Your First Study in 10 Minutes (Quickstart)

This is the path from an empty workspace to a study that has been benchmarked and is sitting in review — the shortest loop through the product, in the order the workflow wants it. It assumes you are signed in as an Analyst (the role that creates and runs studies); if you are a reviewer, the same loop is how you see the work land in your queue.

Before you start: what you need

• The benchmark dump — the comparable companies’ financials for the study period(s), as one Excel workbook (.xls/.xlsx/.xlsm, up to 50 MB), typically exported from the benchmarking source your firm subscribes to. Four columns have to come through: company name, revenue, cost and operating profit. Multi-year dumps are supported; the quantitative screen needs at least two years of data on a company for a multi-year pass to keep it.
• The tested party’s profile — what it does (functions, assets, risks), at least at the level of a paragraph, plus its jurisdiction, whether it supplies goods or services, and its functional role. Quartyl benchmarks this profile against the dump; it is described in the form, not read from the spreadsheet.
• The decision you expect to make — which PLI the study should compute (for example, operating profit on operating costs for a routine services provider) and roughly where the tested party’s margin sits on it. The PLI and the tested party’s margin are entered as parameters in the last wizard step; the method logic is explained in the methods guides.

Step 1 — Create and configure (the three-step wizard)

1. Study Info & Upload. Study name, the tested party’s entity name, the financial year, and a description of the engagement; then the data file and the dump sheet inside it (auto-detected, overridable). There is no database or source selector — the workbook you attach is the population the screens work on.
2. Study Context. The tested party’s profile: jurisdiction, whether it supplies goods or services, its function type (a required multi-select, and the input the screening stages weigh), a nature-of-business description of at least 30 characters, and the scope fields — product scope, channel, accepted and excluded business models, geographies, IP ownership, related party profile, materiality note and any exclusion criteria you want on the record.
3. Parameters & Run. The PLI (margin type), the averaging method — Single Year (Latest) or Multi-Year (Golden Rule — Aggregate Raw Financials) — the percentile band for the range, the tested party’s own margin, and the data filters: maximum related-party-transaction share, minimum and maximum revenue (with the dump’s currency and reporting unit: actuals, thousands, millions or crores) and the minimum employee-cost ratio.

Check the column mapping before you run. On upload Quartyl detects which column is which field — it knows around thirty of them (revenue and its aliases like turnover, net sales or PBDIT; cost; operating profit/EBIT; operating expenses; employee cost; the RPT share; EBITDA; gross profit; total assets and the current-asset lines; NACE Rev. 2 and the national primary industry code; independence indicator; status; website; trade description). Correct any column it mis-reads and ignore any you do not want used. Company, revenue, cost and operating profit must map or the run will not start, and the PLI you pick must be computable from the columns you have mapped.

Save as Draft keeps the study in Draft for later. Run Analysis submits it — Study Initialized, then straight into Processing. Either way the parameters are recorded on the study: they are part of the record, not a runtime preference.

Step 2 — Run the pipeline

The run is an async job — you submit it, the worker picks it up, and the study shows its progress step by step rather than as a bare percentage:

Step	What it does
Queued	Waiting for a worker
Preparing Data	The workbook is read, the column mapping is applied, and the normalized dump is cached for the rest of the run
Quantitative Analysis	The uploaded population is screened: the revenue, RPT and employee-cost filters you set, plus the standing financial and status rejects
Qualitative Screening	Each company’s business description is compared with the tested party’s profile using local embeddings; no language model and no external call is made at this step
AI Screening (FAR Analysis)	The language-model pass: functions, assets and risks read off each company’s description, with a verbatim evidence quote required for every recommendation
Web Research	The optional research pass — candidate websites are found and read, and what comes back enters the report as findings labelled with the URL behind each one. Needs the same plan feature as the deep pass, and a failure here never blocks the run
Advanced AI Screening (Deep FAR)	The stricter second pass, where the firm’s plan enables it — a different model, a fixed schema, and anything it cannot resolve fails closed to flagged
Final Analysis	The statistics on the accepted pool, the arm’s length range, and the conclusion

The deep-FAR step stays in the stepper and is marked (Not in your plan) when the firm’s plan does not include it — the record shows what did not run, not a shorter list. Web Research rides that same single plan feature, and when it does not run the report fields it would have filled stay blank rather than holding a guess.

You do not need to watch it. The study moves to In Review when it completes, and your queue (and the dashboard’s activity feed) tell you it is ready.

Step 3 — Read the comparables grid

The run’s output is the comparables grid: the screened population sorted into four buckets — accepted, almost accepted, flagged and rejected — each company with the recorded reason for its place. This is the work surface of the study in review:

• Verify the accepts. The company is genuinely comparable to the tested party’s profile; the reason says so with the evidence behind it.
• Review the rejects. The reason is specific and factual (not “not comparable”); the company is genuinely out.
• Work the almost-accepted and flagged rows. These are the borderline buckets — the screen’s own way of saying “a human should look”.

The reviewer’s action on any row is one decision: override to accept or to reject, with a reason. That action belongs to the review roles — Manager, Firm Admin, Superadmin — and only while the study is in In Review; as the analyst you read the grid, prepare the case and address what comes back. The grid prompts for the reason and stores it on the override ledger with your name on it — leave the box empty and it is recorded as “no reason provided”, which is exactly as visible to a transfer pricing officer as it sounds. The committed batch is what re-computes the pool and the statistics, and if a second reviewer has moved the same row in the meantime your stale edit is refused rather than silently overwriting theirs.

The grid is where a study becomes defensible: the range is downstream of these dispositions, and the dispositions are the record.

Step 4 — Read the results

With the grid settled, the results view gives you the study’s conclusion:

• The pool statistics for the chosen PLI: count, mean, median, standard deviation, minimum and maximum, first and third quartiles, the interquartile range, the coefficient of variation, the 10th and 90th percentiles, and the Pearson and Spearman correlations between the indicator and its candidate drivers.
• The arm’s length range. Normally the percentile band you set in the parameters (25th–75th by default). Six jurisdictions prescribe their own band and it is theirs that applies: India’s Rule 10CA 35th–65th percentile range, with its six-company minimum, and the 25th–75th convention recorded for the UAE, the USA, the UK, Singapore and Germany. Below three accepted comparables no percentile band is published at all — the arithmetic mean is reported as a point estimate, because a three-point interpolation is not a statistic.
• The conclusion. The tested party’s margin is placed against that band and the study says whether no adjustment is indicated, or an upward or a downward transfer pricing adjustment may be required. Quartyl does not run working-capital or functional-segregation comparability adjustments: the computed quantity is the adjustment itself.
• The supporting scores — the reliability and risk signals the analysis computes from the pool (sample size, dispersion, how tight the interquartile range is against the median, and how many companies in it are loss-making). These are documented weighted rules applied to the numbers in front of them, not a trained model’s opinion.

Inside the range, the analysis is done: submit for review. Outside it, the grid and the parameters are where the cause lives — that is the review to do before submission, not after.

Step 5 — Submit and follow the workflow

There is nothing further for you to send: the completed run puts the study in In Review on its own. From there the workflow runs on roles, not on you. The reviewer (Manager, Partner, Firm Admin or Superadmin) approves — which triggers the Excel workbook, the primary deliverable — or rejects it back to you with the reasons to address, and you re-submit from Rejected. The seven-chapter Word master report, with the branded PDF alongside it, is the on-demand final document: a Partner, Firm Admin or Superadmin requests it once the results exist, with the engagement details, and it does not wait for sign-off. Sign-off and archiving close the study out. The roles page has the complete transition map; the study’s own state stepper shows where it is at any moment.

First-run issues, the ones that actually happen

Symptom	Usual cause	Fix
The upload is refused	CSV is not accepted; the wizard takes an Excel workbook (.xls/.xlsx/.xlsm) up to 50 MB, and each account holds at most 25 stored uploads	Export the dump as Excel, and clear old uploads if the cap is what bit you
A line item computed wrong	A column mapped to the wrong field in the wizard	Correct the mapping (or re-upload) while the study is still Draft or Study Initialized, then re-run; past that, clone the study into a fresh draft
The pool looks empty or tiny	The filters are doing it: min/max revenue, the maximum RPT share, the minimum employee-cost ratio, an extreme indicator band, or fewer than two years of data per company on a multi-year pass	Widen the parameters against the tested party’s actual data shape and re-run
A disposition you disagree with	The screen classified on a description that was thin for that company	It is the reviewer’s row to move: accept or reject it as an override with the reason, while the study is In Review
The study failed mid-run	The worker hit an error, or a job whose worker died was reaped as stale	Re-run it — Failed → Processing is the workflow’s own path

FAQ

How long does a run take? There is no fixed budget to fit into: the worker allows a run up to two hours before cutting it off, and a job whose worker dies is reaped and marked failed rather than left hanging. A study sitting in Queued is waiting for capacity, not stuck — a firm may have at most three analysis jobs queued or running at once, and when the platform-wide queue is full new submissions are refused outright instead of quietly piling up. The step stepper shows progress either way; there is nothing to do while it runs.

Can I change the PLI or the years after the run? Only while the study is still Draft or Study Initialized — that is the window the wizard stays open in. Past that the parameters on a study are part of its record: to change them you clone the study, which copies its identity and parameters into a fresh Draft for you to edit, re-attach the dump to and re-run. The source study keeps its results, comparables and audit trail untouched.

Is the Excel report generated during the run? No — the pipeline produces the analysis; the Excel workbook is built when the study is approved (at report generation), from the settled record. The seven-chapter Word master report and the branded PDF are the on-demand final document, requested by a Partner, Firm Admin or Superadmin any time from In Review onwards.

What if I am not the Analyst — I was invited to review? Your loop starts at step 3: the study lands in In Review, in your queue, and the grid (where you move rows by accepting or rejecting them with a reason), the results and the approve/reject decision are your work surface from there.