Colvin Run Networks, Inc. reposted this
It seems like everytime I look at my feed, there’s a new procurement or Acquisition tool using Generative AI. How are these products differentiating themselves? I think the ability to produce content is definitely convenient, but I don’t think it solves the ultimate government procurement problem — building or buying things that solve a problem, accurately, dependably, and continuously. If anything, used improperly, accelerated content creation could increase the risk that the government accrues by widening gap of knowledge that procurement entities have with the things they’re buying. If the procurement designers don’t understand the words that are being generated, and the Model generating the words doesn’t understand the nuanced details of government environments — it could result in more public funds being wasted, and less civic or Defense capabilities being delivered. Why does this matter? For the last five years, RFPs have been stuffed with commercial buzzwords and newly invented jargon. This has put a major strain on delivery, and has resulted in entire programs resetting and starting from scratch— some for the 3rd or 4th time. We don’t understand or appreciate that behind every buzzword is a scoping/delivery challenge. Words mean everything in an acquisition. Companies literally build their compliance and responses to words — not understanding them deeply enough could lead to catastrophic procurement outcomes. 💡 For B&P, words literally map to currency. How could acquisition shops or GenAI Acquisition Product developers reduce the risk associated with a misalignment and misunderstanding of the content being produced? - Focus on accuracy — ensure that the use of Generative AI is helping scope bids more accurately through Requirements generation/collaboration, Rapid Prototyping, Experimentation, etc. - Find a way to conduct RAG in an extremely thorough and detailed way, using details from previous procurements — successful and unsuccessful. Details are key and so is the schema used to transform the data. - Define niche government environments and settings using vector embeddings to build contextual libraries that are considered when LLMs produce content based on Acquisition prompts. Model ATO stuff, details - Wrap useful processes and workflows around RFP development that forces teams to “pre mortem” their procurement strategies. “Why will it fail?” … “What could have prevented it?” - Always descope…always! Take the complexity of whatever jargon an LLM spits out for an RFP and multiply it by 10, because of the constraints within government environments. - Also, don’t put too much stake in the content related to RFP responses. Now that everyone has a “word butler”, they can accurately describe anything. But, can they build the thing? Lean on tech challenges, pitches, and tangible snapshots of delivery demonstration to augment the decisions that are being made around selecting partners. #government #acquisition #genai