An Open GenAI Business Case Analysis Best Practice: A Methodology for Data Program Transformation

1.4 Integrating the Method Facilitation with AI

Traditional business case analysis methods often involve a linear, waterfall-style process with a team of analysts, consultants, and sponsors. To augment and accelerate this practice, we recommend integrating the methodology itself into a Large Language Model (LLM) or AI Chat Assistant. Our experience in developing the NSDI Modernization Blueprint demonstrated that this approach can:

  • Facilitate the rapid ingestion, digestion, and synthesis of a large-scale desk audit (in our case, nearly 150 documents).
  • Speed up the identification of recurring themes, trends, and patterns across decades of policy and technical documents.
  • Act as a Socratic partner, facilitating gap analysis by asking clarifying questions based on the method’s required outputs.
  • Rapidly draft initial versions of the blueprint and its associated work products and templates.

This AI-assisted facilitation can speed up the desk audit and synthesis process by an estimated 10-20x, allowing human consultants and subject matter experts to focus their valuable time on high-level analysis, strategic refinement, and stakeholder engagement. To achieve this accelerated process, we encourage using an AI assistant of your choice with a structured, multi-prompt approach.

A Proven Prompting Strategy

Prompt 1: Initial Setup and Context Loading First, provide the AI with the complete methodology document and a clear set of instructions for the engagement. This initial prompt is highly effective. Clearly stating the roles, the step-by-step process, and the “hold response” instruction prevented the AI from jumping ahead and allowed for a thorough and patient desk audit. Staying within a single chat thread was also critical for maintaining the deep context established over hundreds of interactions. If possible, you can build your own LLM, RAG, MCP, and Vector Database for future use and sharing.

Please ingest, digest, and apply the attached methodology to generate a report on a specific topic. I will guide you through a phased process. First, I will provide the general problem objective. Next, I will provide numerous documents for a background desk audit. Please hold your response and digest all materials until I give the signal “OK, lets go.” After the signal, you will act as the facilitator, asking clarifying questions based on the method and proposing options to address gaps or risks you identify. Our final goal is to generate a starter report in a collaborative canvas.

Prompt 2: Defining the Problem Next, provide the initial problem statement, scope, and raw needs. Providing the documents in categorized batches (e.g., “Core Strategy,” “Federal Policies,” “Technical Concepts”) was extremely helpful. It allowed the AI to build a more structured understanding of the problem space and formulate more precise questions later in the process.

The attached Purpose, Objectives, scope, raw needs, background, and ideas are the initial input for the problem statement. I will now provide documents in categorized batches for the desk audit to help you organize the research and build out the best blueprint per the method. What we need to do is run the method, use these documents, and draft an initial report focused on the needed components, architecture, services, governance, investment guidance, etc. This input above will help, based on the method, to generate a problem statement that hits the above purpose and objective points and more. Finally, as we do the desk audit, please keep in mind we’ll need bibliography references – so keep that in mind when we do the report.

Prompt 3: The Desk Audit (Iterative) Upload the desk audit documents in batches with clear category tags. Be aware of your chosen AI’s context window and file size limits.

Prompt 4: Initiating the Analysis Once all documents are loaded, give the clear signal to begin.

That was the last set of documents. OK, lets go. Thank you for your patience and diligence. This should provide a rich and detailed foundation for our analysis.

At this point, the AI will begin to facilitate the analysis, asking questions phase by phase, as we demonstrated in the NSDI Blueprint collaboration. Provide answers and include additional attachments to support responses or pointers to attachments. If you need additional documents, you can do such and treat it like Prompt 3. 

Prompt 5: The Refinement Phase After the AI generates the initial blueprint, download the document and create a checklist of desired refinements. You will likely need to modify the template (e.g. headers, table of contents, other moves) and then get familiar with the outputs. Review it and make a task list by phase for the changes you may want to make such as the following examples:

  • Phase 1 – Stakeholders missing, changes; Mapping of Stakeholders to Use Cases
  • Phase 2 – Validate data source lists, add tables
  • Phase 3 –  Further background context on as-is state, history of changes, etc.
  • Phase 4 – Additional target components, mapping Use Cases to Components, specific component expansion or technology references
  • Phase 5 – Add implementation plan, add TCO/ROI Business case section, improved next steps and recommendations
  • General – Ask the AI to review or emphasize specific document more; clean up citations; create bibliography/references summary; expand or shorten sections; revisit executive summary with target emphasis; improved risk section by the template

Then, re-engage the AI for iterative improvements. This iterative refinement process is far more effective than trying to perfect the document in a single pass. It allows the human expert to guide the strategic direction, correct nuances, and add subject matter expertise, while the AI handles the rapid redrafting and integration of information from the extensive source materials. This collaborative loop proved to be the most valuable aspect of the entire process.

We will now begin the document refinement phase. I will provide a series of targeted prompts for specific sections. I will take your output and integrate the changes into a local copy of the document. Please leverage your current work and all the desk audit inputs to assist with these refinements.

Track your checklist, iterate and integrate improvements while editing the document offline (editing in canvas tends to be slower).