa depending on the situation.

RYAN FRANCIS: Obviously, that’s part of what we do at Launchpad is help assess those situations and figure out what’s a good fit. So if you do find a use case that’s a really good fit for AI, being able to have that consistency in terms of lower error rates is huge. The other thing is the turnaround time. So document comes in.

RYAN FRANCIS: It might be off hours and the human’s asleep, and they wake up the next day and then go in and and it’s in a queue. And that queue takes them all day to get through, and they finally get to it. And before you know it, it’s a day before that gets processed. We see this in insurance where a claim comes in, and it takes them weeks to approve it.

RYAN FRANCIS: Right? Because they’re it’s in a queue, and it’s waiting for a human review. We see this in, like, lending where one of our clients will send out a request for a loan to get a pre-authorization, and they get a PDF back from the lender on whether or not that has been approved and what the interest rate is and things like that. And it could take that human a day before they get to that.

RYAN FRANCIS: So with AI, you have this kinda automated set it and forget it. It’s just gonna get taken care of, experience where as soon as that PDF comes in, it’s handled. You can have the AI burst up and burst down. Right? So you could have one AI worker that scales up to a thousand nearly instantaneously. Suddenly, a thousand documents get dropped on your desk. You can scale up the AI to parallelize the work being done. Whereas humans, you can’t really do that.

RYAN FRANCIS: It takes them forever to hire more people and bring them in, and train them and everything like that. So, handling burst demand, being able to scale up and down, is another major value prop of using AI for document processing as opposed to, like, humans. I wanna also add that, like, beyond just document processing is, like, document analysis. And so that’s even going a step further than just, like, saying, I’m gonna go pull certain fields out of a document.

RYAN FRANCIS: It’s going to it’s going a step further and also, like, analyzing whatever that document is about and creating some type of an assessment. Like, a claim is a good example where it’s like, yeah, you could have a document processing engine that just pulls data out about the claim, who is submitting the claim, what the property is that they’re submitting the claim against or about. But then you could also have that AI go a step further and analyze the policy for that person submitting the claim. Look at the coverage on that policy and determine, based on the claim being submitted, whether or not it seems like there’s good justification for that claim being submitted.

RYAN FRANCIS: And even if the AI is not making a final determination, if they’re pulling some information and bullets together so that a human comes in to assess that claim can very quickly see, oh, here’s the coverages. Here’s what the claim is generally about. And, like, they can move a little faster through that process. Like, that still adds a lot of value.

STEFANIE KULBERG: Definitely a lot of value added there. I’d love to hear you share some stories of client work that we’ve done to provide some sort of tangible examples for those unfamiliar with how this can actually work in practice.

STEFANIE KULBERG: I’m thinking of one client in particular where we did a lot of work with them, where they had multiple people analyzing faxes and printed content, and it just took a lot of time. So I’d love to have you walk through that example so that way other people can just kind of understand the practical application. Would you keep me honest here and walk through that specific instance?

RYAN FRANCIS: I think probably taking documents and processing those, that’s basically anytime a doctor or physician needs a lab done for their patient, they would have their office send an order, which comes in the form of a fax document. A lot of times, handwritten on what they needed, you know, what they need. So our software that we built using AI takes that fax document, understands what’s being requested even when it’s handwritten, and creates an order request in their order management system or lab, like, information management system. It creates, like, an order in there, and then a human reviews it, makes sure the data is good, and and and I think they have, like, ten plus people that are just dedicated to taking in these orders, processing them. It’s greatly speeding up that team’s time.

STEFANIE KULBERG: In terms of taking that and applying the concept to something that’s would be applicable across businesses, what would you say that that would what’s the use case that applies across businesses?

RYAN FRANCIS: It’s like PDF ingestion or something, like ingesting data from documents. You know, in an ideal world, you don’t wanna have people sending you documents that you have to process. Ideally, they’re, you know, submitting a form on a website that goes into, like, your database.

RYAN FRANCIS: But the reality is, sometimes you can’t decide how they send you information. Let’s say you’re a brokerage, insurance brokerage. You’re working with a couple of other carriers, and your client is interested in health insurance for their company. You have to go to those carriers, provide them with a bunch of data on here.

RYAN FRANCIS: Here’s the population for my client. Here are the people that they have employed and their ages and all that kind of stuff. You can’t say, like, and I need you to go to my portal and submit your quote in this format, so it comes in a standardized way. Those carriers are gonna send you a PDF whether you like it or not.

RYAN FRANCIS: It’s basically their proposal for what plans they’re gonna offer your employees, which are gonna have coverages and premium amounts and all that kind of stuff. So being able to, as a brokerage, be able to automatically take that data from that PDF and move it into kinda like a standardized structured database where then you can compare and contrast these different options, that would be a major advantage for you as a brokerage to be able to shrink down the time that it takes to go from client interested in a insurance to client has clear options in a presentation from the brokerage.

RYAN FRANCIS: So that’s just, like, an example of a situation where, like, you just don’t carry enough power in that relationship with the carrier to, like, change the way that they give you information. It’s really more like, what are those kinda unique documents that are kinda unique to your business in a way? So if you’re in health care, the, you know, unique document might be the order for your labs. Whereas if you’re in financial services, it might be a loan document.

RYAN FRANCIS: Or if you’re in manufacturing, it might be the engine spec PDF that you get from the manufacturer that you have to process, like in Kawasaki’s case. Because that’s actually a use case for Kawasaki is, like, processing these engine image documents that they get from their manufacturing team.

STEFANIE KULBERG: But I think what’s interesting about what you’re saying is that to give a good use case, at least what I’m hearing, is that a lot of the industries are highly regulated. So you can do this in a way that’s compliant because there are a lot of pen and paper processes that are required. We’ve got HIPAA requirements. You’ve got PHI when you’re dealing with health care, when you’re dealing with financial services. You’ve got all the regulatory issues.

RYAN FRANCIS: I think, like, document processing maybe invoices is the best one because it’s like everyone knows invoices come in all different shapes and sizes, and, like, it’s kind of like an order form a hundred percent. Like, I could go one vertical at a time and just rattle off things, but it’s funny how that works.

STEFANIE KULBERG: Where my mind is sort of going, but previous life legal service platform, so you get a whole case for an upcoming trial you have, and you have seven hundred zillion papers that land on your file. How can you parse that information? It can ingest all that information for you.

RYAN FRANCIS: Mhmm.

STEFANIE KULBERG: And then surface that in a smart, easy way for you.

RYAN FRANCIS: I think the, like triage concept is good as well. Basically, dealing with inbound things that you can’t control as a business. Can’t really control the format it comes in. How do you, like, deal with that in a more automated way? You know, if someone sends you something, you have to, like, do something on it. But the broad topic here is just triaging stuff automatically, you know, with documents being one of the things that you might need to triage.

STEFANIE KULBERG: Thank you, Ryan. This was a very interesting conversation, and I really appreciate your taking the time today. And I hope those of you listening are walking away with some really good insights on how you can take documents and better process them. So that way, you can enrich and deliver a very valued experience, not just for your organization, but for your customers as well.