​​AI-Driven Acquisition: Keeping Humans at the Center​

Welcome to the Government Technology Insider Podcast. I'm your host, Lucas Hunsaker. Government acquisition is stepping into the spotlight like never before. With increased pressures to drive efficiency, reduce fraud, waste and abuse, and ultimately deliver mission outcomes faster, agencies are rethinking how acquisition is done and how emerging technologies like AI and automation can play a supportive role.

In this episode, I sat down with Benjamin Allen, Vice President of Public Sector Solutions at Appian. And in our discussion, he shares how acquisition modernization is evolving in response to new priorities and technological advances. But it's not just about technology. As agencies adopt AI and automation, leaders must also address workforce concerns, maintain human oversight, and avoid the risks of cognitive surrender.

Join us as we discuss how acquisition professionals can embrace modernization while keeping humans at the center, ensuring that technology acts as a partner not a replacement in delivering smarter, more effective government acquisition. I hope you enjoy our conversation today. Well, thank you so much for sitting down with me today, Ben. It is a pleasure to speak with you and I'm looking forward to our conversation.

Thank you for having me. Ben, could you first introduce yourself and explain your role at Appian to our audience? Yeah, I'd be happy to. I'm Vice President at Appian, responsible for our acquisition solutions.

I joined Appian about 20 years ago, and I started in our professional services organization, implementing acquisition solutions for our government customers, both federal civilian and Department of War. And then about seven or eight years ago, I transferred over into our product group and now responsible for the creation of acquisition solutions on our Appian platform. Ben, from your perspective, what's really changing right now in government acquisition? Why has it taken center stage recently?

A lot is going on in federal acquisition right now. I've been working in acquisition for over 20 years. And up until really maybe a year, year and a half ago, acquisition was basically a it was a quiet place in the government. Hard, important work was being performed, but acquisition wasn't really the main event.

And it certainly wasn't the spotlight. And that has completely changed recently. It certainly kicked off with a string of acquisition-focused executive orders that put in motion efforts like the revolutionary FAR overhaul. And we're seeing the biggest rewrite of acquisition rules and regulations in a generation.

The focus is shifting now from process compliance to mission speed and supply chain resilience. A much greater effort is being spent on consolidation and driving common spend through single sources like GSA to allow the government to operate as a single buyer now. And these are all exciting changes, but at the end of the day, change is hard. And it's especially hard when so much of the previous versions of regulations and policy was supported by decades-old rigid technology.

And now they're trying to get that acquisition workforce to work and behave differently with a tool set that's really stuck in the past. So we're starting to see organizations recognize this and demand new acquisition tools. And it couldn't be happening at a better time because we're also seeing so much change in technology with the recent advancement of AI and large language models. It's really this perfect time to modernize because the technology has finally caught up with the mission's needs.

Now where are you seeing AI and automation having the most immediate impact across the acquisition lifecycle Ben I think we seeing applicability really throughout the entire cycle I think AI is making it easier to find information to generate content to conduct robust research to make recommendations about the most advantageous acquisition paths. I can think of AI use cases for every part of the acquisition cycle where you can make meaningful impact. But if I had to pick just one area, I believe the part of the acquisition lifecycle that historically has been the most underserved by technology is requirements development.

The introduction of AI in this phase, the process, it has the potential to make such a tremendous impact. And it's not just in that phase, but all the downstream work that relies on that requirement. Now in acquisition, getting requirements right is absolutely critical. So how can AI help agencies improve that early stage of the process?

I think the biggest way to improve the process is making it easier for program offices to articulate that requirement. Most of these users, they're not acquisition experts, but AI allows us to meet them where they are. Instead of a checklist and a lot of blank templates, we now have AI-driven interfaces that guide the user through the process. So now the application doesn't just say, you need to upload your performance work statement.

An application leveraging AI says, based on your goal of X, here's a draft performance work statement. I've started it based on similar acquisitions, and I have information about your new requirement. I've begun the document. I just need you to finalize it and review it.

That's a tremendous shift from where we were before. Now the application isn't just a system to input data. It's an active participant to help in the creation of that requirement. But hidden in that capability is something important that I also want to highlight.

To make the AI work well, you need to provide it with relevant data. You can't just go to a chatbot and say, write a PWS. It'll do it, but it'll be awful and almost certainly not based on your real requirement. But thankfully, the government, it's sitting on a goldmine of untapped data.

Right now, that data is buried in PDFs, public data sets, old contract files and folders. But with access to it, AI can instantly find similar procurements from the past to help accelerate that requirements definition. It can look at Q&A from similar solicitations and help address questions that were asked before. Look at well-performing acquisitions and look at the contract vehicles and their contract structure and use those in your new acquisition.

In the past, you could never ask a human to just quickly review the past 10 years of related acquisition history and see what's in there that you could leverage. But with AI, you can, and you can do it in a matter of minutes. Additionally, I think AI can really help speed up the overall acquisition time by reducing that back and forth that we see between program offices and contracting. Usually, the program office submits a draft, it waits two weeks, it gets it back covered in red ink, but now AI acts as that first line of review.

It provides immediate feedback on compliance and clarity and consistency. This turns the contracting office back into a strategic partner rather than just a proofreader. They spend their time on high-level strategy because the AI handled the administrative hygiene of the documentation. I couldn't be more hopeful and excited about the impact AI is having on the requirements phase.

There's such a tremendous opportunity here to use AI to improve the quality and the speeds of the requirements being created. Oh, excellent. Let's shift the conversation towards challenges. Now, what are some of the biggest human or organizational challenges agencies face when modernizing acquisition?

Yeah change management is always difficult And any barriers to widespread adoption that what will prevent the agencies from realizing that promise of AI I meet with agency leaders often and the topic of their agencies, the rollout of AI capabilities, it's a frequent topic. Most agencies have started with rolling out a secure agency-wide chatbot. Think of it as a secure chat GPT or Google Gemini. It's general purpose, chat prompt driven, and the ability for users to navigate to it and use it securely whenever they have a task sort of as they see fit.

Initially, they saw tremendous demand and feedback from a very loyal group of users. Each iteration of that capability was immediately dissected by the users, and they had lots of feedback. And agencies were really excited about this. But as the agencies tried to roll out the capability to more users, they noticed the adoption slowed.

The small but powerful users knew exactly how to leverage the AI tools, but the vast majority of their workforce did not. So a typical change management response went into effect. They had training sessions and self-service learning materials. They established champions within the organizations, and that's helped to some degree.

But for organizations looking to modernize acquisition with AI, they need to stop thinking of AI as a sidecar application and start looking at how it can be delivered directly in the acquisition system. Can you tell us a little bit about the concept of cognitive surrender and why it is important to keep humans in the loop with AI? I think cognitive surrender is one of the bigger risks in rolling out AI, and that more people should be talking about it. Too often we get caught up in how much AI can automate or how much AI can create.

I've heard from some agency leaders that have some very lofty aspirations of AI doing everything. Let's fully automate the acquisition process. But I wanna be clear, AI excels with the proper amount of input. And that can be upfront input, it can be from users or historical datasets, or it can be input provided along the process execution path.

But if I simply say I have a requirement for building maintenance, go create my requirements package, or even go create my PWS. An AI can create a very convincing document, but it's also at a very high risk of not matching your actual requirement. Do you want AI making assumptions about the types of maintenance or the location, the types of required service calls, the required response times needed, how quality will be assessed, what security requirements you have? The obvious answer is no, you don't want AI making those assumptions about any of those areas.

So how do you mitigate this? Well, you make sure you have more information up front, and you make sure you involve the human throughout the process, and you divide the reviews and work into smaller pieces that are easier to accomplish and help drive better AI output downstream. If we simply generate a large document in a single shot and say, okay, user, go review this, there's a good chance that the user will either do a fairly cursory job, or they might not even review it at all.

And this is where that cognitive surrender happens. Just assuming that AI is doing good work and the work is accurate and not even reviewing or thinking about it as a human and the ramifications of it are really costly. You might be requesting the wrong performance requirements. Vendors would be bidding on incorrect specifications.

And in the end, the mission is not receiving the right good or service. With cognitive surrender, we go from making the procurement more efficient to actually making it much less efficient. As organizations evaluate AI and applications within their procurement process, they must be keenly aware of that risk of cognitive surrender and how to combat it. Now Ben how do you see AI changing the day work of acquisition professionals and what does that mean for workforce development over time I think how we seeing AI change day of acquisition professionals is not that dissimilar from what we seeing in other industries.

AI is able to make the worker more efficient and AI is able to assume many of the time-consuming administrative tasks and generally free up the worker for more strategic work. Let's use software development as an example first. Many organizations have found that AI greatly accelerates their ability to produce code, but they've also found it doesn't replace the knowledge and experience of senior developers needed to create a robust and scalable application. But the senior developer is now much more efficient, and what they used to delegate to junior developers can now just be assigned to AI.

That's led some companies to let their junior developers go and stop hiring them altogether. But senior developers, they don't grow on trees. They mature over time, first starting as junior developers and then increasing their knowledge through more complex tasks. Then they become the senior developers.

You have to keep that pipeline going, or you're heading for a cliff when those senior developers retire or move on. Now, I'm not advocating for reducing the use of AI so we can give these tasks back to junior resources, but I am highlighting the need to reimagine what their career progression looks like. The government will still need entry-level acquisition professionals if they can mature up from a GS7 all the way to a GS14 or 15, what that enablement process will look like is just going to change.

That's the part that we need to figure out. What does the new process and career progression look like and how can we get more people excited about it? Now as we wrap up our conversation and we start to look ahead, what does a modern AI-enabled acquisition environment look like when everything is working the way it should? I'll start with what I think it doesn't look like.

I don't think it's a chat interface. I view the chat interface like a DOS prompt. For early users of the computer, the DOS prompt was great. It was powerful.

But for the vast majority of users, it was not accessible. It wasn't until Windows and a graphical user interface that the computer really became a powerful tool for the masses. And that's what we need for AI to be meaningful for an organization. We need widespread adoption.

So I'd offer that an AI-enabled acquisition environment is one where the AI is baked into the application. I'm not typing into a chatbot what PSC or NAICS do you suggest for my building maintenance procurement. The application is simply providing its suggestion for the user on the requirement in the application for them to review. And I'm not tasking an AI agent to perform market research.

The agent kicks off automatically when the new requirement is entered and then reports back its findings to review with the user. Building AI into the application ensures it's aware of all the contextual data surrounding the procurement without that user having to manually provide it through copying and pasting. It ensures the AI is used where it's most appropriate without the user having to think, is this a good use case for AI? How would I prompt it?

And it's integrated in a way to ensure auditability and that user reviews and involvement are included throughout the entire process. The true benchmark for success here is simple. You'll know AI is working for your agency when your users don't have to learn how to use it. It's just a tool powering the acquisition capabilities.

Thank you, Ben, for joining me today and a big thank you to our listeners for tuning in. If you'd like to stay up to date on the latest best practices, lessons learned, and proven strategies for leveraging innovative technology in federal, state, and local government, be sure to visit us at governmenttechnologyinsider.com. I've been your host, Lucas Hunsaker, and until next time, so long.