How to Publish Original B2B Research From Your Own Data

Your systems already hold the study

The research your buyers can’t get anywhere else is already in your ERP, your TMS and your ticket queue. Pull one finding from it, with its count, its period and its definition, and publish it. That’s original B2B research, and nobody had to fill in a survey.

We said it in one line in our post on thought leadership: your own data beats anybody’s survey. A survey records what people say they did, and your systems record what actually happened.

4 exports are worth a first look:

• Quoted lead time against actual lead time, by product family and quarter.

• On-time-in-full by region, counted in shipments.

• Returns by failure mode, counted against units shipped.

• Ticket categories and time to first fix, without the free text.

In our post on content calendars, we listed your own data as a lead time to plan for, because somebody has to pull the number, check it and clear it for publication.

One limit holds throughout: a table from one company describes that company’s customers, so say so on the page.

One finding from a pump repair log

Neither business in this post exists. We made both up, figures included.

A service centre repairs industrial pumps. Every job is a ticket with a model, a failed part and a cause. 30 months of tickets give this:

Of 1,184 pumps repaired between
January 2024 and June 2026, 412 had a
failed mechanical seal. In 239 of
those 412 (58%), the technician
recorded dry running as the cause.
Source: repair tickets, all customers,
3 workshops.

A maintenance engineer can use that, because it has a count, a period and a source. It doesn’t name a customer or a technician. A buyer will check it for 2 of the 4 parts that our B2B case study examples post asks of any result: the period and who counted.

57 of the 412 seal tickets had no cause recorded, so they count in the 412 and can’t count in the 239. Say so in the method note.

We can build a white paper around a finding like this one, with your specialists interviewed on what it means. Pulling the export and running the numbers stays on your side.

Who has to say yes to an aggregate

An aggregate needs 4 sign-offs: the contracts, personal data, competition and process secrets. Your lawyer or data-protection lead makes each call, and the texts below are what they’ll read.

A case study needs 3 permissions from one customer. A table across every customer raises different questions.

The contracts. Don’t count until you’ve read the confidentiality clauses in your customer agreements. If the data comes from connected products sold in the EU, the Data Act has applied since 12 September 2025. Article 4(13) lets the data holder use non-personal product data only “on the basis of a contract with the user”.

Article 4(14) limits passing that data to third parties. Whether a published aggregate counts as passing it on is a question for your lawyer, and the contract wording is where to start.

Personal data. Tickets carry contact names, technician IDs and free text, and hashing a technician ID doesn’t make a record anonymous. Under GDPR Recital 26, pseudonymised data that could be re-linked “should be considered to be information on an identifiable natural person”. Drop the free text and the people fields before you count.

Competitors. Our second invented business is a regional LTL carrier. It can publish why its deliveries failed in 2025: 2,786 of 9,620 exceptions in its TMS (29%) carried the reason code for a closed receiving window. It shouldn’t publish next year’s rates.

The European Commission’s guidance on information exchange says “Aggregated and historical information is generally less strategic”. Announcements should stay “limited to the undertaking itself and not extend to the sector or industry”. Paragraph 394 says information can count as historical once it’s “several times older than the average length of the pricing cycles”. Paragraph 398 names a post on a public website among announcements that can still be a concerted practice.

Process secrets. A scrap rate or a process setting can be a trade secret. EU law protects information that “has commercial value because it is secret”. US law looks for value from “not being generally known”. Publication ends both, so the person who owns the process signs it off.

Three tests before a table goes public

Check that no cell stands for one customer, that nothing public can be joined to it, and that 2 releases can’t be subtracted to expose one customer.

The EDPB, the body of EU data protection authorities, released draft anonymisation guidelines for consultation in July 2026. They name 3 criteria: No Record Isolation, No Linkage, No Inference. They’re written for personal data. Ask the same 3 questions about customer companies.

1. Isolation. We’d hold every cell to at least 5 customers, and check that 1 customer doesn’t carry most of a cell alone.

2. Linkage. Region, industry and pump size together can point at one plant that the whole trade knows.

3. Inference. The EDPB warns that “many techniques exist that can reveal information relating to one or several single individuals from aggregate data”. A simple one is subtraction: a post, a deck and a chart cut from one export are 3 releases, so check them against each other.

A cell that fails gets merged or held back. These 3 tests rest on an EU text plus our own floor. Outside the EU, treat them as a yardstick.

A page someone can cite

Put the finding on an open web page, with its method, a CSV and a line telling people how to cite it.

Google’s Dataset guidance asks that the page not be “blocked by a robots.txt file, the noindex tag, or login requirements”. A PDF behind a form doesn’t meet that. If you want the form, gate the long analysis and keep the findings open, a trade-off we worked through in which content to gate.

Write the finding as 1 sentence with its count, period and source, for the reasons in the anatomy of a quotable page. Every number cut from the report keeps the lines we listed in repurposing a white paper. The method note adds the source system, exclusions, minimum cell size and version.

Put the CSV under CC BY 4.0, the Creative Commons attribution licence. Section 3(a)(1)(A) has reusers credit you “in any reasonable manner requested by the Licensor”, so write the line you want. Database rights are covered too, by Section 4. Creative Commons warns, “Our licenses are irrevocable”, so choose before release.

DataCite suggests an order for the citation line: creator, year, title, version, publisher, type, identifier.

[Company] (2026): Mechanical seal
failures in repaired pumps, 2024-2026.
Version 1. [Company]. Dataset. [page
URL]

Sources

• European Data Protection Board, Guidelines 02/2026 on Anonymisation, version 1.0, adopted 7 July 2026 for public consultation. Draft; may change.

• GDPR, Regulation (EU) 2016/679, Recital 26.

• Guidelines on horizontal co-operation agreements, 2023/C 259/01, paragraphs 394, 398 and 411. European Commission, non-binding.

• Data Act, Regulation (EU) 2023/2854, Article 4(13) and (14). Connected products and related services only.

• Trade secrets: EU Directive 2016/943, Article 2(1), and 18 U.S.C. 1839(3).

• Google Search Central, Dataset structured data, last updated 8 September 2026.

• Creative Commons, CC BY 4.0 legal code, Considerations for licensors, Section 3(a)(1)(A) and Section 4.

• Citation format: DataCite Metadata Schema 4.7, released 3 March 2026.

To start, we need the one sentence you want to defend and the system its count comes from. We’ll reply inside 1 working day, with a scope and a price. A white paper is quoted per project. The export comes after we sign your NDA, and whether a figure goes in aggregated, anonymised or not at all is settled at the outline stage. A single finding fits one article at $149.