5 strategies to improve the output quality of Corporate GPTs
Garbage in, garbage out is still a problem in the world of Generative AI, what can you do?
Chat GPT has drastically changed the way we interact, search, and create information. This disruption is reflected in CEO agendas right away, with many adding the adoption of artificial intelligence (AI) to the top of their business priorities. In fact, more than 60% of CEOs across the globe consider AI more important than the internet and fear that not using it can lead to significant competitive disadvantages.[1]
In parallel, more and more employees are testing and using Chat GPT, seeking help and guidance on how to perform various tasks. Unfortunately, only a few companies have successfully implemented Generative AI (Gen AI) applications, aka Corporate/Enterprise GPTs, while others are still struggling with traditional knowledge management issues (e.g. multiple repositories, duplicate and outdated documents, and lack of version control). As a result, open systems like Chat-GPT remain the go-to resource for employees, creating significant risks of confidentiality breaches for companies.
So how can companies set up successful Corporate/Enterprise GPTs and minimize the risks? Adopting data quality standards as a framework can give companies a competitive edge in the Gen AI race. Ultimately, the quality of the data is the foundation of a reliable and high-quality AI application.
Here are some of the crucial dimensions to focus on:
Accuracy: Many companies use platforms such as Box, SharePoint, or Notion as primary sources for their Gen AI applications. However, these platforms are often used for daily work, which includes creating draft documents that are not yet fact-checked, aligned, or finalized. Creating content or information that will be shared company-wide requires multiple revisions and adaptations. Unfortunately, AI is not capable of differentiating between a draft and a final version. When all the versions are kept in the same place, it consolidates them, damaging accuracy.
Validity: Things change quickly in the business world. New trends and regulations are frequently introduced. Going back to archives to reflect all the new changes in old documents is impossible. But if old records are kept in the repository for AI initiatives, your applications might give guidance and suggestions to your teams based on conditions that have altered.
Consistency: We all faced a situation where we came across multiple versions of the same document and needed to ask our colleagues to identify the correct one. When information stored in different locations doesn’t match, AI can generate contradictory results. Moreover, due to its black box nature, it becomes impossible to trace the origin of the information and verify it with the relevant owners.
Completeness: If different teams keep different repositories and rely on disparate systems, without a single source of truth, your Corporate GPT would be training on several incomplete elements and forming answers missing certain insights.
What can you do ?
Here are 5 strategies that can help you improve output quality of corporate GPTs:
1. Leverage a framework that mitigates hallucinations:
While Gen AI usually provides impressive answers to user queries, it isn’t uncommon for it to generate incorrect answers and present them as facts. To mitigate this issue, you can use a framework like RAG (Retrieval-Augmented Generation). This method grounds Large Language Models (LLMs) to a knowledge base, which reduces the chances of providing misleading information and also prevents information leaks.
2. Separate past and future-facing documents:
To create a refined and reliable base for your GenAI efforts, separate documents according to their time orientation and purpose and exclude daily working files to avoid inconsistencies. If you want your Corporate GPT to provide employees with:
- guidance on how to do things: consider creating a base from future-facing documents that offer specific instructions through how-tos, guidelines, SOPs, or best practices.
- information about how things were done in the past: Create a base from finalized records detailing past efforts.
3. Use back-office functionality:
This means that drafts are created and edited in a separate area, and only the final and aligned documents are available on the front end. This way multiple stakeholders can collaborate and provide input on the documents, without the risk of errors or inconsistencies being visible to your Corporate GPT or the end user. Overall, leveraging back-office functionality for document creation and collaboration can help you streamline your processes, reduce errors, and ensure the distribution of the highest-quality documents.
4. Assign owners:
To ensure transparency around the responses of Corporate GPTs, establish a clear information base that specifies the owners of each piece of information. This will help you overcome the challenges presented by the black-box nature of AI technology, allowing you to track back to the source of information provided and creating a more accountable system.
5. Breakdown silos:
When departments work in isolation from one another and use different formats, methods, and software to create and organize information, it becomes harder and harder to create a complete and overarching view. One way to overcome this is by implementing a platform that facilitates teamwork and unifies different departments toward a shared goal. This approach ensures efficient collaboration between different organizational units while maintaining a single source of truth for your Corporate GPT.
In essence, the successful implementation of Gen AI remains a challenge for many companies, as they grapple with traditional knowledge management systems and data accuracy issues. Prioritizing data quality and adopting proactive strategies for managing knowledge will be instrumental in realizing the transformative power of ChatGPT while minimizing associated risks.
Collect and declutter scattered and impractical content<span data-metadata="">
We work with each organizational unit, taking inventory of all guiding documents and create an overview to better manage your data.
- Lean method to asses current state
- Every guiding document reviewed
- 30-50% of guiding documents removed through a high-level quality check
Provide a dedicated space to collaborate and drive change
Once we have decluttered your content, we migrate it to qibri, an easy to use platform where teams can collaborate and align on guidelines and best practices.
- A single space for each guideline/best practice for teams to work towards a common goal
- Clear accountibilty with a single owner per space
- Decentralized and self-organized
