Get Ready for the AI Revolution – AI-Assisted Capture Management

Use Generative AI to Develop Effective, High-Scoring Capture Strategies and Solutions – Q&A Recap and On-demand Video

Join us as we delve into Generative AI’s dynamic impact in revolutionizing capture management. Bruce Feldman, Lohfeld Consulting Principal Consultant, AI and Capture Management Training Specialist, and Christian Ferreira, CEO & Cofounder, Procurement Sciences AI, share insights into how to use Generative AI to develop effective, high-scoring capture strategies and solutions efficiently. And be sure to check out our Q&A session recap below.

Get Ready for the AI Revolution – AI-Assisted Capture Management (watch on YouTube)

Here’s your opportunity to grasp the practical applications of AI, unlocking enhanced efficiency and accuracy for your next capture.

• Understanding AI’s Role in Capture Management: Learn how AI technologies can be integrated into capture management.
• Competitive Advantage: Learn how AI can help produce strategies and solutions that are compliant, responsive, and compelling.
• Practical Tools and Strategies: See how Procurement Sciences’ AI platform facilitates capture.

Recap of our live Q&A Session from Get Ready for the AI Revolution – AI-Assisted Capture Management

(The following Q&A session was edited for clarity.)

Question: What lessons have you learned from helping your customers use these processes and tools to support capture and proposal development?

Bruce Feldman’s Answer: The main thing we’re learning so far is that if you interact with Generative AI in the way it’s optimized for, which is as an assistant or a helper but not an oracle that produces complete and perfect answers you then rely upon, it can really enhance your productivity. You can draft a lot of capture artifacts with Generative AI very well and very quickly, but you still have to take ownership of the final product. You have to check the facts; you have to make sure the case is made the way you desire. Those are the essential aspects I think of using GenAI—the principal lesson learned we’ve gained.

Christian Ferreira’s Answer: I agree, Bruce. I think it’s all about the efficiency gains. There are undeniable gains that you can get from using this new era of technology, but at the end of the day, it still comes down to the people using it as a tool to build the best final product. Allow the AI to help you with the efficiency, but all those efficiency gains you should take and ultimately roll back into building your best possible proposal submission.

Question: Would you discuss AI-assisted capture management’s potential risks and limitations and how these can be mitigated?

Bruce Feldman’s Answer: Building on some of the material in the presentation, a couple of important lessons emerge. One is that if your company is going to commit to trying to use or using GenAI, which I think most companies will, you really need to have an effective way to bootstrap your program. You can’t just send a couple of people to a training class, go buy a platform, and off you go, ready to go to work. You have to be prepared to invest some time and effort in bringing your people up to speed on how to use AI effectively. I think that’s absolutely key. Even as a productivity enhancement tool, just learning your way around the capabilities of the platforms—how they work, the different aspects, and how to use the platform capabilities in a way that best emulates your proposal development or capture workflow—I think really becomes key to getting the most value out of it.

At the same time, doing things that run contrary to how GenAI works can create risks for you. You need to check your facts, for instance, at the end of the day. You need to make sure that you’re not just generating generic content and that you really have company capabilities that differentiate you well embedded in your responses. Those are some of the risks you have to deal with. They’re not really risks in the sense of causing you damage if you operate on the premise that the AI is your partner and a tool, but it’s not the answer machine.

Christian Ferreira’s Answer: I agree. The biggest thing is don’t just use it as an easy button where it can automate a lot of the work for you; it looks very powerful, but there are limitations. You still need to verify the information. It’s going to guide you to where it got the answers from, but at the end of the day, it’s the actual user/operator/human that is going to be the one supervising what the AI is producing. It’s really up to the human to go in there, identify that the information’s accurate, make sure it aligns to their company, and then fact check it to ensure that what they’re submitting is actually what their company is able to perform and has done before.

Question: What are some of the potential issues regarding AI and Controlled Unclassified Information (CUI) or other restricted information?

Christian Ferreira’s Answer: That is a big issue right now. It’s not just CUI data, but ultimately, it’s all company data, and there are a lot of free public tools out there extracting your data for their own use. A lot of people are using their company information and feeding it into these tools. These tools don’t retrain in real-time so right off the bat, it’s not evident that you just leaked a bunch of information. What these public platforms do is they take all the information you put out there, put it into a separate database, and they train their future models on it. It really is a problem across the board right now—what people don’t realize is that the companies that own these models are actually taking your data and will be using it for training future models.

When you move into the CUI level, that’s a whole other level because it’s government-controlled information. But it’s the same thing for anyone accessing government-controlled information—your cyber teams and your CISOs should be dictating how you maintain that information, how you’re storing it, and where you can use it. Any sort of public AI system, I can tell you right off the bat, cannot handle CUI data. If you have a system that can handle CUI data, what that means is it’s probably an on-prem deployment into your infrastructure governed by your CISO team and your cyber team, and there’s going to be very strict training around how to use it and how to use it effectively. The short answer is any time you use any sort of tool in your workforce, you should always be checking with your cyber teams and your CISOs; they’ll be giving you the guidance that you need as far as what you can use, what you can’t use, and also what information can be put into those systems.

Question: Can you talk a bit about how Procurement Sciences stores proprietary information?

Christian Ferreira’s Answer: Ultimately, these large language models, which are really the engine behind the systems, don’t train in real-time. It’s a big misconception that people think that these models are constantly learning and growing in real-time. That’s not the truth. They’re all stateless, which means once they’re done being trained and you put the lid on top of the model, it doesn’t continue to learn. That’s where we use algorithms, as Bruce mentioned in his presentation, called RAG, where we can take your company’s proprietary information and store it in your own private database (it’s called a vector database), and then we have algorithms that allow the AI to go into that database and use the information to build relevant answers without risking any of your data being made public, used for training larger models, or exposed to anyone else in the system.

Question: Where do these tools pull the information from? Will it give suggestions that competitors may see if using a similar software?

Christian Ferreira’s Answer: There are a couple of different sources. As Bruce alluded to, there’s training data, which is how these models are first trained. Those models are typically trained on open-source information from the internet, usually about 30% to 40% of the entire open-source historical internet. You have to think these large language models right off the shelf are extremely powerful. They’re subject matter experts across huge domains.

Obviously, that’s powerful, but it’s not tuned to your company or to government contracting, proposal writing, etc. That’s where that RAG algorithm comes in, where ultimately, you can bring additional data into the system, whether it’s from open source, federal APIs, internet searches, scraped websites, or uploaded information from your company—there are multiple different sources that you can build on top of. You take that training data, but then you upload and then store it separately for the training of their future models, which ultimately are then made public where others can leverage and access your information that was leaked.

Question: Is Procurement Sciences’ AI platform collaborative?

Christian Ferreira’s Answer: It is. We are building more into it right now. Collaboration is key in proposal writing, and we know this strongly as most of our team came from a government contracting background. We are adding some things in there so it matches more closely to more traditional word processing tools that are cloud-based, where you can work together on a document. There are further things we’re adding to it right now to make it much more of a collaborative platform across the entire life cycle for management of opportunities as well.

Question: I understand that my data will be siloed and not used to train other companies or groups, but will it continue to learn within my company’s vector?

Christian Ferreira’s Answer: By default, no. It comes back down to these large language models being stateless, so they don’t actually continue to learn. The vector database is different than the large language model. We do have algorithms that we’re building right now that do build an ability for it to continue to learn about your company, your users, what you’ve done in the past, what you’ve bid on, what you haven’t bid on, etc. But by default, large language models do not continue to learn with you. That’s more of a proprietary algorithm that we’re working on.

Question: How does one inform or manage the language in AI-generated responses, for example, exclude terms on Lohfeld’s legendary list of words to avoid?

Bruce Feldman’s Answer: In some of the platforms, you can set up what I would call a system prompt, which becomes a fixed part of your identity inside the system. You can, for instance, set up a system prompt that says Gen AI, I never use these words. If you want to give it a system prompt that says, I prefer to use these sorts of words, you can say, I want to use verbs that are transitive verbs that take a direct object. Word selection is absolutely essential, especially when you’re directing the GenAI, because different words don’t mean the same thing. The word list does not mean the same thing as the word summarize, for example. That’s an obvious example.

When you want to get a specific output format, you have to pick your verb carefully to direct the GenAI to do the task you want and not the task you think you want. What I recommend is as part of your GenAI proficiency development, you have to calibrate your language to the GenAI and get used to how it thinks and talks in response to you so you can adjust how you instruct it and query and prompt it to do what you need.

Christian Ferreira’s Answer: I want to add onto that. That was a great answer, Bruce. What I would say is for people who haven’t used GenAI before, the easiest way to think about it is just like you would talk to a human and give them feedback—you can do the same thing to these GenAI tools. Most of these GenAI tools have a generic chat interface on top. They also have more advanced tools, but using those chat interfaces, you can tell these AI systems, just like I would tell Bruce, I don’t like how you wrote this; I don’t want you to include these words. It’s just like talking to a human. The short answer to your question is you can just give it that list of keywords and say up front, remove these, never use these, and it’s going to follow that exact command for whatever it’s building for you.

Bruce Feldman’s Answer: I think it’s worth noting the large language models that form the heart, the creative part of these GenAI tools; they’re language-triggered. They have a natural language processing front end and back end, and they create a very detailed and complicated mathematical representation of language and how words are associated with each other. Word selection, especially in your prompts and your instruction to the GenAI, becomes really essential. One of the things we teach is to give the GenAI lots of context and lots of background on what you want to know but keep your direction in the task relatively short and to the point. That way the GenAI has the right kind of language and word associations that steer your request to the right place in the model to give you the kind of information and response that you want.

Question: Do you have any advice for a team looking for AI software and what to look for?

Bruce Feldman’s Answer: One of the things we’re doing at Lohfeld right now is looking at doing some benchmarking on comparing AI platforms. We think that it’s essential that we understand how they vary, what they can and cannot do, and how well they do it.

There’s really, I think, very little information out there on the comparative performance and capability of the tools. There’s a tremendous amount of subjectivity because again, we’re judging the way AI manipulates ideas and uses language. It really becomes very much a subjective set of elements, but we’re going to try to get our arms around that by looking at some of the typical tasks a proposal writer or a capture manager might propose to have GenAI help with and then compare responses. Then we’ll probably do some benchmarking against public platforms as well. Again, not using any proprietary data, but you want to compare it to the standards like ChatGPT and the new Google tool called Gemini, that sort of thing.

Christian Ferreira’s Answer: To piggyback on top of that answer, there are a lot of different factors that play in. Obviously price. What is your price range? How many people are you trying to get on the platform? Are you trying to use it for just one or two people, or are you trying to empower your entire workforce? Do you need an on-prem solution? Do you need a multi-tenant? That ties into what that total price tag is going to look like. Functionality is big as well. There are a lot of AI tools out there right now. Are you trying to use an AI tool just for proposal writing where there are a bunch of tools popping up, or are you really trying to automate the end-to-end process? Do you really want to get into the intricacies of capture or business development, specifically into government contracting or specifically into commercial contracting? All things to think about. There are a lot of tools.

The biggest problem we see now is that we were one of the first ones to market. Now there’s a whole bunch of tools out there, and all that’s going to do is make it more complicated for companies to select the right tool, because there’s so many out there, there’s a lot of white noise. Just look for people that have been proven; look for people that actually have customers, especially in the government contracting realm. You don’t want to be one of those top 5 or 10 customers for a company in its very early days. That’s very risky, especially if they’re using your data, regardless of what they say about data security, privacy, and not training. I would probably stay away from trusting an early small company like that until they get the first couple of guinea pigs out of the way.

Bruce Feldman’s Answer: Another related question is to what extent should you invest when you start to bring GenAI into your platform? Not just for proposal work, but for really any knowledge worker type task? There are a number of articles from companies like Gartner and McKinsey on how to bring AI into your enterprise. What’s the most effective way to do it? You might want to do some research, do some searching, and go see what they recommend for your scale and size of enterprise.

Question: Does your software produce a capability matrix? If not, is that on your roadmap?

Christian Ferreira’s Answer: It does. We have two different tools. We have a traditional tool where you can just upload the document and it shreds it for you, and we also have the capability through the AI chat. The biggest difference is that through the AI chat, you have a lot more flexibility as far as what you want it to do because you can be very creative. Typically, requirements or compliance matrix, it’s going to be the requirement number, what the requirement was, and what that compliance keyword was. That’s the most basic level, and you can do that with traditional shredders where you upload a document, and it builds a CSV for you with each field.

Using the AI chat tool with a custom prompt gives you the ability to add on top of that. You can actually go in there and tell the AI to identify liability risks, help me match to past performances, give tips to differentiate my company, tips to respond, things along those lines. There’s a lot more power when you actually go through the AI chat. Then we also have that traditional pre-AI method that some people like, which is to upload a document, spit out a CSV, and then go in there and add their finishing touches to it.

Question: Can you talk more about what training you provide if onboarding the software to your team?

Bruce Feldman’s Answer: We have a proposal for onboarding, which we and Procurement Sciences do jointly. We recommend you have a certain size team that engages in the environment with actual proposal data. We have a 3-day training class that’s designed to give people a thorough grounding on what Generative AI is, what its capabilities are, what its limitations are, and then some fundamentals on how you can use AI-assisted tools—really from any platform—to support your workflow. Again, be it capture or proposal-related work.

At the end of that, you’re at a place where you can really self-start. We recommend that in the environment in which you’re working, you have some people with you who understand the proposal process very well and also have some experience using GenAI to facilitate it because learning through trial and error can be pretty painful. You need a mentor or two in the environment to help you with that. The onboarding doesn’t really need to go much more into documentation or formal training; at least, Christian and I would contend that. I think once you reach a certain point in the curve, you self-start, you learn as you go and you can ascend the curve pretty quickly.

Christian Ferreira’s Answer: I agree. I think there are a lot of training classes popping up right now. They’re very generic though, and that’s where it’s good to learn the generic information, but if you’re using GenAI specifically for government contracting or specifically for proposal, business development, or capture, that’s definitely where a partnership with Lohfeld comes in because they offer that class, that methodology, how to use AI effectively, safely within your teams specifically for those use cases versus more the generic mindset of how can I use a generic tool for just generic purposes.

Question: What tools are you aiming to add to the system that will facilitate capture?

Christian Ferreira’s Answer: We have a pretty interesting roadmap. For us, our advantage was we were one of the first to market. All of our team comes from government contracting, so we understood the pain points very closely. We are at the tip of the spear for AI. AI is changing every single week. There are new massive breakthroughs taking place so often. We’re looking at those every single day and asking ourselves questions—how can we apply it to the end-to-end workflow that a contractor or government contractor facilitates within that kind of capture, proposal, and BD process? I don’t want to share too much of our secret sauce in our roadmap, but there are a lot of exciting, very powerful things coming out in the near future.

By Beth Wingate, President, Lohfeld Consulting Group, APMP Past CEO, APMP Fellow Lohfeld Consulting Group has proven results specializing in helping companies create winning captures and proposals. As the premier capture and proposal services consulting firm focused exclusively on government markets, we provide expert assistance to government contractors in Capture Planning and Strategy, Proposal Management and Writing, Capture and Proposal Process and Infrastructure, and Training. In the last 3 years, we’ve supported over 550 proposals winning more than $170B for our clients—including the Top 10 government contractors. Lohfeld Consulting Group is your “go-to” capture and proposal source! Start winning by contacting us at www.lohfeldconsulting.com and join us on LinkedIn, Facebook, and Twitter.