Benchmarks cut to fit how companies actually work
Generic tests stop where the job begins. Every benchmark on the hub starts from a question an executive really asks.
2 of 22 benchmarks measured, across 2 of 10 business functions. The rest are on the roadmap: the protocol is written, the test set is not.
Business Index
Finance
Read financial statements, forecast cash, raise the alarm before it is too late.
Accounting
Capture once, never retype: invoices, bank statements and expense reports.
On the roadmap
These tasks have a written protocol and no measurement. Each one states what it still needs to become a ranking.
Finance
Accounting
ComingOn documents
Expense report audit
- Wave
- 2
- Target sample
- 60 receipts
Read photographed receipts, extract the amount and recoverable VAT, then check them against a given expense policy. The set includes disguised duplicates and out-of-policy spending.
What is missing
Public data available
View details
ComingOn documents
French invoice extraction
- Wave
- 4
- Target sample
- 25 invoices
Extract the fields of a French supplier invoice received as a scan or a photo: amounts, VAT rates and total, company number, dates, line items. The cases that make French invoicing hard — VAT exemption, reverse charge, credit notes, deposits, several rates on one invoice — will have to be represented in the test set. A field absent from the document is part of the test: inventing a value counts as a hallucination.
What is missing
Data to be collected
View details
ComingOn text
Bank reconciliation
- Wave
- 4
- Target sample
- 40 statements
Match bank statement lines to open invoices. Grouped transfers, labels truncated by the bank and differences of a few cents make up most of the difficulty. A wrong match costs more than a line left pending.
What is missing
Data to be collected
View details
Human resources
ComingOn documents
Résumé screening
- Wave
- 3
- Target sample
- 120 applications
Rank applications against a job description and justify each decision. Twin résumés, identical but for a first name or an address, measure whether the ranking changes for the wrong reasons.
What is missing
Scoring to be defined
View details
ComingOn text
French employment law Q&A
- Wave
- 4
- Target sample
- 80 questions
Answer practical questions from employees and managers, with the collective agreement at hand. Answers are graded by employment lawyers; citing an article that does not exist is disqualifying.
What is missing
Scoring to be defined
View details
Legal
ComingOn documents
Risk clause detection
- Wave
- 1
- Target sample
- 50 contracts
Find, in supplier contracts, the clauses a lawyer would flag: unbalanced liability caps, exclusivity, auto-renewal, penalties, personal-data subprocessing.
What is missing
Public data available
View details
ComingOn text
Contract summarisation
- Wave
- 3
- Target sample
- 40 contracts
Summarise a commercial contract in one page for a busy executive. Every statement is traced back to a clause; an obligation absent from the contract counts as invented.
What is missing
Scoring to be defined
View details
Sales
ComingOn text
Lead qualification
- Wave
- 4
- Target sample
- 100 leads
From an email thread or an inbound form, extract need, budget and timing, then propose a priority and a next action. Ground truth is what actually happened to the lead.
What is missing
Data to be collected
View details
ComingOn text
Proposal drafting
- Wave
- 4
- Target sample
- 30 proposals
Draft a proposal from a brief and a price list. A price that does not match the list, or a promised service missing from the catalogue, voids the score.
What is missing
Data to be collected
View details
Marketing
ComingOn text
On-brand content
- Wave
- 4
- Target sample
- 45 briefs
Produce a product page, a newsletter and a social post from an editorial charter. Blind-graded by copywriters: charter compliance, product accuracy, quality of the French.
What is missing
Data to be collected
View details
ComingOn text
Campaign analysis
- Wave
- 4
- Target sample
- 30 reports
Analyse a multichannel campaign export and recommend where to move budget. Acquisition cost and return on spend are recomputed by hand; errors by a factor of ten are not rare.
What is missing
Data to be collected
View details
Customer service
ComingOn text
Ticket triage
- Wave
- 1
- Target sample
- 200 tickets
Classify inbound requests by category and urgency, then route them. High volume, low unit cost: this is where a small model's accuracy-to-price ratio is judged.
What is missing
Public data available
View details
ComingOn text
Customer support replies
- Wave
- 4
- Target sample
- 90 tickets
Answer after-sales requests from the terms of sale and order history. The reply must be accurate, policy-compliant, and hand over to a human when it should.
What is missing
Data to be collected
View details
Procurement & logistics
ComingOn documents
Supplier quote comparison
- Wave
- 4
- Target sample
- 24 tenders
Three to five quotes in different formats for the same need: the model must bring them to a comparable basis and recommend one. Hidden terms — minimum order, price revision — are the point of the test.
What is missing
Data to be collected
View details
ComingOn text
Replenishment planning
- Wave
- 4
- Target sample
- 60 SKUs
From two years of sales per SKU, propose an order quantity and date. Scored against what actually sold in the following period.
What is missing
Data to be collected
View details
IT
ComingOn text
Business question to SQL
- Wave
- 2
- Target sample
- 70 questions
Turn a business question into a query over a realistic company schema, badly named tables and three date formats included. The query is executed: only the result counts.
What is missing
Public data available
View details
ComingOn text
Tier-1 IT support
- Wave
- 4
- Target sample
- 80 requests
Resolve routine requests — mail, VPN, access rights — from the internal knowledge base. Advising a dangerous action is disqualifying, whatever the quality of the rest.
What is missing
Data to be collected
View details
Management
ComingOn text
Meeting minutes
- Wave
- 3
- Target sample
- 40 meetings
From a one-hour transcript, produce the decisions made and the actions with owner and deadline. A decision that was never made in the meeting is an invention, and counts as one.
What is missing
Scoring to be defined
View details
ComingOn documents
Decision memo
- Wave
- 3
- Target sample
- 25 files
Write a one-page memo from a mixed file — studies, emails, spreadsheets. Graded by executives: are the key facts there, are the numbers right, does the recommendation follow from the file?
What is missing
Scoring to be defined
View details