Corpus isn’t a word that I’ve seen very often, at least until the last 18 months. If you were looking for a “collection of written texts,” it usually followed with a fine-tooth combing of an extraordinarily long series of documents to find a single passage. If you were lucky, you knew the section the passage was in. If you were not, you were looking for context clues to narrow down the possibilities. In my experience, the information was never in the write place.
Now, corpus is synonymous with training data. The “collection of written texts” can now be stored in a GPT of our own making. Hopefully, making life easier for all involved. While this use case will be very popular in regulatory settings, let’s step beyond compliance, ESG, and legal industries to focus on how Generative AI can help improve bespoke manufacturing.
Requirements gathering isn’t a step for a lot of bespoke manufacturing. It is the focal point of the process. Teams are given reams of documentation around the size, weight, strength, and, of course, what the product must accomplish. If requirements change, which they often do, the requirements team must comb through the new documents to understand and integrate the new requirements, while adjusting to changes in old requirements.
As you can imagine, the process is arduous and time intensive. We think there is a better way.
Generative AI has primarily been used, thus far, as a chat interface. A user types in a query, and the chatbot reaches into its knowledge base, its corpus, to fashion a response. This is a helpful step forward for users, but it falls short of being transformative. A transformative step would be what we refer to as Ask, Answer, Audit.
While utilizing the same basic framework, the framework improves and optimizes on a chatbot response. A GPT is programmed to identify and pull requirements forward. In the corpus, it may state that the casing has to be 5.5 inches by 8.2 inches. Within the output document, the GPT will state, the dimensions of the case must be 5.5 inches in width by 8.2 inches in length. More importantly, it also provides a location for this piece of information. “Found in Document Physical_Requirementsv2.docx. – Page 5 line 7.”
Now, the requirement is not only captured, but it is auditable. There is no guessing on whether the information is correct. And auditing the accuracy is not hunting for a needling in a haystack.
As for changes in requirements, it can track those, too. If in PhysicalRequirementsv7.docx , the dimensions have changed to 5.3 inches and 8.0 inches, the GPT can find the original requirement and add an addendum to the original requirement, with a new audit trail, of course. Tracking the additional changes can help have conversations with the customer about the number of changes and how it has affected costs of the project.
This use case is a game changer for bespoke manufacturing. It can completely change the process and project timeline for teams. In companies that I’ve talked to, they have three teams that work together to build their products: requirements, design, and build. But the requirements team takes up the vast majority of the project timeline and need to be consistently consulted with to manage changes to the design and build phases.
Integrating Generative AI into a bespoke manufacturing company would shrink the requirements gathering phase dramatically and refine the actions into a requirement auditing phase. Four months of manually processing paperwork could be contracted into about two weeks.
Let’s take a look at what the Generative AI-enabled process may look like:
1. Team receives the requirements documents and loads them into the GPT
2. Team hits run on the requirements gathering program. GPT condenses documentation into 30-40 pages of requirements.
3. Requirements team audits the 30-40 pages of requirements. Adjusts the requirements, as needed.
4. Once the initial requirements document is complete, the bulk of the team is released to the next project. A small skeleton crew stays with the project for consistency and further adjustments.
5. When changes come in, the requirements are amended for changes and additions. Both or annotated to ensure team sees the adjustments and can audit them appropriately.
I believe there are two primary opportunities to drive ROI with the Ask, Answer, Audit use case: increase manufacturing output through more projects and tracking changes.
In my conversations, it seems like the requirements team is usually the bottleneck in a bespoke manufacturing organization. Not only is it the bottleneck, but it is a larger part of the organization. Additional personnel mean additional costs to be managed. More importantly, an additional design or build team member can provide a lot more marginal work output to a single project. By increasing the ability of the requirements team to grow their output, you are essentially making it the marginal cost cheaper for the company to take on new projects.
Pricing is difficult for many bespoke manufacturers. At the beginning of the project, there is a budget. With the inevitable changes, the budget is busted. But how can you present the evidence to your client for additional resources for the project. Step 5 in the new process means that every time a set of changes are provided by the client, the team near immediate feedback in how the requirements have changed for the project. Sometimes these are small changes. Many times, they are not. Depending on the size of the change order, you may be able to have the customer an amended bill in 24 hours. The quick turnaround changes the conversation with the client; you are not having a conversation about the cost two weeks after it arrived but 24 hours. Their changes have real time costs.
A lot of the focus from Generative AI has been on chatbots. Usually, how do we simplify interactions, both internally and externally? By looking at how we can rebuild processes to streamline work with Generative AI, and quickly be able to check the work it has completed, we find a way to not only improve outcomes, but limit the downsides of model hallucinations create internal and external headaches when the models are not properly monitored.
The great news, as briefly mentioned at the beginning, is that this use case goes well beyond bespoke manufacturing. It holds a great deal of promise in regulatory/legal fields. We have talked to clients about the opportunity in AML/KYC and ESG areas just to name a couple. In reality; Ask, Answer, Audit is the use case we should be looking for when trying to get the most from Generative AI in the short term.
Learn how the Expected Value Framework can transform decision-making by encouraging leaders to confront uncertainty, while fostering adaptability, resilience, and proactive risk management, empowering decision-makers to thrive in a dynamic and unpredictable world
Building a robust data strategy demands balancing technology's promise with security. Core technologies, SMEs, security checks, and maintenance considerations create a well-rounded approach for organizations.
Discover the power of the "Ok, Ok, Wait, What?!" approach to uncovering insights in your business and your data.
