---
title: "Ask your company: natural language reports and alerts for leadership"
description: "Natural language reports for leadership: ask how you are tracking against target and get data from every team, the risks and a proposed plan, plus alerts."
url: https://addabra.com/en/blog/ask-your-company
lang: en
date: 2026-10-11
category: "Departments"
keyword: "natural language reports"
publisher: Addabra
alternate: https://addabra.com/blog/preguntale-a-tu-empresa
---

# Ask your company: natural language reports and alerts for leadership

> Natural language reports for leadership: ask how you are tracking against target and get data from every team, the risks and a proposed plan, plus alerts.

Sunday, 9 p.m. Richard, CEO of a medical equipment distributor, is preparing for Tuesday's board meeting. He wants to answer one question: "How are we tracking against the quarterly target?" To answer it, he has to request the sales report, the collections report, the inventory report and the operations report. Each one arrives in a different format, with a different cutoff date, and the collections report won't show up until Monday afternoon. Richard ends up building a deck with numbers that are two weeks old.

That's the problem well-built **natural language reports** solve: leadership can ask the company what it needs to know, in plain words, and get an answer with data from every department, the visible risks and a proposed plan. Not in a week. In minutes.

## Why information reaches leadership late

Almost no midsize company has a data shortage. It has too much data, scattered across systems that don't talk to each other.

- **Every department tells its own story.** Sales reports what closed; finance, what was invoiced; operations, what shipped. All three numbers are true and none of them match.
- **The report depends on one person.** Someone in each department exports, cleans, pastes and sends. If that person is on vacation, the report doesn't exist.
- **Dashboards answer what someone anticipated.** They work for the usual questions, but Sunday night's question is almost never on a dashboard.
- **Numbers arrive without interpretation.** Knowing you're at 72% of target doesn't tell you whether it's serious or what to do about it.

The result: leadership decides on stale information or on gut feel. Neither scales.

## What natural language reports for leadership are

The idea is simple: instead of requesting reports, you ask questions. "How are we tracking against the quarterly target?" "Which large accounts haven't ordered in 60 days?" "Why did shipping costs go up in the northern region?" The system understands the question, queries the sources, cross-references the information and answers.

Behind that simple experience is an agentic system with several parts:

1. **Read-only connectors** to your systems: CRM, ERP, accounting, help desk and critical spreadsheets.
2. **A company brain**, the business memory where the definitions live: what "closed deal" means at your company, each department's target, which accounts are strategic. We explain it in depth in [your company brain](https://addabra.com/en/blog/company-brain-business-memory).
3. **Department agents** that know how to read sales, finance or operations data, each with its own context.
4. **An analyst agent** that takes the question, splits up the work, gathers the answers, spots contradictions and writes the response.
5. **An evaluator** that checks every figure has a source and that the text doesn't claim more than the data supports.

That structure is what separates it from a chatbot plugged into a database: shared memory, business rules, verification and traceability.

## What a good answer looks like

A useful answer for leadership always has the same shape. Here's an illustrative example of what Richard would get:

> We're at 71% of the quarterly sales target with 64% of the quarter elapsed. Collections are at 58%: three large accounts with overdue invoices make up most of the gap. Inventory for the monitor line covers less than three weeks of demand. Main risk: if those three accounts don't pay before month-end, cash won't cover the scheduled purchase. Proposal: a call from leadership to the two largest accounts this week, and pull forward the monitor order with the backup supplier.

Notice the pieces:

| Element | What it answers |
|---|---|
| Status | Where we stand against target, by department |
| Cause | Why we're there, with concrete data |
| Risk | What could go wrong and when |
| Proposed plan | What the system would do, so a person can decide |
| Sources | Where each number comes from and its cutoff date |

The plan is a proposal, not an order. Leadership accepts it, changes it or discards it. If any action means touching something (emailing a client, moving an order), it goes through approval by the right person.

## Weekly reports and alerts: let the company speak first

Asking is powerful, but the real value is not having to ask. The same system can work in two more modes:

- **Automatic weekly report.** First thing every Monday, leadership receives the state of the quarter in the same structure: status, causes, risks and proposal. Same format every week, so comparisons are easy.
- **Threshold alerts.** Rules you define: "Tell me if a strategic account cuts its orders," "Tell me if overdue receivables exceed a certain amount," "Tell me if a critical delivery slips more than two days." The alert comes with context, not as a bare number.

Alerts should be few and relevant. If the system flags everything, soon nobody reads it.

## A framework to implement it without getting lost

1. **Write down the ten questions you ask most.** The real ones, the Sunday night ones. That's your initial scope.
2. **Define every term.** What counts as a sale, an active customer, an on-time delivery. If two departments define it differently, decide which one wins.
3. **Identify the official source for each data point.** One data point, one source. Everything else is reference.
4. **Decide who sees what.** Role-based permissions from day one.
5. **Start with reading, not acting.** First the system informs; automated actions come later, and with approval.
6. **Review quality every week.** Compare answers against the manual reports for the first few weeks. That's where poorly written definitions surface.

## Common mistakes

- **Connecting everything without defining anything.** A model with access to ten systems and no definitions produces convincing, wrong answers.
- **Accepting figures without sources.** If the answer doesn't say where a number came from, don't take it into a board meeting.
- **Not being able to audit.** You should be able to see what the system queried to produce an answer. That traceability is what we call [AI agent observability](https://addabra.com/en/blog/ai-agent-observability) on the blog.
- **Treating it as a dashboard project.** It's not another dashboard. It's a system that reasons over your data with your rules, and gets better the more you use it.

## What to do on Monday

Take a sheet of paper and write down the questions you asked yourself last month that you had to wait to get answered. Next to each one, note how many departments had to get involved and how long it took. That list is the best requirements document there is.

At Addabra we build this kind of system to fit your company: we connect your sources, work with your team to write the business definitions into the brain, design the department agents and alerts, and support it in production. It's a natural part of a [Company OS](https://addabra.com/en/blog/company-os): the layer where leadership talks to its company.

## Frequently asked questions

### Where does the data in the answers come from?

From the same sources your teams use today: CRM, ERP, accounting, support and spreadsheets, connected with read-only permissions. Every answer cites where each figure comes from so you can verify it.

### What happens if the agent is missing data or the numbers contradict each other?

It should say so. A well-designed system would rather answer "I don’t have collections data updated since Tuesday" than make up a number. Contradictions between departments are surfaced as findings, not hidden.

### Can everyone in the company ask the same questions?

No. Answers respect each role’s permissions: a regional manager sees their region and leadership sees the consolidated view. That rule is defined when the system is designed, and it is logged.

## About Addabra

Addabra is an AI-software-first software factory that designs and builds custom agentic systems for companies: one agent per process, a shared business memory, orchestration, human approval by role, auditing and cost control. Free diagnosis: https://addabra.com/en/blog/ask-your-company#contacto