The problem with using LLMs for analysis isn’t that they’re bad. It’s that they’re not built for consistency — and that’s quietly undermining every insight your team produces.
Same data. Same prompt. Different answer. Every time. Here’s why that matters — and what to do about it.
You’ve added AI to your workflow. Your team is using ChatGPT, Copilot, or something similar to speed up analysis. You’re getting outputs faster than ever.
So why can’t you confidently stand behind them?
In this webinar, Sergio Gomes at Relative Insight runs a live demonstration that exposes one of the most overlooked risks in AI-assisted analytics: LLMs are non-deterministic. Feed the same dataset into the same prompt twice and you won’t get the same answer. Same tool. Same data. Different conclusions.
That’s not a bug. It’s how these models work. But for any team using AI to analyse customer feedback, it has serious consequences — for reproducibility, for stakeholder confidence, and for the credibility of your entire insights function.
Duration: 5 mins | Available Now | Ideal for: CX, insights and research leaders
What you’ll see in this Webinar
Sergio runs a live, side-by-side demonstration — two tabs of ChatGPT, the same dataset, the same prompt — and shows exactly how non-determinism plays out in practice.
1. The same prompt, two different outputs Two simultaneous ChatGPT runs on identical data return similar-but-not-the-same themes. Ad volume and placement becomes ad frequency and intrusiveness. App performance and technical reliability shifts slightly. Conclusions change at the margins — and the margins are where decisions get made.
2. What non-determinism actually costs you It’s not just an inconvenience. When results shift depending on who runs the analysis, or when they run it, you lose the ability to reproduce findings, track trends over time, or answer the question every stakeholder eventually asks: “How did you get to that?”
3. A deterministic alternative, demonstrated live Sergio then loads the same dataset twice into Relative Insight — one run from the morning, one from twenty minutes before recording — and shows how a defined, consistent analytical framework produces identical outputs both times. Same themes. Same metrics. Same conclusions. Every time.
Why this matters for your insights program
The appeal of LLMs for analysis is real. They’re fast, flexible, and easy to access. But for any team that needs to:
- Track customer sentiment week on week
- Show consistent signals to leadership or operations
- Defend their methodology when results are challenged
- Build a repeatable, auditable reporting program
…non-determinism is a structural problem that speed alone can’t solve.
If your team can’t recreate their results, they can’t build confidence in them. And if leadership can’t trust the outputs, the insights stop driving action — no matter how good the underlying analysis is.
What you’ll learn
- Why AI-generated analysis varies — even when the prompt and data are identical, and what that means for the reliability of your outputs
- What non-determinism looks like in practice — a live demonstration that makes the problem immediately visible and hard to ignore
- The difference between flexible and consistent AI — and why that distinction is critical for any team responsible for trackable, reportable insights
- How a defined analytical framework eliminates variability — and what it looks like when the same data always returns the same answer
- What an audit trail actually means for CX analytics — and why being able to show your working is the foundation of stakeholder trust
Who should watch this webinar?
This webinar is built for the person in the middle — the one who bridges customer data and This is for anyone whose insights need to hold up to scrutiny — from a colleague, a director, or a board:
- Heads and VPs of CX or Customer Insights whose teams are using Gen AI tools but producing outputs that vary run to run
- VoC program leads who need consistent, reproducible signals to track performance over time
- Insights and analytics managers responsible for the accuracy and defensibility of their team’s outputs
- Research leaders who’ve been handed AI as a productivity solution — and are now grappling with what it does to analytical rigour
If you’ve ever re-run an analysis and got a slightly different answer — and wondered which one to believe — this webinar is for you.
