// Finance & Strategy

AI built around how a finance team works.

For the people who have to close the month, explain the margin and answer the board. Built on the documents and legacy systems you already have, so the number you report is the number you can trace.

CaptureReconcileCloseReportForecast
// Built for
01Finance directors and controllers
02FP&A and commercial analysts
03CFOs of owner-managed groups
04Strategy and corporate development
// From live deployments

Real numbers from real operations.

Six figures
Of proven procurement saving, evidenced line by line
$126
Variance between a rebuilt dashboard and the platform total, on +$26M
4 in 10
Purchase orders arriving on or before their promised date
0
Arithmetic failures across 267 independent checks

Every figure traces to a working paper. Where a number is unproven, we say so.

// Your working cycle

Four phases, from the document to the board pack.

Finance in an industrial business is mostly the work of proving things. Here is where that work breaks down, and what changes when the proving happens once rather than every month.

01

Capture & Reconcile

Invoice · ticket · GRN · ledger

Most of the evidence finance needs arrives as paper or PDF, and somebody keys it. The systems it has to agree with were never designed to talk to each other, so the reconciliation is done by eye, on volume, once a month. What that method finds is the differences big enough to notice, which is not the same as the differences that matter.

// Automated reconciliation

Quarri reads the documents as they arrive and matches them to the ledger line by line, so a difference is found by the process rather than by somebody being thorough that week.

  • Invoices, tickets and GRNs read, not retyped
  • Purchase, receipt and invoice matched per line
  • Differences surfaced rather than absorbed
  • Every figure traceable back to its document
Document captureSystems reconciliationAudit trail
Stickered lumber stacked in a shed, the stock a finance ledger has to value
// Documents · structuredpaper in

Documents read off the page and matched to the ledger, so the audit trail is a by-product of the process rather than a separate job.

02

Close & Control

Accrue · reconcile · post · close

The close runs on a workbook that a small number of people understand, and the rules inside it live in cells rather than anywhere documented. It works, until the person who built it is away, or the business changes shape, or somebody asks why a number moved. Then the answer takes days to assemble and rests on the same workbook nobody can check.

// AI-powered document extraction

Quarri drafts the accrual from live data and holds the rules where your team can read and edit them, so the close stops depending on one workbook and one person.

  • Accruals drafted from live data each cycle
  • Rate tables and mappings held openly, not in cells
  • Unmatched items aged in view, never absorbed
  • Postings written back as a file finance can use
Accrual automationRules you can editClose controls
Debarked logs loaded on a trailer, the movements a month-end accrual has to capture
// Accrual against invoicegap $0.01

One release cycle reconciled at document level, surfacing an over-accrual credit that a monthly comparison would have netted away.

03

Margin & Mix

Product · customer · branch · mix

Group margin is a number everybody has and nobody can act on. Underneath it sit customers bought at one price and served at another, products that lost money quietly for a year, and internal transfers that make a branch look like a business. Reported monthly and in aggregate, a mix problem and a rate problem look identical.

// Margin intelligence

Quarri decomposes margin into customer, product and rate effects, so a compression you can fix is told apart from one you merely have to explain.

  • Margin split by customer, product, branch and rate
  • Internal transfers excluded from trading margin
  • Duplicate customer identities merged before reporting
  • Retention and cohort movement measured, not assumed
Margin by customerMargin by productMix decomposition
Aisles of pallet racking, the inventory a margin analysis has to account for
// Margin bridgemix, not rate

At one manufacturer, inter-branch transfers had turned a healthy lifetime margin into a reported loss until they were separated out.

04

Cash & Forecast

AR · AP · cash · forecast

Cash is the number the board asks about and the one assembled last. Receivables age in one system, commitments to suppliers in another, and the stock in between was bought with money you can no longer use. Each is knowable; together they are a week of work, which is why the forecast is usually a shape rather than a position.

// Dynamic forecasting

Quarri holds receivables, commitments and stock value in one place, so a cash position is read rather than rebuilt.

  • Receivables aged by customer and by limit
  • Credit exposure flagged against agreed terms
  • Cash tied up in stock, valued per SKU
  • Board-ready views exported, not rebuilt
Receivables ageingCommitted spendCash forecasting
A roadside deck of cut logs, working capital sitting on the ground
// Receivables against terms1 breach

One account was carrying open receivables at more than four times its agreed limit, visible only once exposure was read against terms rather than in total.

// Why it can do this at all

AI at the foundation. Not retrofitted.

Most software for this chain was built as a form. Quarri was built as a model of your operation.

A fixed platform
  • Their fields, your words mapped in
  • Exceptions become change requests
  • Paper is a keying job
  • Every run costs a model call
  • New questions join the queue
  • Reports you wait to receive
  • Scale by hiring around the gaps
Quarri
  • Your words, from day one
  • Exceptions become rules you edit
  • Paper read at volume, deterministically
  • 100x cheaper per execution
  • Questions asked in plain English
  • Dashboards and scripts you build yourself
  • Scale by pointing it wider
// Across the whole cycle

Ask it in plain English. Answer the board in minutes.

Because the ledger, the documents behind it and the operational data all sit in one governed layer, a question can cross the whole business instead of stopping at a system boundary. The questions worth asking are usually the ones nobody could previously afford to ask.

  • Any question across the cycle, in plain English
  • Numbers computed, so the answer matches the ledger
  • Board packs and dashboards your team builds itself
  • Alerts when a margin slips or an account breaches terms
Natural language queriesBoard-ready exportsAlerts on exception
// Margin by customerone question

Margin movement by account, computed from the same documents the audit trail rests on.

// In practice

A lumber and millwork manufacturer

// Before

Group margin was reported monthly from a workbook nobody outside finance could check. Purchasing ran across a long tail of suppliers with no like-for-like comparison, and the same customer appeared under several spellings.

// With Quarri

Quarri matched purchase, receipt and invoice line by line, tracing a unit-of-measure error that had inflated purchase spend for years and surfacing a six-figure procurement saving. Separating inter-branch transfers from trading activity turned a reported loss back into a healthy lifetime margin. The rebuilt reporting was checked against the platform total and agreed to $126 on over $26M, with zero arithmetic failures across 267 independent checks. Read the case study.

Coming soon

Agents that close and chase.

Everything above works out the answer and hands it to you. The next step is letting it act. An agent that drafts the accrual when the cycle opens, flags the items that will not tie before you go looking, and chases the invoice that has quietly gone past terms. You approve a decision instead of assembling one.

  • Accruals drafted the day the cycle opens
  • Items that will not tie flagged before you look
  • Every action proposed for approval, never taken quietly
// In designnot yet deployed

A design sketch rather than a screenshot. Not deployed today, and every action would come to you for approval.

See it on your own data.

Live in two weeks, on the systems you already run.