AI Is Coming for Contracting

AI isn’t likely to take your job, at least for the foreseeable future. But it certainly is going to change it.

By Annaliese Trenchfield

Pilot tests across government and industry demonstrate that artificial intelligence (AI) has a big future in contracting. It isn’t likely to displace human contracting professionals. But it has begun reshaping their jobs.

AI soon may be automatically flagging high-risk clauses in your solicitations so you can remove them before they can trigger protests. It is being trained to draft contracts that are accurate and consistent in their wording, structure, and formatting. In the near future, it could produce high-quality documents for your review tailored to specific agency needs and handle post-award and closeout activities.

Already, AI is combing vendor data to produce determinations of responsibility in 1/60th the amount of time it takes a contracting specialist. It is reviewing solicitations for insertion of Section 508 accessibility requirements, and pulling information from the Contractor Performance Assessment Reporting System (CPARS) to assist in source selections.

AI isn’t likely to take your job, at least for the foreseeable future. But it certainly is going to change it, most observers believe.

RPA, AI, and ML

Companies and federal agencies increasingly have adopted robotic process automation (RPA), artificial intelligence (AI), machine learning (ML) and natural language processing (NLP). The use of these technologies has grown as organizations seek to modernize operations, in-crease efficiency, reduce costs, and provide more accurate and timely services.

AI adoption has more than doubled during the past five years, according to a December 2022 McKinsey survey of 1,492 participants representing a range of regions, industries, and company sizes. “In 2017, 20 percent of respondents reported adopting AI in at least one business area, whereas today, that figure stands at 50 percent,” the survey found.

According to a 2020 report by Deloitte, instances of RPA, AI, and ML in government tripled between 2018 and 2019, totaling more than 3,000 by the end of 2019. This trend is expected to continue.

RPA uses software bots to automate repetitive, manual tasks. These bots are programmed to mimic the actions of a human user in performing a variety of tasks, such as data entry and processing or interfacing with existing systems to automate processes.

Federal agencies use RPA to automate document processing, data entry, and analytics. The Army developed a determination of responsibility assistant bot (DORA) that pulls information about vendors from government databases and websites and creates reports for contracting specialists in less than five minutes as opposed to the hour it takes to perform the process manually. DORA was modeled on the Internal Revenue Service (IRS) procurement office Contractor Responsibility Bot.

AI is a branch of computer science that involves the development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.

In 2019, the Department of Homeland Security (DHS) Procurement Innovation Lab (PIL) began an effort to use AI to aid contracting professionals in applying contractor past-performance information during source selections. Using a DHS commercial solutions opening pilot program, the PIL engaged five companies to create software-as-a-service demonstrations using AI to identify which records in the Contractor Performance Assessment Reporting System (CPARS) contain the information most relevant to a given source selection. The Artificial Intelligence for Past Performance Project also sought data-driven, evidence-based recommendations about opportunities to improve CPARS data quality.

ML is a branch of artificial intelligence that enables machines to learn from data and make predictions. The Department of Defense (DoD) is applying ML to develop predictive models used to schedule maintenance on equipment. In October 2022, the Army Materiel Command (AMC) struck an $85 million deal with Palantir Technologies to forecast equipment maintenance needs and end of life via predictive modeling software. AMC seeks to improve the Army supply chain by cutting the time that missing parts or equipment breakdowns damage readiness. DoD AI spending grew from just over $600 million in fiscal 2016 to $2.5 billion in 2021 for more than 600 projects as of April that year, according to the Government Accountability Office.

The advent of AI and ML has revolutionized the way humans interact with technology and machines. AI and ML are being used in a variety of applications from facial recognition to voice recognition to natural language processing (NLP). NLP involves the use of words, gram-mar, and context versus structured data, such as appears in spreadsheets.

The General Services Administration created a solicitation review tool (SRT) that uses natural language processing to assist AI in reviewing bid solicitations. It identifies those related to information and communication technology, and then ensures they include Section 508 requirements to improve accessibility.

Since 2018, the IRS procurement office has used AI to review solicitations and contract documents with its Contract Clause Review Tool, which identifies missing, outdated, or incorrect provisions and clauses. Clause reviews went from taking six hours when performed by contracting staff to six minutes with the bot.

The IRS DATA Act Bot combines AI, NLP, optical character recognition, and RPA to read and modify contracts and related documents. Its name was derived from the 2014 Digital Accountability and Transparency (DATA) Act. The IRS created it to clean up Federal Procurement Data System data. Since then, it has been used to add telecommunications security language to 1,466 IRS contracts, and review and correct coding in COVID 19-related transactions at less than a quarter of a second per fix.

In 2017, the Health and Human Services Department (HHS) began prototyping Accelerate, which combined blockchain, AI, RPA and ML to apply a neural network to “analyze contracts and to show discrepancies in prices paid for the same goods across the department. A proof of concept using 18 months of FY 2016-2017 HHS spend data showed the differences in prices paid by HHS organizations for each good or service bought.

The test discovered more than 200 contracts with VMware with significant price variation across the department. HHS entered discussions with VMware and won a consolidated contract to achieve significant cost avoidance.

Accelerate was designed to enable HHS contracting professionals to employ category management to make purchases as a single enterprise, leveraging the full HHS demand to reduce prices. Using the program, HHS combined all its contracting data in a single database structured by governmentwide spending categories.

“Natural language processing (NLP), an AI tool, was used to assess all of the contract language and data in order for the team to ‘teach’ the tool how to restructure the data into logical ‘drill down’ from general categories to specific products/services and their pricing,” according to a 2018 report on the BUYSMARTER initiative of which Accelerate was a part.

Additionally, combining and analyzing contract data centrally allowed HHS to analyze contractor performance departmentwide and alert contract managers to violations of contract terms and conditions. Accelerate included use of a chatbot to answer contracting staff questions.

Accelerate machine learning combined with linear regression allowed HHS to look at past performance to predict which contract types might encounter problems. The program offered recommendations for changing policy or personnel to avoid the issues.

Natural Language Limits

Despite the many AI and ML advances, limitations persist. One of the main ones is the inability to fully understand natural language. AI and ML are not yet able to process natural language in the same way that a human can. This can render NLP and other AI/ML capabilities incapable of effectively interpreting and responding to natural language.

NLP and other AI/ML capabilities also struggle to understand context. They lack the ability to interpret and understand the complexities of human communication, so they are not able to accurately interpret and respond to nuances. AI and ML are not yet able to accurately predict human behavior. They can process large amounts of data and information, but they cannot yet forecast how humans will interact with it – for example, how humans will use a product or service or whether a marketing campaign will succeed. Finally, AI is only as good as the data put into it, and procurement and contracting data, still largely input by human hands, remains notoriously incomplete and messy.

Using AI to Change Contracting Itself

Despite their limitations, AI and ML hold the promise of significantly streamlining government contracting by identifying and prioritizing contracts, analyzing bids and contract language, and speeding up the review process. AI and ML also can be used to identify and mitigate potential risks, such as fraud or waste, during the review process.

Using AI to analyze large amounts of contract data, such as historical information, pricing and performance data, and contract terms and conditions, can identify areas for improvement improving the process itself, as HHS Accelerate and the IRS AI programs demonstrate.

HHS Accelerate shows how AI and ML can be used to better manage and monitor contracts to ensure they are being executed in accordance with agreed-upon terms and conditions. Accelerate then generates recommendations that can be used to improve contract performance and draft better future contracts.

An AI Editor Looking Over Your Shoulder

The Defense Department’s Chief Digital and Artificial Intelligence Office (CDAO) is working together with Trenchant Analytics LLC to build a contract writing system powered by AI that will aid humans in rapidly writing user requirements, calls to industry, solicitations, and other transaction agreements.

The goal is to greatly accelerate the government contracting process. The CDAO is responsible for adoption of data, analytics, and AI to generate DoD decision advantage. It works to ensure the department’s technology is effective, secure, and efficient, and to develop digital and AI capabilities.

The CDAO-Trenchant contract writing system will enable humans to rapidly write contracting documents. This is a significant task, as these documents involve complex language and require a deep understanding of the government contracting process. The system will employ NLP to automatically generate documents that are accurate and consistent in their wording, structure, and formatting.

Using AI, the system will be able to rapidly produce high-quality documents that are tailored to specific DoD needs, eliminating many of the delays and inaccuracies that often occur in government contracting. CDAO and Trenchant Analytics hope to drastically reduce the time it takes to complete a contract, freeing up contracting personnel to focus on more complex tasks.

Mobius Logic Inc. and the Air Force 341st Contracting Squadron presented at the NCMA 2022 Government Contract Management Symposium in December 2022, demonstrating a developmental AI system for identifying protest-prone clauses in contract solicitations.

The company is building the prototype under a Small Business Innovation Research contract with the 341st Contracting Squadron. Collaborating with the squadron’s contracting officers and specialists, Mobius plans to introduce AI tools into the contracting process beginning at the solicitation phase and concluding with post-award and project close-out activities.

Augmenting, Not Replacing Humans

AI, ML, and NLP are not replacements for human capabilities; they are augmentations. The promise of RPA, AI, and ML technologies in government contracting lies in their ability to automate administrative tasks and provide greater insights into the data associated with a con-tract. They can expedite processes and reduce the time and effort associated with managing contracts, analyze contract data to identify trends and patterns, provide insight on contract performance, and allow agencies and companies to make more informed decisions.

The use of RPA, AI, and ML in contracting also can improve data accuracy and reduce the risk of errors. In addition, AI and ML can identify discrepancies or inconsistencies in contracts, allowing organizations to catch errors earlier and reduce the risk of protests.

The future of AI has been the subject of hot debate since the introduction of generative AI to the general public via DALL-E-2 and ChatGPT late last year. Generative AI uses unsupervised learning algorithms to create new digital images, video, audio, text or code. ChatGPT uses material on the Internet to generate new written products in response to human prompts.

DALL-E-2 does the same for images. Both were released for public use in 2022. ChatGPT demonstrates that AI can produce consistent, cogent, and persuasive text that closely mimics human writing. The catch is that it can only do so accurately when the information already exists on the Internet.

“It generates responses in extremely oversimplified terms, by making probabilistic guesses about which bits of text belong together in a sequence, based on a statistical model trained on billions of examples of text pulled from all over the internet,” according to The New York Times.

Faced with a direction to write about a new subject, the AI sometimes will make up authoritative-seeming text and even populate it with official-sounding sources, including fictional names and titles. On other occasions, generative AI will produce near-gibberish. That is why OpenAI, ChatGPT’s creator, included an accuracy disclaimer when it opened the AI for public use in December 2022.

Similar problems dog DALL-E-2, OpenAI’s system for creating realistic images and art from descriptions in natural language. Users have noted that it returns images demonstrating gender bias and racial stereotypes and its pictures often are overly sexual. It has problems rendering faces and simply cannot fulfill some user requests. When AI can create the art users request, however, it creates fascinating and even prize-winning pieces.

While these programs are not yet being applied to contracting, their very public shortcomings are cautionary about the need for very specialized skill in refining prompts for generative AI and in carefully and closely training it for use in highly regulated environments such as procurement.

Still, its very trainability and capacity for “learning” from its mistakes make it likely that AI will improve rapidly. No doubt faster than our ability to imagine all the next uses to which it will be put. It cannot write a contract today, but who knows about next year or the next decade? CM

Annaliese Trenchfield is the pseudonym for Trenchant Analytics LLC, whose president, John Ferry, tuned and prompted OpenAI’s Generative Pre-trained Transformer 3 (GPT3) to produce the body of article, and the DALL-E 2 extension to generate the imagery. Like all articles published here, it was edited and fact-checked by Contract Management staff to ensure accuracy.


“The state of AI in 2022—and a half decade in review,” Quantum Black, McKinsey, December 6, 2022,

Chirico, Elizabeth, “The Modern Perspective, Army ALT Magazine, May 28, 2021

Innovation Alley: Department of Homeland Security, Procurement Innovation Lab,

“Artificial Intelligence: Status of Developing and Acquiring Capabilities for Weapon Systems,” Government Accountability Office, GAO-22-104765, February 17, 2022,

Solicitation Review Tool, Section, General Services Administration,

Kanowitz, Stephanie, “IRS deploys bots to streamline procurement processes,” GCN, January 29, 2021,

Krebs, Cory D., “How Can the DoD Adopt Commercial-Style Artificial Intelligence for Procurement?” Naval Postgraduate School MBA Professional Project, June 2020,

“The BUYSMARTER Journey: Our Successes Thus Far,” Summer 2018, Reimagine HHS, BUYSMARTER,


Roose, Kevin, “The Brilliance and Weirdness of ChatGPT,” The New York Times, December 5, 2022,

Roose, Kevin, “An A.I.-Generated Picture Won an Art Prize. Artists Aren’t Happy.” The New York Times, September 2, 2022,