Every AI analytics tool claims to understand natural language. But ask one "show me premium stones" and watch what happens. Most tools silently guess what "premium" means — and frequently guess wrong. You get results, but for the wrong question. And you don't even know it.
We built something different. Today we're introducing Conversational Intelligence — a multi-layer AI system that understands your intent, asks when it's unsure, and guides you to insights you weren't even looking for.
The Problem: Silent Misinterpretation
Here's what typically happens when you ask an AI analytics tool an ambiguous question:
Showing stones where category = 'Premium'...
The tool guessed that "premium" means a category column value. It was wrong. "Premium" in the jewellery industry could mean:
- High clarity grades (IF, VVS1, VVS2)
- Ideal or Premium cut grade
- Price above $5,000 per carat
Three completely different queries. Three completely different answers. The tool picked one silently, got zero results, and left you wondering what went wrong.
How Lookato Handles It
With Conversational Intelligence, the same question gets a very different response:
I need some clarification before proceeding.
Lookato recognized "premium" as ambiguous — it appears in the business glossary with three possible interpretations. Instead of guessing, it asked. One click, and you get exactly the right query.
The Intelligence Pipeline
This isn't a simple keyword lookup. Conversational Intelligence runs a multi-layer pipeline on every query, powered by GPT-5.4:
Layer 1: Nano Classification (~200ms)
A lightweight GPT-5.4-nano model classifies your query in milliseconds. Simple questions ("total revenue last month") resolve instantly and skip to query planning. Complex or ambiguous questions escalate to the full resolver.
Layer 2: Intent Resolution
The full resolver analyzes your question against your business glossary, dataset schema, and conversation history. It produces ranked hypotheses with confidence scores:
Show a monthly sales trend for Great Gems over the most recent 12 months.
This coaching banner appears in real-time while you wait for results. Each assumption fades in one at a time, giving you something to read — and verify — while the system works.
Layer 3: Binding Constraints
The resolved intent flows to the query planner as binding constraints — not suggestions. When the resolver determines that "sales" means the lookato_lattice_sales dataset and "last 12 months" means voucher_date >= toStartOfMonth(now() - interval 12 month), the planner must use these exact terms. No re-interpretation. No drift.
Smart Follow-ups: The System Leads the Conversation
After every answer, Lookato suggests what to explore next — based on what the data actually shows, not generic prompts.
Great Gems Ltd generated $14.5M in sales across 12,641 stones. Gross margin is a solid 21.8%.
These aren't random suggestions. The system saw that Great Gems has strong revenue but variable margins, so it suggests investigating margin patterns. Click any chip and you're one step deeper into your data.
The Business Glossary: Your Language, Codified
Every company has its own vocabulary. "Premium" means something different to a jeweller than to a SaaS company. "Fast-moving" might mean high transaction count to one team and quick sell-through to another.
The Business Glossary is where you teach Lookato your language:
| Term | Ambiguous? | Interpretations |
|---|---|---|
| Premium | Yes | High clarity (IF, VVS), Ideal cut, Price > $5K/ct |
| Quality | Yes | Clarity grade, Cut quality, Color grade |
| Performance | Yes | Revenue trend, Sell-through rate, Price/ct trend |
| Fast-moving | Yes | >3 transactions/period, Sold within 7 days |
When a user asks about any of these terms, Lookato shows the clarification card with the exact options from the glossary. No guessing. No silent misinterpretation. Admins manage the glossary through the admin UI — no SQL required.
Shareable Reports with Context
Every answer can be exported as a beautifully formatted PDF. But here's what makes it useful: the PDF includes conversation context.
When you share a follow-up answer like "So what are we missing?", the PDF shows the preceding questions and answers that led to that point. Your stakeholder gets the full story — not a decontextualized data point.
The context shows all prior questions (for flow) and the last two answers in full (for substance). Older answers show "..." to keep the PDF concise.
Powered by GPT-5.4
Conversational Intelligence runs on OpenAI's latest GPT-5.4 model family:
- GPT-5.4 for query planning and multi-step analysis
- GPT-5.4-mini for intent resolution and narrative generation
- GPT-5.4-nano for ultra-fast query classification (~200ms)
Each model is chosen for its strength: nano for speed, mini for cost-effective reasoning, and the full model for complex analytical planning.
Try It
Conversational Intelligence is available now for all Lookato users. Ask an ambiguous question and see the clarification card in action. Check the coaching banner to see how the system interpreted your query. Click a smart follow-up to explore further.
Your data has answers. Now the system actually understands the questions.