Introduction

Bringing large language models (LLMs) into a business is never a one-size-fits-all process. Each customer, each workflow, and each goal is unique. To make sure the solution fits, I always start by asking questions to be answered first. Here’s a look at the kinds of questions I raise with customers at the very start of any LLM integration project.

My goal is to share my process so you can learn from it and see what works in real projects. If you’re considering using LLMs in your business, this is a great place to start. I’m also inviting you to reach out if you’d like to discuss your own needs or want guidance from an experienced consultant. In fact, I’m thinking of making this the first in a series of posts, each one walking you through a different step of my consultancy process. So you can follow along and apply these ideas to your own work.

Key questions:

Data control and integration

  • How do you want your data to interact with the LLM? (For example: Should the LLM access your data in real-time, or do you want to keep your data separate and private?)
  • Do you need to keep full control over your data, or are you open to some integration?
  • What types of data will the LLM access (structured, unstructured, sensitive)?
  • Where is your data stored, and are there any compliance requirements?
  • How often does your data change, and how should updates be handled?
  • Who should have access to the data through the LLM?

Business model and output

  • Should the LLM’s responses reflect your business model? (For instance: Do you want to highlight sponsored products or services in the output?)
  • How would you like to influence the ranking or prioritization of certain results?
  • Are there specific products, services, or messages you want to highlight or avoid?
  • Do you have compliance or regulatory requirements for public-facing content?
  • Should the LLM generate new content or only use existing templates?
  • How will you measure if the LLM’s output aligns with your business objectives?

Handling user input

  • How should the system respond to vague or incomplete user questions? (Example: “I’m looking for something quiet in nature.”)
  • What’s most important for your users—clarity, speed, or personalization?
  • Should the LLM ask follow-up questions to clarify user intent?
  • Are there specific user segments with different needs or behaviors?
  • What languages or dialects should the LLM support?
  • How should the system handle inappropriate or unexpected input?

Human oversight

  • Do you want a human to review or approve certain actions or outputs from the LLM?
  • Where should the “human in the loop” be placed in your workflow?
  • What criteria determine when human review is required?
  • How will feedback from human reviewers be incorporated into the system?
  • Who is responsible for monitoring and managing LLM outputs?
  • What is the escalation process for issues or errors?

Model selection and flexibility

  • Do you have preferences for a specific LLM provider or model (like OpenAI, Claude, Llama, Mistral)?
  • How important is it for your solution to be flexible or model-agnostic?
  • Are there budget constraints that affect model choice?
  • What are your requirements for latency and response time?
  • Should the system support future model upgrades or changes?
  • Are there open-source or on-premise requirements?

Trust, reliability, and brand integrity

  • How should we handle mistakes or “hallucinations” from the LLM?
  • What level of control do you need over the language and tone of the AI’s responses?
  • Are there topics or phrases that must be avoided?
  • How will you monitor and report errors or inappropriate outputs?
  • What is your process for updating or correcting outputs?
  • How do you want to handle user complaints or disputes?

Tool and feature integration

  • What external tools or features would improve your user experience? (For example: Route planners, booking systems, or calendars.)
  • How should these tools interact with the LLM?
  • Are there existing systems (CRM, booking, analytics) that need to be integrated?
  • What are your authentication and authorization requirements?
  • How should errors from external tools be handled?
  • What is the plan for maintaining and updating integrations?

Personalization and proactive suggestions

  • Should the AI make proactive suggestions based on user behavior or preferences?
  • How much context or memory should the AI retain across user sessions?
  • What user data can the LLM use for personalization?
  • Are there privacy or consent requirements for personalization?
  • How should the system balance personalization with user privacy?
  • How will you evaluate the effectiveness of proactive suggestions?

Real-time data and commercial opportunities

  • Do you need the AI to work with real-time data, like availability or seasonal offers?
  • Should the system support promotions, events, or time-sensitive deals?
  • How will the LLM access and update real-time data?
  • What are the sources and reliability of real-time data?
  • Are there business rules for when and how to surface commercial opportunities?
  • How should expired or outdated offers be handled?

Measuring impact

  • What metrics or analytics do you want to track to measure the AI’s effectiveness?
  • How will you define success for this project?
  • How often should performance be reviewed?
  • What tools or dashboards will you use for analytics?
  • Who is responsible for monitoring and reporting results?
  • How will feedback be collected and acted upon?

Vendor lock-in and costs

  • Are you concerned about being locked into a specific vendor, platform, or tool?
  • How important is it to manage costs, for example, by limiting LLM usage or using smaller models?
  • What is your expected usage volume and budget?
  • Are there exit strategies in case you need to switch vendors?
  • How will you track and control ongoing costs?
  • Are there licensing or subscription terms to consider?

Why these questions matter

Starting with these questions helps me understand my client needs, risks, and opportunities. It also ensures that the LLM solution they want to build is practical, secure, and aligned with their business goals. By focusing on the answers, we can avoid common pitfalls and create a system that truly works.

Curious about how these questions would apply to your business? Let’s connect and walk through them together.