Feature Ask Any Business Question in Plain English

Ask anything. Get a
verified answer instantly.

Plain-English questions, reasoning shown, sources cited. When a query fails, Zenzo repairs and retries — automatically. When data lives across systems, it plans the join.

60s

From question to verified cross-source answer

Source federated join planner with predicate pushdown

~60%

LLM-call reduction via semantic + prompt cache

Auto

Self-correcting loop repairs failed SQL up to N times

How It Helps

Built for every question your team needs

Trusted, sourced answers in seconds — from real data, with reasoning shown before every query.

💬

Plain English queries

No SQL, no schema knowledge. Ask in the language your team uses.

⚡

Reasoning chain before execution

Zenzo shows what it understood and what sources it will check. Your team confirms. Then it runs.

🎯

Cross-source intelligence

CRM + billing + product DB + helpdesk in one query, one answer.

📋

Source attribution

Every answer shows which sources were queried, last sync time, and confidence.

🔄

Verified playbook memory

Confirmed queries become playbooks — pre-validated by your analysts.

🛡️

PII masked before AI sees data

Only masked schema metadata is sent to the LLM. Row data never leaves the server.

Reliability

Failed queries fix themselves

The Self-Correcting SQL Loop v2 catches engine errors, asks the LLM to repair, and retries — with PII-sanitised traces, in-attempt cost gating, and full per-attempt telemetry.

🩹

Auto-repair on dialect drift

Typo, wrong column, missing quote? Zenzo repairs and retries — up to N times — without bothering you.

🔐

PII-safe error traces

Engine errors are sanitised (quoted literals, tuples, big numbers stripped) before persistence. Sensitive WHERE values never land in logs.

💰

In-attempt cost gate

Every repair LLM call is cost-estimated up-front. Runaway prompts cannot overshoot your per-day cap.

📊

Rescue-rate KPI

7-day rolling KPIs — attempts, rescued, rescue rate %, cost-gated — visible in Settings.

🪪

Identity-bound traces

Every trace carries the caller's RuntimeIdentity for end-to-end provenance.

🔁

Stop or retry — your choice

Configurable max retries; manual override always available.

Federation

Join data across N sources — without an ETL

The Federated Query Planner replaces brute-force pandas merges with a sqlglot-based plan that pushes WHERE filters down to each source, transpiles per-dialect, and refuses to push predicates over masked columns.

🌐

N-source planner

Postgres ILIKE, MySQL backticks, Snowflake quoting — transpiled per dialect, executed in parallel.

⬇️

Predicate pushdown

Filters land at the source, not in memory. Less data over the wire, faster answers.

🔒

Privacy-gated pushdown

Predicates touching masked columns stay on the firewall side — your masking policy cannot be bypassed by a join.

📈

Cardinality-aware

LRU-cached COUNT(*) estimates (TTL 5 min) drive smaller-side-first joins.

🛟

Structured fallback

7-day fallback-ratio counter; if the planner is uncertain, it falls back to pandas with a visible badge.

🔍

Plan visibility

Every answer carries a "Query plan" disclosure — what was pushed, what was joined, what was masked.

Memory

"Filter that customer" — works tomorrow too

The Conversational Memory Graph remembers the entities the copilot resolved for you across sessions. Pronouns like that, those, them stay anchored — and GDPR-safe forget purges every trace in one call.

🧠

Cross-session entity recall

Customers, regions, products, orders — resolved once, remembered across days.

🪶

Confidence-aware

resolved / low_confidence / none — the UI surfaces a chip when the copilot isn't sure.

🗑️

GDPR-safe forget

Forget an entity → every chat reference is purged so the copilot cannot re-learn from history.

Live Examples

Real questions. Verified answers.

Every answer includes its reasoning chain, source attribution, and confidence level.

"Which enterprise accounts are showing churn signals this quarter?"

Salesforce CRMStripe BillingProduct DB

14 Enterprise accounts match all 3 signals. ARR at risk: $2.1M. Top flag: renewal 18 days, logins ↓68%.

"What is our NRR trend for the last 6 months?"

Stripe BillingPostgreSQLCRM

NRR: 118%. Enterprise: 128%. Growth: 108%. Net expansion from Q2 upsells (+$184K).

"Which features drive the fastest trial-to-paid conversion?"

Product DBCRMBilling

Teams activating API connector before day 3 convert at 3.8× baseline. AHA moment is earlier than assumed.

"Why did support ticket volume spike 34% this week?"

ZendeskITSMError Logs

218 tickets trace to a Tuesday deploy. 3 Enterprise accounts affected. One root cause, one fix.

"For the first time, our CFO stopped questioning the numbers in our weekly review. The reasoning chain gave her confidence to act on them immediately."

— Director of Revenue Operations, Series C SaaS