Innovations: Industry Perspectives on Contracting With the Defense Department
Introduction by Anne Laurent, NCMA Director of Professional Practice and Innovation
Public outreach, non-FAR deal-making, lots of intellectual property flexibility, intriguing problems, and continuous touch points with buyers and military service members are the attributes that can make the Department of Defense an attractive customer for companies selling the most advanced technologies. Andrew Bowne did a terrific article for the February 2021 issue on the results of his research on what artificial intelligence firms find off-putting about selling to the Pentagon and what can overcome it. In this column, he shares interviews with 15 AI company leaders during the research. We’re hearing more from tech companies about barriers DOD creates for the upstarts it wants to attract, but it’s still unique to have so many voices aggregated in one place. Like his study, Bowne’s interviews should be mined for many lessons as the need for national security innovation intensifies.
BY ANDREW S. BOWNE
The Department of Defense (DoD) is at risk of failing in defending the United States and its allies in the age of artificial intelligence (AI).1 The National Security Commission on Artificial Intelligence (NSCAI) warned that the United States would not regain its leadership position unless it took whole-of-nation actions to employ AI responsibly.2
The report makes clear the government cannot achieve this mission alone – it needs the help of industry.3 However, the DoD faces challenges in attracting committed partners in industry.4 According to the NSCAI, the traditional procurement process is too slow, too complex to scale, and too challenging to convince venture capital to back startups that attempt to work with the DoD.5
The conclusion by the NSCAI is shared by Congress. The recently signed 2022 National Defense Authorization Act has urged the DoD to reduce the barriers to entry for commercial AI firms.6 This continues a trend of Congress issuing legal requirements in-tended to enable the Defense Department to leverage commercial innovation in AI and make it easier for those innovations to enter the DoD via contract.
To better understand the perspective of commercial AI firms regarding defense contracts, I conducted interviews with a cross section of the AI industry.7 To encourage candid discussions, I committed to strict confidentiality, so interviewees are not identified. The interviews provided personal perspectives on the problems and best practices company leaders have encountered in contracting with the DoD.
Understanding the perceptions and opinions of commercial AI firms can lead to better practices and policies that can make the DoD a more attractive customer.
Whether they worked at traditional defense contractors or startups that had never competed for a DoD contract, the interview-ees highlighted differences between commercial contracting and defense contracting. Many interviewees shared the views of Con-gress,8 the Government Accountability Office,9 the Office of the Secretary of Defense,10 and the NSCAI that contracting with the DoD can be difficult for industry. The most cited pain points in working with the DoD were the length of the process, complex intellectual property (IP) rules, lack of meaningful communication with end users, overly specific and inflexible requirements, and lack of expertise in AI and machine-learning capabilities and limitations.
Lengthy Lead Time
“We filed a proposal sometime around two years ago. We’re still waiting on a decision,” an interviewee from a large public corporation reports. While not all contracts take years to award (many factors can contribute to the delay including funding delays, acquisition requirements, or litigation), long lead times to award a contract are a significant concern for AI firms.
According to multiple startup founders, long lead times mean they risk running out of funding before contract award. Other early-stage firms explained the lengthy process also discourages venture capital, limiting options for startups wanting to work with the DoD. “Startups go out of business waiting for their government contracts,” said an executive from an Austin-based startup.
Because AI advances so quickly, lengthy processes can lead to obsolete technology. “By the time this contract actually gets executed, in many cases, the things that we promised, we wouldn’t even want to deliver because they are legacy software,” says an interviewee from a large public corporation.
Many AI firms are aware the DoD has authorities, such as commercial solutions openings11 and other transaction agreements (OTA),12 that can streamline the process to be more aligned with commercial requirement statements, timelines, and negotiability. Some companies seek out those opportunities exclusively. “If we didn’t have SBIR13 and OTA, it’s just too long of a process. It’s too protracted. It’s too many hoops you have to jump through,” says an executive from a Boston-based startup.
While there are many considerations leading to contract award that must be weighed by the contracting team, it appears that too long of a lead team can risk losing innovative AI firms as competitors for contracts with the DoD.
Complex IP Framework
The DoD intellectual property framework is daunting, especially to new entrants to the defense market.14 “For all the emerging technologies firms, the burden to learn the government IP system, including everything that comes along with it, is just a hill too far,” says an executive with significant defense contracting experience. One executive captured the general view of all interviewees. “IP is a critical factor. IP is fascinating for AI companies, and it poses unique problems.”
Because the DoD licensing and data rights framework is often very different from commercial licensing, many interviewees say their firms had to hire expensive attorneys to ensure they weren’t inadvertently releasing their most valuable IP. “If you’re not careful in creating restrictive licenses, then you’ve suddenly thrown your stuff out there for all to use, which would be deadly,” says one interviewee. Others feared the DoD could release their IP not through negligence or malice, but rather due to a firm’s failure to fully understand the consequences of the data rights package provided in a DoD contract.
IP fears lead some firms to avoid the DoD altogether. “Not knowing what you don’t know is part of what is going to keep the average commercial-oriented startup away,” an interviewee observes.
The importance of IP to AI firms leads to a desire to negotiate commercial-like licenses rather than traditional data rights frameworks. As one executive explains, “IP is our number one concern because IP is our company. So, when I see a defense requirement, and I’ve seen two of them recently, that says, ‘If you’re not going to give us government purpose rights or unlimited rights15 on day one, don’t bid,’ then that’s a no-bid. As soon as I see government purpose rights, unlimited rights, as a condition, I’m out, unless I have an opportunity to negotiate.”
As IP is an important consideration for both industry and government, it is important for the requiring activity to work with the contracting officer and legal counsel to identify requirements and intended use of the AI model early. It is also important to under-stand the need to communicate those requirements with industry.
Some uses may only require data rights in model data and not necessarily the algorithm or parameters that may be proprietary. However, other use cases may require at least government purpose rights, especially use cases that are intended to make predictions that will inform command decisions on personnel or operations. Several executives at AI firms explained that understanding why the DoD is requesting data rights can help during negotiations.
Rules and Procedures
Firms with limited experience with the DoD recognize the opportunity afforded by a defense contract but view it as limited to traditional defense contractors. One AI startup executive avoids working with the DoD because of a perception that the contracting process is complex and dominated by traditional defense contractors.16
“How do you create the kind of marketplace that somebody on the commercial side would be able to access?” the executive asks. “My impression is that if you’re going to sell to DoD, then you need to know the players, you have to know a whole vocabulary as well as have an old-boy network to know who’s doing what.
“And if you’re sitting in Silicon Valley, you have no idea what the old-boy network around some subdomain of defense contracts looks like at all. The only way [the DoD will] get Silicon Valley-type startups to start engaging is if there’s a really proactive push to lower and demystify the barriers so startups can do business more or less as they do today. That means finding a way to do it such that they don’t have to know what FAR is and all the other acronyms.”
The DoD efforts to address the perception that the system is for traditional contractors have been noticed by commercial AI firms. Organizations such as AFWERX17 and the Defense Innovation Unit18 that have endeavored to engage with the broader commercial innovation base were known to each of the interviewees, even ones that never competed for a DoD contract.
While the opinions about these organization and their outreach are positive, most AI firms remain intimidated by the law, regulations, and policy that apply to defense contracts. Reducing the use of jargon in solicitations or using industry standards were cited by the interviewees as helpful in lowering those barriers.
Interviewees universally noted that many DoD buyers lack AI knowledge. Data pipelines, compute, storage, model development and deployment reportedly are poorly understood. While most interviewees strongly preferred commercial contracting processes compared to the defense model, there was not necessarily a desire for the DoD to mimic the commercial sector completely. “If the DoD understood the tech, whether or not they behave like a business would be irrelevant. They would buy technology the way that technology should be bought,” one interviewee explains.
The lack of AI knowledge is compounded by a perceived lack of education about available procurement best practices. One interviewee surmises that a better understanding of AI would help inform better DoD procurement practice for AI, including drafting requirements, selecting a contract vehicle, and source selection evaluation. “The [DoD] doesn’t know how to buy [AI],” the interviewee says. “They don’t know how to assess it. Things get sold because they’re well marketed, not because they’re technically superior or meet some technical thresholds. It’s no secret. The National Security Commission on AI has talked about it.”19
However, there is optimism about DoD organizations that focus on providing education in technology and acquisition. As one interviewee remarks, “I think [the problem] is an education problem, which is what the [Department of the Air Force/MIT] AI Accelerator is doing.20 I have high hopes.”21
What Attracts AI Firms?
According to the interviewees, AI firms are attracted to contracts that enable open communication and collaboration between contractor and customer to solve exciting problems. AI companies seek flexibility in performance and freedom to negotiate essential terms and conditions. Some DoD organizations already recognize that flexibility and collaboration are critical to attracting AI firms and can achieve technological goals.
According to the interviewees, flexibility to experiment and iterate potential solutions and collaboration with the end user are essential contract attributes for an AI firm. Negotiating clauses, notably IP, is strongly favored over regulation-driven and requirements-focused contracts. Open-ended problem statements are very popular, hence the typically high praise for AFWERX’s Small Business Innovation Research (SBIR) open topics,22 the DIU commercial solutions openings,23 and the Defense Advanced Research Projects Agency (DARPA) prize challenges.24
Many interviewees explained that the traditional acquisition requirements approach with detailed specifications is antithetical to developing and deploying machine-learning models. As one executive explains, “nobody can define a solution today within are-as of AI,” so detailed specifications can lead to confusion or a missed opportunity to find an innovative solution. Instead, interviewees would prefer that the DoD let offerors know the capability gap and allow them to determine whether AI can solve it.
Open communication is critical for AI firms to understand the data pipeline, the problem to be solved, and computing and storage infrastructure. An executive with extensive experience working with the DoD explained that contracting for AI “requires insight into data type, model type, and end user. I can go into J.P. Morgan, and there’s a team of 60 data scientists, and we can sit down with them for hours.” The executive explained that J.P. Morgan-type communication is critical to understanding the training data, the use case, and the how best to model the problem, but rare in the DoD.
Open cross-functional communication within the DoD between the requiring activity, end user, and contracting team can help potential contractors better understand and fulfill the government’s needs.
Collaboration and working with service members are attractive to many AI firms. For most interviewees, the culture at AI firms favors collaborative efforts over purely transactional contracts. This is why many of the AI firms interviewed prefer other transactions to FAR-based contracts.
A Silicon Valley-based startup executive explains, “[w]e like OTs because we have more interaction with the end customer. You don’t really get into a blackout period like [requests for proposal] until the final phase, so you can ask some questions.” Asking the end user questions helps firms understand the nature of the DoD problem and brainstorm solutions.
The ability to negotiate, iterate, move quickly, and use commercial-like processes make OTs attractive. “When you’re working in an OT, you’re breeding a little more thinking outside the box, which is another positive,” says an interviewee.
However, several interviewees explain that many OTs do not fully leverage the flexibility of the authority. Agreements officers often default to inserting FAR clauses, they say, limiting the flexibility and ability to collaborate and negotiate that AI firms find attractive.
In addition, several interviewees described concerns about transitioning successful OT prototypes to ongoing funding. Al-hough OTs for prototypes can result in follow-on production agreements,25 transitioning to a program of record is difficult, several interviewees reported.26 This problem is more a product of programming, budgeting, and funding than contracting. The perception that the DoD touts contracts such as SBIR or OTs as a pathway to programmatic funding can lead to disillusionment, however, if many of those awards don’t advance beyond a prototype or minimum viable product.
Eric Schmidt, NSCAI and Defense Innovation board chairman and former Alphabet CEO famously told Congress, “The DoD violates pretty much every rule of modern product development.”27 While it appears that some leaders at commercial AI firms would agree, there is room for optimism.
What AI firms find attractive – contract flexibility, ability to solve challenging problems, collaboration with end users, and ultimately delivering AI-enabled capabilities that help save lives and protect liberty – can be realized under current contract law. However, attaining national security goals appears to require educating the total force, including contract professionals, about the capabilities of AI and how to leverage flexible contracting authorities that attract commercial AI firms. CM.
Andrew S. Bowne is a Ph.D. candidate at the Adelaide Law School, University of Adelaide, Australia. This article is based on research conducted as part of his Ph.D. studies.
Disclaimer: “The views expressed are those of the authors and do not reflect the official guidance or position of the United States Government, the Department of Defense or of the United States Air Force.”
Statement from DoD: The appearance of external hyperlinks does not constitute endorsement by the United States Department of Defense (DoD) of the linked websites, or the information, products, or services contained therein. The DoD does not exercise any editorial, security, or other control over the information you may find at these locations.
1 See National Security Commission on Artificial Intelligence (NSCAI), Final Report (March 2021) 19-20.
2 Ibid 23-7.
3 Ibid 1.
4 Ibid 65.
5 NSCAI, Interim Report (November 2019) 60.
6 Fiscal Year 2022 (FY22) National Defense Authorization Act (NDAA), Pub. L. 117-81, § 227.
7 The interviewees comprised business decision makers, such as chief executive officers, executive vice presidents, directors of sales, directors of business development, senior managers, chief operating officers, and directors of federal or defense programs at commercial firms that use or provide AI-enabled solutions for defense-relevant use cases. The firms represented ranged from start-ups of fewer than 10 employees to some of the largest technology firms in the world. Firms ranged from self-funded to publicly traded corporations with the intent of understanding the impact cash flow and source of funds have on the business calculus of a commercial AI firm in deciding whether to compete for a DoD contract. They also varied in experience with the DoD. The spectrum of experience with the DoD was selected for the interview sample: firms that had never considered working with the DoD; firms that attempted but have been unable to secure a DoD contract; firms that perform some DoD contracts that make up a small proportion of their business; dual-use firms with roughly equal commercial and defense portfolios; and defense contractors whose primary, if not only, customer is the DoD. Finally, the firms were based in a variety of regions, including Silicon Valley, Austin, Boston, Denver, New York, Seattle, and Washington, DC. To encourage candid discussions, the research is committed to strict confidentiality. This article focuses on the industry perspective. For the perspectives of DoD contracting officials, see e.g., Stan Soloway, Jason Knudson, and Vincent Wroble, Other Transaction Authorities: After 60 Years, Hitting Their Stride or Hitting the Wall (2021) 7; Danielle C. Tarraf et al, The Department of Defense Posture for Artificial Intelligence (RAND, 2019) 56.
8 See Senate Armed Service Committee, FY2022 NDAA Report (2021) 251.
9 U.S. Government Accountability Office, Military Acquisitions, DOD is Taking Step to Address Challenges Faced by Certain Companies, GAO-17-644 (July 20, 2017) 9.
10 See Office of the Under Secretary of Defense for Acquisition and Sustainment, Other Transactions Guide (2018) 4-5.
11 Commercial solutions openings are a streamlined solicitation method that allows an agency to seek innovative solutions from the commercial market. Class Deviation—Defense Commercial Solutions Opening Pilot Program, 2018-O0016 (June 26, 2018).
12 Other transaction agreements (OTA) are contracts for research or prototype projects under statutory authority that are exempt from traditional procurement laws and regulations, such as the FAR and Bayh-Dole Act. 10 U.S.C. §§ 2371, 2371b.
13 Small Business Innovation Research (SBIR)is a program that encourages small business to perform federal research and development with the potential for commercialization to stimulate innovation. 15 U.S.C. § 638.
14 See NSCAI (fn. 1) 62.
15 See DFARS 252.227-7013 and -7014 for definitions of license rights.
16 See NSCAI (fn. 1) 62 (explaining that the “prospect of bureaucratic snarls deters companies from working with DoD; it is economically irrational for many startups to even try”).
17 AFWERX is a partner for commercial technology innovation to rapidly field high-value commercial and military capability and expand the network of innovative Airmen and Guardians. https://afwerx.com/.
18 DIU partners with organizations across the Department of Defense, from the services and components to combatant commands and defense agencies, to rapidly prototype and field advanced commercial solutions that address national security chal-lenges. https://www.diu.mil/.
19 The NSCAI has emphasized that the DoD has an “alarming talent deficient” in AI and recommends overhauling talent management to build digital fluency. NSCAI (fn. 1) 120-4.
20 The USAF-MIT AI Accelerator is designed to make fundamental advances in artificial intelligence to improve Department of the Air Force operations while also addressing broader societal needs. https://aia.mit.edu/.
21 Disclaimer: The author is currently assigned to the Department of the Air Force/MIT AI Accelerator, though this interview was conducted before this assignment. For more information about the DAF/MIT AI Accelerator, visit https://aia.mit.edu/. The DoD AI Education Strategy, written by the Joint Artificial Intelligence Center, details the DoD’s education plan, recognizing the total force requires continuous education and upskilling to meet the DoD’s AI Strategy objectives. DoD AI Education Strategy (September 2020) 2-4, https://www.ai.mil/docs/2020_DoD_AI_Training_and_Education_Strategy_and_Infographic_10_27_20.pdf. Several organizations within the DoD, such as RAPIDx in the Department of the Air Force and the Defense Acquisition University, have start-ed to provide training to program managers and contracting officers on industry outreach and innovative contracting practices.
22 AFWERX developed the Small Business Innovation Research (SBIR) Open Topics to increase the efficiency, effectiveness, and transition from research to operational capability. AFWERX, “SBIR Open Topics,” https://www.afwerx.af.mil/sbir.html.
23 The Defense Innovation Unit developed their Commercial Solutions Opening (CSO) as their solicitation and competitive evaluation process focused on simplicity for the commercial solutions provider. See Defense Innovation Unit, “Work with Us,” https://www.diu.mil/work-with-us/commercial.
24 Prize challenges encourage thinking outside the box that starts with a goal without predicting who or which approach is most likely to succeed, making way for novel approaches that might otherwise seem too risky to pursue. DARPA, “Prize Challenges,” https://www.darpa.mil/work-with-us/public/prizes.
25 Other transactions for prototype can result in a follow-on production award without further competition if certain require-ments are met. 10 U.S.C. § 2371b(f).
26 The National Defense Authorization Act for Fiscal Year 2022 created a commission akin to the NSCAI on Planning, Program-ming, Budgeting, and Execution reform. FY22 NDAA (fn. 6) § 1004.
27 Mark Wallace, “The U.S. Air Force Learned to Code—and Saved the Pentagon Millions,” Fast Company (July 5, 2018) https://www.fastcompany.com/40588729/the-air-force-learned-to-code-and-saved-the-pentagon-millions.