Maximize opportunity while protecting your data

Maximize opportunity while protecting your data

Many managers experience challenges retaining knowledge and lessons from project to project. Most understand the importance of an accessible knowledge base however there is a lot of friction in collecting, storing and retrieving quality data.

Part 1 went over the underutilised qualitative data in project analytics and the potential use-cases for LLMs in project management.

Empowering PMs with the right knowledge improves their ability to manage risk, and sets up the project for success. The opportunity is amplified for large enterprises, like those with international and fragmented structures. LLMs provide a natural, accessible interface to consume large amounts of diverse data sources.

Protect your data

Does all this accessibility to sensitive data risk your information security? Not necessarily. There are now several enterprise-grade SOC 2 options for your stack (eg. vector store and LLM), designed for sensitive use-cases. Forward-thinking companies have been working on internalising these tools with security and privacy in mind:

  • McKinsey developed generative AI tool ‘Lilli’ using 100,000 documents from its deep experience and intellectual property. Link

  • PwC launched conversational agent ‘ChatPwC’ to empower employees with access to its knowledge systems. Link

  • Einstein GPT, the next generation of Salesforce’s Einstein AI system, is an agent for customer relationship management (CRM). Link

Since ChatGPT is trained on user responses a competitor may accidentally see your sensitive info. Newly-released ChatGPT Enterprise finally allows access to OpenAI models with secure and encrypted SOC 2 compliance.

To maximise data control and ownership, you will want a series of compliant systems that include vector storage, prompt+retrieval engines and observability. There are various options for each system that excel depending on your preference for cost, control and/or open-source.

You can also try fully turn-key solutions (PAC Project Assistant) which can be self-hosted on your own company servers for ultimate ownership and security.

Ready, Set, Grow

🔑 Grow your knowledge base

  • Your scanners should be overdrive digitizing physical document archives and decades of memos, technical specs, and proposals, etc. Any unique knowledge and experiences will differentiate your models and its peformance against competitors.

🤖 Build vector store infrastructure

  • Vector stores hold your information in a way accessible to the new AI models. They allow LLM’s to talk with your data through searches for semantic similarity, i.e. finding documents with similar content, people with similar experience, projects with similar risks etc.

🚧 Focus on privacy and data sources with high ROI

  • Identify your key P&L drivers and prioritize the high impact use-cases + data. Start with memos, technical specs, minutes, risk & issue logs. Although meeting recordings could be used for training, initial efforts are better focussed on larger, higher quality datasets like memos & specifications.