prices_06-10.csv
Prices in pence, not pounds
unit_price 1299 · yesterday 12.99
Schema: passVelraq: flag
stock_06-10.csv
Yesterday's file, sent again
same hash as stock_05-10.csv
Schema: passVelraq: flag
orders_06-10.csv
Half a file that looks complete
4,812 rows · usually 9,600+
Schema: passVelraq: flag
None of these crash your pipeline. That's the problem.
Velraq flags the supplier files that load fine but are wrong, before they turn into wrong prices, wrong stock or a customer complaint. Each file is checked against that supplier's own history.
Send copies of your top 5 suppliers' past files and get a report within 48 hours, before Black Friday on 27 Nov. Nothing connects to your systems, and your suppliers change nothing.
Why it matters
Nobody needs a tool to save 30 minutes.
You get files from dozens of suppliers every day. Most are fine. The ones that aren't still load, so nothing tells you until the damage is done.
01
Stop hearing about bad supplier data from your customers first.
02
Stop being blamed when the 9am report is wrong.
03
Get suppliers to fix their own mistakes, with the evidence in hand.
And AI shopping assistants now read your product data too: a wrong price or stock level from a supplier file gets repeated to shoppers before anyone looks.
How it works
It learns each supplier's normal, then flags what doesn't fit.
Watch
Reads supplier files where they already land. Suppliers change nothing.
Learn
Builds each supplier's normal from their own past files: row counts, price ranges, which products and warehouses usually appear.
Flag
Good rows go through, only the bad rows are held, and the report goes out on time with a label like:
Example
1,204 supplier prices held, using yesterday's. Everything else updated.Fix and score
Drafts a note to the supplier with the evidence, and updates their Data OTIF score.
Todayon your old files, with Velraq Scan.In the pilotautomatically, every morning.
You choose what happens to a bad file.
In the pilot
Watch onlyFlag it, change nothing.
Hold bad rows DefaultGood rows load. Bad rows wait for a fix.
Hold the whole fileNothing from it loads until it's fixed.
It starts in watch-only mode for 2 weeks. Our target after that: no more than 1 false alarm per feed per month. One click marks anything as fine.
In the pilotwatch-only start and the false-alarm target.Today"This is fine" marks in the scan report.
For engineers: how "normal" is learned
Each supplier's normal comes from its own recent files: the usual row count, each column's typical values and how often it's blank, the values a status column takes, the products and warehouses that are usually there, and when the file usually lands.
A file is flagged when it falls well outside that supplier's normal, or matches a known failure pattern for stock, price and product files: a byte-for-byte resend, a unit switch, dates that didn't move, products that quietly dropped out.
Nobody writes rules per supplier. When something is fine, you mark it, and where it makes sense the new level becomes normal. The price checks start on a supplier's 6th file and are at full strength from about the 21st, so a few months of files per supplier is best.
Today I run Velraq Scan on copies of your files. In the pilot the same checks run on each new file as it lands.
Tested so far
A whole price file multiplied by mistake still loads.
VAT added twice, a currency conversion, a doubled column. Every price still looks like a price, so nothing complains. Velraq's price checks are built to catch it, and to tell it apart from a genuine supplier price change.
95-96 in 100
price files multiplied by mistake were caught.
x1.2 to x2, on 13 months of a UK online retailer's real sales files, with the errors injected by us. About 3 false alarms per 100 clean files. The basic checks alone caught 7 to 25 in 100.
From file 6
a new supplier's files are checked.
On ONS shop prices (2012 to 2014) arranged by us as monthly supplier files, 4,258 of 4,260 files with every price x1.2 were held among a supplier's 6th to 20th files, against about 6 in 10 before we taught the checks to start this early. In that period about 7 in 100 good files were held too.
85% and 96%
mistake or real price change?
When a whole file moved by one factor, 85 in 100 injected scaling errors were labelled mistakes and 96 in 100 simulated price rises (re-rounded to price points) were labelled price changes. A pointer, not proof: a person confirms each one.
Each test's rules were fixed and approved before it ran, on data held back from design, and it was scored once. The errors were ours, injected into real public data rather than supplier price lists, and these checks haven't yet run on live supplier files. That's what the free check is for.
ONS price quotes: Office for National Statistics. Contains public sector information licensed under the Open Government Licence v3.0.
See it work
A report on 30 days of supplier files.
On 30 days of sample files from 3 fictional suppliers, the scan found 11 problems. A schema check alone would have caught 1.
Sample data from fictional suppliers
10 of 11 problems would pass a schema check
89files checked
11problems found
1caught by a schema check alone
Brightway Prices
Hollowmere
Larkhaul Dispatch
See the sample report
Example
Tue 06-10· 214 files checked· 1 flagged· 2 late
Most days look like this. Quiet means checked and clean, not doing nothing.
Data OTIF
Score your suppliers on their data, like you score their deliveries.
Data OTIF % = files on time, in full and correct / files expected
| Supplier | Files expected | On time | In full | Correct | Data OTIF |
|---|---|---|---|---|---|
| Larkhaul Dispatch | 30 | 100% | 97% | 93% | 90% |
| Brightway Prices | 30 | 97% | 97% | 97% | 90% |
| Hollowmere | 30 | 97% | 93% | 87% | 83% |
Take it into your next supplier review.Which supplier sent files late, which sent half files, which sent prices in pence. With dates and evidence for each.
Hold them to account.Big retailers already hold suppliers to account for paperwork and order accuracy. Velraq gives you the same leverage for the data they send.
Todayin the scan report.In the pilotmonthly and automatic.
Your data stays yours
Copies only today. In the pilot, none of your rows leave your account.
Today
You send copies, I run the scan.
Redacted files are fine: swap names, but keep prices and stock as they are. The scan runs offline with no uploads to any service, and your files are deleted within 7 days of your report. Happy to sign your NDA. How files are handled.
In the pilot
The checker runs inside your own AWS account (London region).
Only counts and check results leave it, to build the scorecard. Suppliers change nothing.
For engineers: what runs where
Today: Velraq Scan is a Python script I run offline on the copies you send. It imports nothing that talks to a network, and the HTML report it writes loads nothing from the internet.
In the pilot: the checker runs inside your AWS account in the London region (eu-west-2), next to where supplier files already land. Rows, values and product codes stay there. What leaves is per file: supplier, date, row count and which checks passed or failed, to build the scorecard.
How to start
Start with your old files.
Step 1 Today
Free pre-Black Friday check.
Send a few months of files from your top 5 suppliers, ideally 25 or more each (redacted is fine). You get the report within 48 hours, with a one-page note per supplier you can forward to them.
Step 2 The pilot
Free 3-month pilot on 3 supplier feeds.
£0 for 3 months
Velraq runs on 3 of your supplier feeds, every morning, in your own AWS account. After that, £1,200/month for up to 15 feeds, only if it met the measures we agreed.
How the pilot works
Week 1: the checks run over your last 3-12 months of files and we agree what success looks like. Week 4: we review catches and false alarms together. Month 3: you decide. Nothing to pay until then.
FAQ
Questions people ask.
Why not write our own checks?
You can, and for schema checks you should. The silent problems need each supplier's history, and someone keeping 30+ suppliers' limits right. That's the work Velraq does.
Do our suppliers need to change anything?
No.
Is our data safe?
For the free check you send copies (redacted is fine). The scan runs offline and your files are deleted within 7 days of your report. In the pilot it runs in your AWS account, and only counts leave.
What does the free pilot commit us to?
Nothing binding. Up front we agree the success measures, a decision date and who decides. If it meets them, it's £1,200/month for up to 15 feeds. If not, you walk away.
What if it flags something that's fine?
It starts with 2 weeks in watch-only mode. One click marks it as fine, and it learns.
We already use dbt tests / Great Expectations / Soda.
Those mostly check rules you write, after data has landed. Velraq checks each supplier's file against that supplier's own history before it lands, and gives you the evidence to take back to the supplier. They work side by side.
What files does it read?
Today CSV, text (.txt) and Excel (.xlsx).
In the pilot JSON and XML too. EDI (EDIFACT, EANCOM, TRADACOMS): after your EDI translator.
Not yet PDF.
Adwait Kumbhare
Data engineer working with Python, SQL, AWS, Snowflake and Databricks, with an MSc in Data Science and Analytics from Brunel University London. I built Velraq Scan, and I run every report myself.
LinkedInTalking to 10 data & ops people in UK retail, ecommerce and logistics.
20 minutes. No pitch.