DataLift User Manual

How to turn a Purchase Order into a production-ready CSV.

01 Welcome

DataLift takes a Purchase Order from Zara or Bershka and helps you produce the CSV used to manufacture the labels. Instead of typing every field by hand, you upload the PO PDF, the composition document, and the screenshots from the label generator. DataLift reads them with AI, fills the form, and you review and correct anything that needs a human eye.

The process is split into six steps. You always move forward, but you can go back at any moment if you spot a mistake. Nothing is sent anywhere until you click Export in the last step.

Before you start

Have ready: the PO PDF, the composition document, and the screenshots of the labels generated in the Inditex label generator. Without these inputs DataLift cannot produce a complete output.

02 Logging in

Open datalift.pages.dev in Chrome or Edge. Enter the email and password you were given. If you forgot the password, contact the administrator — there is no self-service reset.

Login screen with email and password fields — Fig. 01Login screen. Use the credentials provided by your administrator.

Once inside, you land on the home screen. Click New Order to start processing a PO.

03 Step 1 — Setup

Tell DataLift which brand the order belongs to. This is the most important choice in the whole flow because it controls every downstream prompt, validation, and CSV column.

Pick one of:

ZARA — then choose the sub-brand: Maag, Ecru, Vilet or Dub.
BERSHKA — no sub-brand.

Fig. 02Brand picker. Each brand triggers a different processing profile.

Important

If you pick the wrong brand and continue, the AI will use the wrong prompt and most fields will come back incorrect. Going back to fix it forces a full reprocess.

04 Step 2 — Upload files

Step 2 shows four upload cards. The exact set depends on the brand you picked in Step 1. Fill the cards that apply to your order:

PO — the Purchase Order PDF received from Inditex.
Labels — the label reference list (when the customer provides one separately).
Composition — the composition document for the garment.
LG — the screenshots captured from the Inditex Label Generator tool.

Step 2 with Bershka selected, four empty upload cards visible — Fig. 03Upload cards. Files stay on your machine until DataLift sends them to the AI.

When the required cards are filled, click Process. DataLift sends the files to the AI and waits for the structured response. Typical processing time is 30 to 90 seconds depending on the size of the PO.

Tip

If you have several screenshots of the Label Generator (different labels in the same order), upload all of them in the LG card. DataLift matches each screenshot to its barcode automatically.

05 Step 3 — Review & references

This is the longest step. DataLift shows everything it extracted from the PO and asks you to review and complete the data before moving on. Step 3 has three areas you must check: the overview, the warning banner, and the label references.

Overview

The main view shows the data grouped into three blocks:

General Information — order number, season, customer codes, destination and currency.
Prices — unit prices by currency.
Order Table — the full breakdown of references and quantities.

Read each block carefully. Every field is editable — correct anything that does not match the PO before continuing. The currency drives the hemisphere used in care labelling (GBP, USD and EUR are Northern; AUD, CLP and BRL are Southern), so do not skip it.

Step 3 overview with General Information, Prices, and Order Table visible — Fig. 04Step 3 overview after a successful extraction.

Warning banner

If the AI flagged anything for human review, a blue banner appears at the top of the page with one or more messages. Typical examples:

Price 'black' — a price field came back with an unexpected value.
No label references — no references could be matched from the inputs.

Each message has a dismiss button. Read the message, fix the underlying data, and dismiss it. Do not dismiss a warning without fixing what it points to — the banner is the only place these issues are surfaced.

Blue warning banner with manual review messages — Fig. 05Manual review banner. Dismiss each message only after fixing the underlying data.

Label references

The Label References section is where you confirm the label codes that will go on the garment. It has:

Three dropdowns — for the standard label slots. Pick the right code for each.
Manual reference field — for any extra reference that is not in the dropdowns. Type it as it appears on the PO.

Valid references start with SP, HP, CL, ST or WT, and have Z as the third letter. DataLift filters out invalid prefixes (PL, WP, MR, RA) and the excluded range CLZCALL010–CLZCALL019 automatically.

Label references section with three dropdowns and a manual reference field — Fig. 06Label references. Use the dropdowns first; add manual references only if needed.

06 Step 4 — Composition & care

Step 4 covers two things on the same screen: the composition of the garment (split by part) and the care instructions.

Composition

Use only the principal or aggregated composition. Ignore breakdowns of recycled fibres.
Each part must add up to exactly 100%.
Multiple parts are separated by | (space-pipe-space) when written into the CSV.
A typical garment has two parts (for example, body + lining). More are possible.

Care instructions

All care text appears in UPPERCASE English. The order of the icons follows the GINETEX standard:

WASH
BLEACH
TUMBLE DRY
IRON
DRY CLEAN

If the source document uses the Korean KS K 0021 order (which swaps IRON and TUMBLE DRY), DataLift reorders the icons automatically. You do not have to fix this by hand.

Step 4 with two composition parts and care icons — Fig. 07Composition parts and care icons on a single screen.

07 Step 5 — Size range

DataLift lists every size present in the PO with the corresponding quantity. Check that:

The sizes match the size range on the PO exactly. Missing or extra sizes mean the AI misread the table.
Quantities add up to the PO total.
Each size has the correct barcode and S.Code (positions 12 and 13 of the 14-digit barcode from the Label Generator).

If you uploaded Label Generator screenshots in Step 2, an extra Validation column appears showing whether each size matched a screenshot. If you did not upload LG screenshots, the column is not shown — that is normal.

Step 5 size table with quantities and optional Validation column — Fig. 08Size range. The Validation column only appears when LG screenshots were uploaded.

Reference codes

If you see a reference starting with PL, WP, MR, RA, or one inside CLZCALL010–CLZCALL019 here, something is wrong upstream. Go back to Step 3 and review the label references.

08 Step 6 — Export CSV

The final screen is the export. Confirm the summary looks right and click Export. The CSV is saved to your Downloads folder with a name based on the order number. Send it to production through your normal channel.

Export screen with download button — Fig. 09Export. The order is also saved to your local archive in case you need it again.

Order archive

DataLift keeps a local archive of every order you process. You can open it from the home screen to review or re-export past orders without reprocessing.

09 Warnings

A few things will always cause trouble. Watch for them at every step.

Wrong brand in Step 1

The brand controls everything downstream. Changing it after Step 2 forces a full reprocess.

Dismissed warnings without fixing them

The Step 3 banner is the only place AI-flagged issues are surfaced. Dismissing without fixing produces a clean-looking but wrong CSV.

Composition that does not sum to 100%

Production will reject the label. Fix the percentages before exporting.

Mismatched barcode and S.Code

The S.Code is positions 12–13 of the 14-digit barcode. If they do not match the screenshot, the AI matched the wrong screenshot to the wrong size.

Care icons in Korean order

DataLift normally fixes this, but if you see IRON before TUMBLE DRY in the output, fix it manually and report it.

Real PO data in the wrong place

Never email a CSV outside SML and never share screenshots that include customer barcodes. POs are confidential.

10 FAQ

Can I process several POs at the same time?

No. One order per session. Finish or cancel the current one before starting another.

My PO has two garments. What do I do?

Process each garment as a separate order, with its own composition and its own Label Generator screenshots.

The AI got the country wrong. Why?

Some POs hide the destination country in the small print. Edit it manually in Step 3 — that is what the review step exists for.

What does Price 'black' in the warning banner mean?

The AI returned a price field with an unexpected value. Open the Prices block in Step 3, find the row it refers to, and correct the price.

Can I edit a CSV after exporting?

Yes, the file is plain text. But the recommended path is to fix the data inside DataLift, export again, and discard the bad CSV. That way the archive stays clean.

Where are my past orders?

In the local archive, accessible from the home screen. The archive lives in your browser, so it is tied to the machine and browser you used to process the order.

Does DataLift work offline?

No. It needs to reach the AI service to extract data from your files.

Which browsers are supported?

Chrome and Edge on Windows. Other browsers may work but are not tested.

11 Troubleshooting

Symptom	Likely cause	Fix
Login fails with a correct password	Account disabled or wrong email	Contact the administrator.
Processing takes more than three minutes	Large PDF or AI service backlog	Wait. If it fails, retry once. If it fails twice, contact the administrator.
Warning banner shows "No label references"	References missing from the PO or not detected	Add them by hand in the Label References section of Step 3.
Composition shows 99% or 101%	AI misread a digit	Edit the percentages manually so each part sums to 100%.
A size is missing from Step 5	AI missed a row in the PO size table	Go back to Step 2, reprocess. If it happens again, add the size by hand.
S.Code does not match the screenshot	Screenshots assigned in the wrong order	Re-upload the LG screenshots in Step 2 in the correct order.
CSV download does not start	Browser blocked the download	Check the browser’s download notification and allow it.
Archive is empty after closing the browser	Private/incognito window or cleared site data	Use a normal window and do not clear site data for `datalift.pages.dev`.

12 Glossary

PO: Purchase Order. The document Inditex sends with the list of labels to produce.
CSV: Comma-Separated Values. The plain-text file format DataLift exports for production.
LG: Label Generator. The Inditex web tool that produces the barcode and reference numbers for each label.
GINETEX: European standard that sets the order and meaning of care symbols: wash, bleach, tumble dry, iron, dry clean.
KS K 0021: Korean care-labelling standard with a slightly different icon order. DataLift normalises it to GINETEX.
Hemisphere: Northern or Southern, derived from the destination currency. Affects which seasonal label is used.
S.Code: Positions 12 and 13 of the 14-digit barcode produced by the Label Generator. Identifies the size variant.
Reference: Label code on the PO. Valid prefixes: SP, HP, CL, ST, WT. The third letter must be Z.
Sub-brand: For Zara only: Maag, Ecru, Vilet or Dub. Determines the specific prompt and field set used by the AI.
Manual review banner: Blue banner at the top of Step 3 listing issues the AI could not resolve. Dismiss each message after fixing the underlying data.
Order archive: Local list of past orders kept in the browser. Survives reloads but not a cleared cache.