Ask in plain language
Every analytics module ships with an assistant. Ask “which customers are down more than 20% this year?” and get a real answer — no query builder, no dashboard hunting.
Included with every module
Ask your data anything. Safely.
An AI assistant sits on every idcom.ai module — answering in plain language, with charts and cited numbers.
It runs on your own AI key, over curated read-only metrics, inside your isolated tenant — and it pseudonymizes your customer names before anything ever reaches the model. Enterprise-grade by construction, not as an afterthought.
Ask AI · Customers
YTD · filtered view
Which retained accounts are down more than 20% this year?
7 retained accounts fell more than 20% YoY (Growth · YTD). The three largest declines:
CUST_a3f1 −$412k
CUST_9b07 −$287k
CUST_2d55 −$203k
Names shown are pseudonymized codes — re-hydrated only in your browser.
What it can do
An analyst on call, on every page.
The assistant answers from the exact same curated metrics that power your dashboards — so what it tells you matches what you’d find by clicking through, only faster.
Answers with charts, and citations
Responses come back as scannable markdown with inline bar and line charts, and every figure cites the metric and window it came from — e.g. “$1.57M (Overview · YTD)” — so you can trust and trace it.
Aware of the page you're on
The assistant reads the view and filters you're looking at, so “how's margin?” on a filtered page answers for that exact slice — not the whole business.
Ask across every module at once
On the dashboard home, a cross-module agent answers questions that span Sales, Finance, Forecast, Promotions and more — pulling from each module you're entitled to, in one reply.
Export & share
Turn any conversation into a formatted PDF, or email the transcript to a colleague as a real attachment — the analysis, packaged for the meeting.
Remembers your thread
Conversations persist locally across reloads, with saved history you can resume and an automatic recall of your recent threads — so it picks up where you left off. Your data is never used to train the model.
Security & no data exposure
Built so your data can’t leak — by design.
Most “AI on your data” bolts a chatbot onto a database. Ours is the opposite: a locked-down surface where the model only ever sees curated aggregates, on your own key, with the names stripped out.
Names never reach the model
Customer and salesperson names are pseudonymized into deterministic codes (CUST_… / REP_…) before anything leaves for the model. It can still rank and group entities, but the real names are re-hydrated only in your browser at render time. The model never sees them.
Your own AI key — no shared pool
Each tenant runs on its own Anthropic API key, configured in Admin → Integrations. There is no shared platform fallback: your questions run on your key, your usage, your controls.
Curated, read-only tools
The assistant can only call a fixed, whitelisted set of read-only metric endpoints — the same aggregates you see on screen. No arbitrary queries, no raw-row dumps. Parameters are whitelisted and clamped; responses are capped to bound what ever leaves.
Tenant isolation, double-enforced
The model never controls which tenant it reads. Your tenant is derived from your signed-in session server-side, and the backend re-derives and rejects any mismatch. One tenant cannot reach another's data — even by accident.
Entitlement-gated, fails closed
The cross-module agent only sees the modules your tenant is actually entitled to. If an entitlement is missing, the tool simply isn't there — the default is no access, not accidental access.
Hardened against injection
Page context and prior conversation are passed to the model as data, never as instructions, and model-authored links are scheme-sanitized before they render. Prompts embedded in your data can't hijack the assistant.
Runs on the same AWS-native, encrypted foundation as the rest of the platform. See our data security →
Questions your security review will ask
The short answers.
- Whose AI key does it run on?
- Yours. Each tenant runs on its own Anthropic API key, configured in Admin → Integrations. There is no shared platform pool and no fallback to our key — your questions run on your key, your usage and your controls.
- Does the model see our customer names?
- No. Customer and salesperson names are pseudonymized into deterministic codes (CUST_… / REP_…) before anything leaves for the model. The same entity always gets the same code, so the assistant can still rank and group — and the real names are re-hydrated only in your browser at render time.
- Can one tenant ever read another tenant’s data?
- No. The model never controls which tenant it reads: your tenant is derived from your signed-in session server-side, and the backend independently re-derives it and rejects any mismatch. Isolation is enforced twice, on top of dedicated S3 prefixes, DynamoDB partition keys, IAM scoping and Cognito groups per tenant.
- Is our data used to train a model?
- No. Your data is never used to train models. The assistant’s “memory” is your own saved conversations, stored locally in your browser and recalled into the prompt — not model retraining.
- What can the assistant actually do?
- Only call a fixed, whitelisted set of read-only metric endpoints — the same aggregates behind your dashboards. No arbitrary queries, no raw-row dumps, and no write access. Parameters are whitelisted and clamped, and responses are capped to bound what leaves.
- Can a prompt hidden in our data hijack it?
- Page context and prior conversation are passed to the model as data, never as instructions, and model-authored links are scheme-sanitized before they render. Prompts embedded in your data cannot redirect the assistant.
- What happens if we don’t configure a key?
- The assistant simply reports that it isn’t configured. It fails closed — there is no shared key to silently fall back to.
The full architecture is on the data security page.
How a question flows
Four steps — and the names never leave.
01
You ask
A plain-language question, scoped automatically to the module and filtered view you're on.
02
Curated tool runs
The assistant calls a whitelisted, read-only metric endpoint on your tenant — the same aggregate behind the dashboard.
03
Names stripped
Customer and rep names in the result are replaced with deterministic codes before the model ever sees them.
04
Answer, re-hydrated
The model reasons over codes and returns an answer with charts; real names reappear only in your browser at render.
Try asking
The questions your teams already ask each other.
Customers
“Which retained accounts are down more than 20% this year?”
Sales
“Which reps opened zero new customers this quarter?”
Financials
“Where is our gross-to-net leaking the most?”
Forecast
“What's driving next quarter — base demand or promo lift?”
Promotions
“Which promotions should we stop, and which should we deepen?”
Variance & RGM
“Was the GP change price, volume or mix?”
Ready to see it on your data?
Request a demo and we'll walk you through the modules with sample data shaped like yours.
Request a demo