Your company’s back office, run by AI

Perimeter automates the routine work in finance, operations and sales: reports, reconciliations, follow-ups and alerts. You stay in charge: every number shows where it came from, so your team can focus on what matters.

Book a demo See how it works
Connect your agent Claude OpenAI Ozor
Sales by channel · Monday meeting 2.4 s
Build a sales-by-channel dashboard for Monday’s meeting and tell me what changed this week.
Reading business definitions
Querying ERP + e-commerce
Checking against the accounting close
SQL behind this answer select channel, sum(net) as sales from sales where week = 35 group by 1 order by 2 desc
What do you want to know about your data?

Sales by channel

Updated2 min ago
Data throughSep 14
Net sales$0M+13% vs week 34
Average ticket$0K+4% vs week 34
Stockouts0%+0.4 pts
Net sales by channel · last 8 weeks
W28W29W30W31W32W33W34W35

Trusted by leading teams

  • CCU
  • Carozzi
  • Claro
  • Salcobrand
  • VSPT
  • VTR
  • Biotec
  • Tika

Perimeter is AI built to run your company. It takes on the repetitive work, keeps an eye on the numbers and gives your team back the time to decide.

Explore the platform
SAP Salesforce Snowflake Google Sheets HubSpot Shopify PostgreSQL BigQuery Slack

See the whole business

Connect your sources, monitor what matters and dig into what changed and why. Every result traces back to the query that produced it.

Q4 marginif demand
drops 4%
Hold price—Recommended
Targeted promo—Recommended
Raise price 3%—Recommended

Decide the next move

Model scenarios, challenge assumptions and keep the tradeoffs in plain sight. The plan stays current as conditions change.

Every Monday 08:00
Queries sales and margin 0.8 s
Checks against the close 1.2 s
Drop >8%: asks for review waiting
Sends to the sales committee done

It doesn’t just plan. It executes

It runs workflows inside the controls you set, with visibility into every step. When human judgment is needed, it stops and asks you.

What changes when nobody has to wait for the analyst

Cost
-90%

Lower analysis cost from day one

Everyday exploratory questions move to agents you can verify, and the data team focuses on what actually needs human judgment.

Speed
4 min

From days of waiting to minutes for an answer

The question that used to go into the data team’s ticket queue now gets answered in the same meeting where it came up.

Traceability
100%

Of answers come with their SQL visible

Every figure comes with the query that produced it. If the number matters, anyone can check where it came from before deciding.

The Perimeter platform

Five layers that connect AI to the context and data it needs to do reliable operational work.

Secure, permissioned access to your systems, enforced read-only, and your business definitions encoded. Deploy in our cloud or yours: your data stays yours.

The rules, exceptions and definitions for every table live with the agent. It knows what “active customer” means to your team, which sales count and which suppliers are excluded.

The model, the tools and the reasoning approach are chosen per task. When a cheaper model answers better, we use it.

Before answering, the agent checks its figures against control sources and runs the evals your team wrote. If something doesn’t add up, it says so instead of making things up.

Roles, per-source permissions, per-user scope and a log of who asked what and when. Any connection can be revoked in one click without affecting the rest.

Connected sources
Snowflakeread-only
SAP · ERPread-only
Budget 20263 sheets
Salesforceread-only
Business definitions
Active customerbought in last 90 days
Net salesgross − returns
Own channelstores + e-commerce
Excludenon-compliant suppliers
rule learned from chat · Sep 12 “don’t count transfers between warehouses”
Routing by task
Automatic
Long reasoninghigh
Report writingmedium
Simple querylow · 14× cheaper
Checks before answering
Matches the accounting close
Team evals · 24/24
Data coverage1 store missing
note the Chillán store reported late. The figure is shown without it, with a warning.
Access and audit
Sales committeesales · margin
Store managerstheir store only
Supplierstheir category
Query log1,284 this week
01 — Context

Agents that know your business’s definitions

Agents with context

Load the rules, definitions and exceptions for every table, so the agent knows exactly what “active customer” means to your team.

Agent memory

Tell it something once in chat—like “exclude non-compliant suppliers”—and it’s saved and applied automatically from then on.

Validated with evals

Write the questions whose answers you already know. The agent goes into production when it passes them, not before.

What dashboard do you need?
Operations dashboard live
On-time deliveries92%
Stockouts2.3%
Coverage78%
Perimeter agent
All about context
02 — Dashboards

Dashboards built from one prompt, in view of the whole team

Real-time dashboards

Describe what you want to monitor. Charts are chosen, wired to live data and laid out on a dashboard, or a full data site with a shareable link.

Share and embed

The same dashboard lives where your team already works: a link, the intranet or an iframe in an internal tool. You decide which filters viewers control and whether the page requires a signed token before it loads.

Automatic insights

Every chart comes with a summary of what changed: which segment moved, by how much, and what it means if the trend holds.

Sales dashboard Key metrics and acquisition channels
Sales$0M+13% vs last week
Churn0%−6% vs last week
CAC$0−17% vs last week
Revenue by plan · monthly trend
MarAprMayJunJulAug
Channels
$83Mtotal
  • Direct sales
  • Distributors
  • E-commerce
  • Marketplaces
All about dashboards
03 — Automations

Monday’s report, without losing Monday

Scheduled reports

Describe the report once. It runs on your schedule and arrives already written to the right people.

Run history

Every run is recorded: when it fired, what each step returned and why, without digging through logs.

Human-judgment steps

Add an analysis step inside the workflow and define when it should stop for approval. If the number falls outside the range you set, it waits for a person before sending anything.

What do you want to automate?
Scheduled · Monday 08:00
Building dashboard with live data…
Sales report · week 35 live
Sent to the sales team
Workflows
All about automations

Built for every team that lives by the numbers

Wherever there’s a question nobody got around to answering, Perimeter goes to work. Here’s how each team’s day changes.

Sales

The forecast corrects itself when the month shifts, and tells you which account moved it.

See the use case

Finance

The close explained line by line, with the margin variance attributed before anyone opens Excel.

See the use case

Operations

Stockouts caught before the store reports them, with the SKU, the warehouse and the likely cause.

See the use case

Marketing

Where to put the next dollar, matching what was spent against the sales that actually happened.

See the use case

Category

What to restock, what to clear and when, using real sell-through by store instead of the national average.

See the use case
The real questions

What each team asks it on any given Tuesday

The pipeline stops being yesterday’s snapshot

The forecast lives in the same conversation where it gets discussed. Every figure comes with the account, the date and the movement behind it, so the committee debates the decision, not the data.

Which accounts explain the drop in the October forecast?
Show me customers whose ticket dropped two months in a row.
Which reps close faster when the discount exceeds 12%?
Tell me every Monday if any large account dropped more than 15%.
Monday 08:00The committee arrives with the forecast already explained.
By accountEvery variance traced to a specific movement.
No ticketsThe follow-up question gets answered on the spot.

The close is explained before anyone asks

The margin variance arrives attributed by product line, channel and customer, checked against the accounting close. When a source is missing, the agent flags it instead of filling the gap.

Why did gross margin fall 1.8 points this month?
Separate the price effect from the mix effect in the drop.
Which customers are over 60 days past due and still buying?
Compare actual spend against budget by cost center.
ReconciledEvery figure is checked against the accounting close.
AttributedPrice, mix and volume separated, not a lone total.
AuditableThe SQL for every line is one click away.

Stockouts show up before the complaint does

Perimeter matches inventory, sales and shipments at the SKU and warehouse level. When a pattern goes out of range, it alerts you with the likely cause and the history of every time the same thing happened.

Which SKUs will stock out in the next 10 days?
Why did on-time deliveries drop in the southern region?
Which suppliers missed deadlines more than twice this quarter?
Alert me if a distribution center’s coverage drops below 15 days.
By SKUAt the level where restocking decisions are made.
EarlyThe alert arrives before the store’s complaint.
With the causeNot just what happened: why, and when it happened before.

Spend decisions made with sales right next to them

Each platform’s spend is matched against the sales that actually happened in your own systems, not the number the platform reports about itself.

Which campaign brought in customers who bought again?
Compare CAC by channel using ERP sales, not the platform’s.
Where are we spending on audiences that already buy from us?
Build the weekly spend report and send it on Fridays.
Real salesReturn is calculated against your own systems.
Repeat buyersTells new customers apart from existing ones.
WeeklyThe report builds and sends itself.

Real sell-through, store by store

The national average hides the decision. Perimeter goes down to the store and SKU level to tell you what to restock, what to move between stores and what to clear before price decides for you.

Which products sell well in one region and poorly in another?
How much inventory has gone over 90 days without a sale?
Which store-to-store transfers would avoid clearance this season?
What happened the last time we launched this category in winter?
By storeThe decision is made where the sale happens.
Before markdownExcess stock is caught weeks ahead.
With historyCompares against similar past seasons.

Enterprise-grade security

A tool that lets hundreds of people look at the company’s data needs more governance, not less.

Read-only, always

Perimeter can only read. Any attempt to modify or delete information is blocked before it reaches your data.

Encrypted credentials

Connection keys travel encrypted, are isolated per account and are never shown or stored in the browser.

Your data stays yours

Your data is used only to answer your questions: it doesn’t train models and isn’t shared with anyone. On Enterprise plans, processing can stay in your cloud and your region.

Role-based access control

Roles, per-source permissions and a log of everything that was queried, who asked for it and when.

Per-user scope

Each person sees only the data meant for them, including when dashboards are shown inside your own application.

Secure integrations

Every connection has the minimum permissions it needs and can be revoked in one click, without affecting the rest of the team.

Connected to the systems where your numbers live

Connect everything,
ask anything

Databases, warehouses, spreadsheets and business apps, all queryable from a single conversation in plain language. No exporting, no copy-pasting by hand.

Ask about a connector
Databases, warehouses and data lakes
PostgreSQL
MySQL
SQL Server
BigQuery
Snowflake
Databricks
MongoDB
Redis
Redshift
Supabase
+20
Business apps
Salesforce
HubSpot
Shopify
Mercado Libre
Falabella
Cencosud
Walmart
Ripley
VTEX
WooCommerce
Bsale
SAP
Odoo
Zendesk
Jira
Stripe
Google Ads
Meta Ads
Sheets
Slack
Notion
Asana
+100
MCP connections
Claude
OpenAI
Ozor
+5

AI-powered BI, built for the gap left by traditional BI and generic AI

How Perimeter compares with traditional BI tools and generic AI assistants.

Feature Perimeter Traditional BI Generic AI assistants
Implementation timeDaysWeeks or monthsImmediate
Agent with business contextTrained on your rules— Not availableGeneric, no context
Natural-language answersAccurate and governed— Not availableFrequent hallucinations
DashboardsOne prompt, real timeBuilt by handStatic, one-off
Alerts and reportsIncluded, one promptManual setup or add-on— Not available
Governance (roles, audit)Built inEnterprise plan only— Not available
Data sourcesDatabases, apps and spreadsheetsDepends on connector licensesManual upload, no large data
VerifiabilitySQL visible in every answerTechnical users only— Not available

Questions we get asked

What data teams ask before trying it, answered straight.

What would you use it for?

Not to replace your dashboards. A dashboard answers a question someone planned for, and that’s still useful. Perimeter exists for the other questions: the ones that come up on a Tuesday, matter for an hour and never justified opening a ticket.

That’s where analysis backlogs come from, and where most teams give up and export to a spreadsheet. Perimeter doesn’t replace your reporting layer: it removes the queue in front of the questions that never warranted building one.

Does it replace my data team?

No. It takes the repetitive work off their plate: the same five queries they get asked for every week on Slack. Your team goes from answering questions to defining what the numbers mean, and those definitions are exactly what makes the agent answer well.

What if the answer is wrong?

You see it. Every answer shows the query that produced it, so the error is visible instead of hidden in a number. And you only fix it once: tell the agent what was wrong, it’s saved as a rule and applied from then on.

Do I need clean data before I start?

No. It connects to your data as it is, with no cleanup project first. What you do need is to write down your definitions—what an active customer is, which sales count—and that happens during implementation, in hours, not months.

How long does implementation take?

The first results come out the same day we connect your sources. What takes a bit longer is loading your business context and testing the agents with questions whose answers you already know, before opening it up to the rest of the team.

What if my question spans two different systems?

That’s the normal case, not the exception. Databases, spreadsheets and business apps are queried from the same conversation, so a question that mixes ERP sales with ad spend gets answered without anyone exporting anything.

How much does it cost when the whole team uses it?

The price doesn’t spike when you add people who only ask questions. That’s the difference from traditional BI, where every new license pushes people back to asking a coworker for screenshots.

What AI didn’t change

Bad data still produces bad answers, now faster and more confidently, which is worse. That’s why Perimeter’s agents load your definitions before answering, and why every answer shows its query.

Governance matters more, not less: a tool that lets hundreds of people look at the company’s data needs enforced read-only access, per-person permissions and a log of who asked what. Someone still owns what the numbers mean; what changes is that they write it down once, as rules the agent applies, instead of re-encoding them in every report.

How to evaluate this category, including us

Most trials fail because they test the wrong question. Ask any tool something that already has a report and it will look good, because the answer is a lookup.

Test the questions that never became a report, and hold us to the same checks as anyone else: can you see the query? What happens with the second question? How do you correct it? What does it cost as usage grows? Does it reach your data as it is?

Your data already has
the answers

Tell us what you want to answer and we’ll show you Perimeter on your own sources.

Perimeter is implemented together with your team: we connect your sources, load your business context and test the agents with questions whose answers you already know before putting them to work.