How it works
Ask your data. Act on the answer.
MyDataTalk turns a plain-English question into SQL, runs it against your database, and answers in words and charts. But an answer isn’t the finish line — it’s the start. Save it, build a dashboard around it, send it to your team, set it to watch itself and alert you when something moves, or fold it into a report that keeps itself up to date.
Point it at Postgres, MySQL, Snowflake, or a spreadsheet.
A plain-English question becomes SQL, an answer, and a chart.
Save answers as insights, ranked by impact with a next step.
Describe a dashboard once; every tile is planned and run for you.
Send it, or set an alert that watches the metric for you.
Turn insights into a report that keeps itself up to date.
Step 01 — Connect
Your data, where it already lives
Add a connection once and it’s ready to query. Credentials are encrypted at rest, and the schema is indexed so the model only ever sees the tables relevant to a question — not your whole database on every ask.
Step 02 — Ask
Plain English in, real answers out
Type a question the way you’d ask a colleague. MyDataTalk writes the SQL, runs it, and replies with a written summary and the right chart — no query language required.
- See the work. Every answer can show the exact SQL it ran — nothing is a black box.
- Multi-source. Ask across several connections at once; it routes to whichever answered.
- Context-aware. “Why did that spike?” knows what “that” was.
- Grounded. Add notes and documents to the Knowledge Hub once; every answer can pull from them.
Step 03 — Understand
The Insights Hub
Save any answer as an insight and it becomes something you act on, not just read. Each one gets an AI-written recommended action and an impact value the hub ranks by, then moves through three lanes as you work it.
Freeze accounts C7162498 & C3512977 pending review.
OpenInvestigate SKU-4521 returns driving the drop.
In progressFixed broken verification link; conversions recovered.
ResolvedStep 04 — Build
A dashboard in one shot
Describe what you want — “a fraud monitoring dashboard” — and MyDataTalk plans a handful of concrete questions, runs each one, and lays out the charts. Prefer to build it by hand instead? Pin any chat answer straight onto the board.
- One prompt, several tiles. It proposes the questions first, so you approve the plan before anything runs.
- Diagnoses, not just charts. A “what changed” question renders the baseline → current → delta breakdown, not a raw table.
- Kept fresh. Turn on auto-refresh and each tile re-runs its query on a schedule.
Step 05 — Act · Send
Get it in front of the right people
Send an insight straight to a destination — formatted per channel, with the chart embedded in email. Connect your own Gmail to send from your address, or paste a Slack/Teams webhook. Webhook URLs are encrypted and never shown again.
- Formatted, not raw. Bold and a bar chart in email; clean text in Slack and Teams.
- Send as you. OAuth Gmail so replies come back to your inbox.
Step 05 — Act · Alerts
Make a metric watch itself
Don’t wait to remember to check. An alert runs a metric query on a schedule and notifies a destination when something crosses a line — or when it simply behaves unusually. It fires once per crossing and re-arms when things return to normal, so the channel stays signal, not noise.
Fire when a value crosses a fixed line — fraud above $1M today.
No number to guess. Fire when a metric strays more than 3σ from its own recent history.
Compare like-for-like — same hour of day or same weekday — so a Monday spike isn't flagged for being Monday.
Send each firing immediately, or batch them into one hourly or daily summary per channel.
Step 06 — Tell the story
A report that updates itself
Pick a set of insights and MyDataTalk writes them up as a narrative — what’s happening, what changed and why, what it means. Turn on auto-refresh and it stays a living document: when a linked insight moves, the story notices and rewrites itself, picking up where the last version left off instead of starting over.
- Grounded in evidence. Every claim traces back to a saved insight's answer or diagnosis — and your Knowledge Hub, if it's relevant.
- Keeps continuity. A regeneration is fed its own previous narrative, so it reads as an update, not a rewrite.
- Full history. Every version is kept, so you can see how the story — and the underlying numbers — evolved.
False positives rose to 142 this quarter, a 42% increase driven largely by the EU region…
Reference
The vocabulary
- Connection
- A database or spreadsheet MyDataTalk can query. Credentials are encrypted; the schema is indexed for relevance.
- Insight
- A saved answer with a recommended action and an impact value, tracked Open → In progress → Resolved.
- Knowledge Hub
- Notes and documents you add once; MyDataTalk searches them for context when answering a question or writing a story.
- Dashboard
- A grid of chart tiles — planned and run in one shot from a prompt, or pinned by hand from chat.
- Destination
- Somewhere insights and alerts are sent: email, Gmail, Slack, or Teams.
- Alert
- A metric query on a schedule that notifies a destination on a threshold or an anomaly.
- Anomaly (σ)
- How far a value sits from its own recent average, in standard deviations. Above your sensitivity, it fires.
- Digest
- Batched delivery — one combined message per channel, hourly or daily, instead of many pings.
- Story
- A narrative built from a set of insights. With auto-refresh on, it's a living document that rewrites itself as those insights change.