// How it works

The timber data platform is a layer, not a migration.

Your ERPs, databases and spreadsheets stay exactly where they are. Quarri governs them into one layer and makes them usable by any leading model, without a migration project or a five-year platform bet.

// Keep

  • Your ERPs, mill systems and databases
  • The spreadsheets your team trusts
  • How your people already work

// Add

  • A governed layer over all of it
  • Safe read + write into your systems
  • AI, dashboards and automation on top
// Where Quarri sits

Under the model. Over your data.

General-purpose models sit on top and do the reasoning. Quarri is the layer beneath them that holds your data, your definitions and your workflows, and it reads and writes back into the systems you already run.

// Horizontal AI

General-purpose models reasoning

Open source Your own agents

Quarri an MCP plugin

Product layer what your team uses
Document extractionReconciliationForecastingSmart SchedulerMargin intelligenceBuild
Infrastructure layer deterministic execution
100+ agent toolsVersioned workflowsReusable across teams
Context layer
Semantic modelCompany glossaryMemory of work
Data layer
WarehouseAgentic modellingRole-based data accessRead + write
Connectors & ingestion
Daily syncAPIs & databasesScanned documentsExternal feeds
// Your sources
Paper & handwritten Legacy ERP Mill systems Spreadsheets Databases Scanned PDFs External data

Model your data once, then any workflow builds on top. One source of truth, flexible, fast to build and cheap to run.

// The process

Five steps, done with you, not sold to you.

Software that needs a data team is not much use to an operator who does not have one. So the connecting, modelling and encoding is done by us, inside the fee, and it is where most of the early value comes from. This is what happens, in order.

01 Week 1

Connect what you already run.

Legacy ERPs, mill systems, databases, spreadsheets, scanned PDFs and paper. Nothing moves and nothing is replaced: Quarri reads from the systems you have and writes back into them. Around half a day per source, absorbed in the SaaS fee.

  • Legacy ERP
  • Mill systems
  • Spreadsheets
  • Scanned PDFs
  • Paper & handwritten
  • External data
02 Week 1

Clean it and model it once.

Every source is cleaned, deduplicated and structured into one queryable layer. The rules your team carries in its head get written down: rate tables, department mappings, the exceptions everybody knows and nobody documented. Model the data once, and every workflow after it builds on the same layer.

  • One queryable layer
  • Rate tables
  • Mappings
  • Exceptions, documented
03 Weeks 1 and 2

Encode your rules, in your language.

Your definitions go into the model: species, grades, units, cost centres, the metrics you actually run on. The reconciliation logic goes in as rules rather than prompts, so the matching, converting and reconciling runs as encoded workflows. Replayable, auditable, and with no model call per document, which is why a number in front of your board is never guessed.

  • Semantic model
  • Company glossary
  • Rules, not prompts
  • Replayable
04 Week 2
First workstream live

Automate what carries risk.

The first workstream goes live in production, not in a sandbox. That is what "live in two weeks" means. We start with the manual processes where an error costs money: reconciliations, accruals, extraction from paper. Each one gets validation checks, so a mistake is caught by the process rather than by somebody noticing.

  • Reconciliations
  • Accruals
  • Extraction from paper
  • Validation checks
05 Month 2 onward

Hand it over, then keep building.

Your power users learn to build, not just to read. Standalone workbooks get retired as people query the data directly, and the team that runs the operation stops queuing behind whoever owns the spreadsheet.

From month two: development days each month for new automations, a monthly call to set priorities, and maintenance of everything already running. Each workflow built is reusable, so the second and third are faster than the first, and what we learn on one process carries into the next.

  • Development days
  • Monthly priorities
  • Maintenance included
  • Forecasting next

Your contribution to data organisation, cleaning and catching improvements in process flow is two thirds of the value.

Customer, forestry operation
// What every step leaves behind

Everything built is reusable, shareable and yours to edit.

A reconciliation built for the finance team's month end does not stay with the finance team. It is versioned, documented and available to operations the next morning, and anyone with access can open it, read exactly what it does and change it. That is the part that compounds.

Shared across teams

Built once in finance, reused in operations, no rebuild and no second version drifting out of sync.

Versioned, not fragile

Every change is tracked and every edit runs against a verification harness before it lands. If something breaks, roll it back.

Open, not a black box

Your people can read the logic, see which rule produced a number and change it themselves. No ticket, no vendor.

Outputs persist

Workflows saved, dashboards live, data captured and written back into your ERP rather than exported and forgotten.

// Why deterministic

Encoded logic on the data. AI on the reasoning.

Quarri is a deterministic layer beneath probabilistic AI. Every figure below traces to a live deployment.

0
LLM calls per document on extraction, fully replayable and auditable
~1 sec
To re-solve a full order book, within 1% to 1.6% of a brute-force optimum
<30 sec
Full reconciliation runtime, replacing a 4-minute previous build
125
ERP tables synced daily at a single mill deployment

See it on your own data.

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

Request a demo → Read the FAQ →