Governed product categorisation

Turn messy product feeds into category outputs teams can actually run.

AiCategoryEngine combines workflow, review, and run history in one operating path so catalogues move faster without becoming harder to trust.

Faster classification Move large catalogues through one repeatable path instead of fragmented manual steps.
Governed workflow Keep products, rules, review, run, and history connected in the same operating model.
Traceable decisions Make outputs easier to inspect, approve, and explain when edge cases appear.
Platform flow

One path from raw catalogue data to controlled output.

Instead of stitching together spreadsheets, prompts, and ad hoc review, teams work through one visible sequence.

1
Upload products and logic Bring catalogue data and workbook rules into a structured run setup.
2
Review weak evidence before run Catch ambiguity early instead of discovering problems after output is delivered.
3
Run, inspect, and keep history Preserve runtime, output, and lineage so each run is easier to assess and repeat.
Measured Runtime and token usage are captured as part of the run record.
Reviewable Edge cases can be held for manual judgement instead of hidden behind automation.
Operational The product is shaped around repeated usage, not one-off output generation.
Public operator tools

Matching, dedupe, and branding workflows available without sign-in.

These tools sit outside the authenticated categorisation workflow so operators can compile matching results, build dedupe workbooks, audit completed matching, or work through brand fixes quickly from the home page.

User output

Matching Results Compiler

Create the user-facing matching workbook from one or more auto-matching result files.

  • Compiles retailer sheets into one workbook
  • Preserves user review traffic-light tabs
  • Supports optional ALR context
Open compiler
Post-matching review

ALR Matching Auditor

Compare matching results with account-level reports to see what was matched, skipped, deleted, or left unresolved.

  • Uses before and optional after ALR workbooks
  • Surfaces exacts, skipped reverses, and not-found outcomes
  • Outputs a downloadable audit workbook
Open auditor
Account dedupe

Dedupe Workbook Builder

Compare incoming products with an account-level report and create the upload-ready workbook.

  • Adds retailer matches to existing products
  • Groups true new products into one row each
  • Shows what is already in account and what URL it matched
Open dedupe builder
Admin branding

Brand Helper

Open the focused brand-review surface used with the Admin Toolkit helper extension.

  • Uses the AiCategoryEngine page format
  • Pairs with the bundled browser extension
  • Keeps brand work separate from workbook tools
Open branding tool
How it works

A workflow that shows what is happening, what needs review, and what has already run.

The product is strongest when the process stays legible. Each stage has one job, and each handoff is visible.

01

Prepare the run

Upload the product file and workbook logic in one place, with the fields and mappings needed to start cleanly.

  • Products file preview before run
  • Logic file preview before save
  • Clear run setup instead of hidden defaults
02

Review what looks weak

Thin evidence and ambiguous rules surface before output is committed, so operators are not debugging blind results.

  • Review-gated rows stay visible
  • Manual edits are part of the path
  • Confidence is used to guide attention
03

Run with history attached

Completed runs retain timing, usage, output, and lineage so the platform can be assessed as an operating system, not just a tool.

  • Runtime is recorded per run
  • Output stays linked to the run record
  • History supports repeatable delivery
Live proof

A compact view of how the platform is being used.

The page shows enough live operating data to prove the system is active and measured without forcing visitors to parse a full dashboard.

Completed runs -- Recorded in the current summary window.
Rows processed -- Total rows handled across completed runs.
Average throughput -- Observed rows per minute.
Total tokens -- Inference plus translation usage.
Why it wins

The difference is not just faster output. It is a cleaner operating model.

The strongest contrast is between hidden automation and visible workflow. Teams can move faster when they do not have to guess how outputs were produced.

What teams are usually stuck with

Manual files, scattered logic, and black-box classification make it hard to know what changed, what failed, or what should be reviewed.

  • Too much work happens outside the product
  • Review starts after outputs are already produced
  • Run history is incomplete or too hard to read
  • Cost and quality are discussed after the fact

What this platform changes

The workflow is visible from setup to output, with review, runtime, and lineage attached to the run instead of reconstructed later.

  • One path for product file, logic, review, run, and history
  • Human judgement stays in the loop where evidence is weak
  • Each run is easier to inspect, compare, and repeat
  • Performance and usage are measured as part of delivery
Controls

Control is part of the product, not a promise layered on top.

The platform is designed so teams can understand who has access, what was reviewed, and which run produced the final output.

Access and entry

Authentication sits in the real product flow rather than as an afterthought around it.

  • Protected sign-in path
  • MFA support where enforced
  • Clear separation between public site and product access

Review before output

Weak evidence can be surfaced and handled before a run is treated as finished.

  • Reviewable rows stay visible
  • Manual judgement is part of normal operation
  • Confidence is used to direct effort, not hide uncertainty

Run lineage

Each run keeps useful timing and output detail attached, making repeatability and assessment easier.

  • Completed runs are preserved in history
  • Output stays linked to the run record
  • Measured usage remains visible after completion

See it against a real catalogue and workflow.

The fastest way to assess the platform is with your own product structure, your own logic, and the same review path your team would actually use.