Original research · 4 markets · 89 purchases
What actually comes
before a B2B purchase?
We checked 89 real purchases against each buyer's own year before. Here is what held.
Mason / Research · Published 9 November 2026
Part of a larger body of work: 67 B2B sellers researched and 92,315 search results read. Of the 46 signals we pitched whose record gives a verdict, 1 held as first worded, 29 more held once we narrowed them on the evidence, and 16 did not hold. Read the benchmark.
We read the 12 months before 89 B2B purchases in four markets, then the 12 months before that, for the same company. Most popular buying signals, like a new leader, a launch or funding, were just as common a year earlier. What rose changed how the business runs: a new system, a new team, a new route to market.
THE QUESTION
Did the same thing happen
the year before?
If it did, the event describes the company, not the moment. That is fine for picking accounts and useless for timing outreach.
HOW WE CHECKED
Two years for every purchase.
Same company, same sources, same rules.
For each purchase we read the 12 months before it, then the 12 months before that, and counted each event in both.
01 / The test
Why the year before matters
Every signal list looks convincing when you only look backwards. Take a company that bought. Look at its last year. You will find a new hire, a launch, a press release, a partnership. Something always happened.
The question nobody asks is: did the same thing happen the year before, when the company was not buying? If it did, the event describes the company, not the moment. It is still useful for picking accounts. It is useless for timing your outreach, which is the whole reason people pay for signals.
So for every purchase we read two years: the 12 months before the purchase (the buying year) and the 12 months before that (the year before).
Same company, same sources, same rules.
02 / Method
How we did it
The purchases are real.
They come from the seller's own closed deals, dated by its own records, and from competitors' public customer pages, dated by a page that names both sides.
The sources are public.
The buyer's own website and store, job posts, staff profiles, press releases, company registers, filings, certificate logs (new web addresses) and app reviews. It is the same research method we use for clients.
An event counts only if it has a date
and falls inside the window. A year-only date never counts toward timing.
We only count companies we could actually see.
If a source did not reach back far enough for a company, that company leaves the denominator for that year. That is why some rows read "of 17" against "of 14".
Our rule for "rose"
The event shows up in at least 15 percentage points more of the buyers in the buying year than in the year before. Studies B, C and D use that rule. Study A reports its counts as they came.
| Study | Who bought, and what | Purchases |
|---|---|---|
| A | Mid-sized companies in Germany, Austria and SwitzerlandA commerce or content platform project from a digital agency | 28 |
| B | US Shopify brandsAn outside ecommerce bookkeeping and accounting firm | 20 |
| C | Direct-to-consumer brands in the US, Canada and UKAn email and SMS retention agency | 19 |
| D | Robot makersA robotic hand or gripper | 22 |
03 / Finding 1
The popular signals were just as common a year earlier
Leadership changes, launches and funding are on every signal list. Here is how often they showed up in the buying year against the year before, for the same companies.
buyers with leadership activity in the buying year
in the year before
| Event | Buying year | Year before |
|---|---|---|
| Leadership activity Study A | 26 of 28 | 26 of 28 |
| A marketing or retention leader starts or leaves Study C | 13 of 19 | 11 of 19 |
| International activity Study A | 25 of 28 | 25 of 28 |
| A new product line Study B | 6 of 20 | 11 of 20 |
| A first-ever product category Study C | 6 of 16 | 7 of 16 |
| A new robot or product line Study D | 11 of 17 | 7 of 10 |
| The SKU count steps up Study B | 7 of 20 | 6 of 20 |
| A funding round or a loan Study C | 4 of 17 | 6 of 17 |
| An email or lifecycle specialist leaves Study C | 6 of 19 | 5 of 19 |
A launch at a product company is a heartbeat, not an event.
In Study B, new product lines were more common in the year before (11 of 20) than in the buying year (6 of 20). Brands that launch a lot are the kind that end up needing outside accountants. That tells you who to call. It says nothing about when.
04 / Finding 2
What did rise changed how the business runs
Six events rose by 15 points or more in Studies B and D, and two in Study A point the same way on rougher counts. None of them is a headline. Each one changes the work inside the company: a system to run, a team to manage, a new route to money.
brands put in a finance or inventory system in the buying year
in the year before
Read these with care. The samples are small: with 28 deals, only effects of about 2.5 times or more stand out from noise. Outside money in Study B rests on one deal whose date we corrected. The web-address row in Study A counts mentions of a theme, not 12 confirmed buying decisions. Every count on this page is in one CSV.
| Event | Buying year | Year before |
|---|---|---|
| A core system changes (ERP, product data, CRM or warehouse) Study A | 13 of 28 | about 6 of 28 |
| A finance or inventory system goes in Study B | 5 of 20 | 1 of 20 |
| An AI or robot-software stack is adopted Study D | 13 of 20 | 7 of 19 |
| A production, field-service or data team is stood up Study D | 6 of 17 | 1 of 14 |
| A new sales channel opens (a distributor, a national retailer, a store abroad) Study B | 5 of 20 | 1 of 20 |
| First people or a venture in a new country Study D | 6 of 17 | 3 of 17 |
| Outside money arrives that wants the numbers Study B | 4 of 20 | 1 of 20 |
| Development, test and staging web addresses appear Study AAbout 5 of the 12 were the build itself; without them, about 7 against 5. | 12 | 5 |
The pattern we trust most is about order, not any single event.
In Study A, a core system changed first and the platform came second, mostly 1.5 to 11 months later. The earliest trace was usually a person or a web address, not a press release.
A cutover or consultant role starts, or a new address like a test product-data site.
A core system changes: ERP, product data, CRM or warehouse.
The press catches up.
The platform purchase.
05 / Finding 3
One market had no timing signal at all
In Study C we checked 11 events against the year before for brands that hired an email and SMS retention agency. None rose by 15 points. The leadership move, the lost specialist, the new SMS channel, the new store platform, the new country: some were a little higher in the buying year and some a little lower, but none by 15 points. The old retention agency being replaced, the one you would bet on, showed up at 1 of 18 against 1 of 16.
That is a result, not a failure.
It means that in this market, public events tell you which brands fit, and the timing has to come from somewhere else: a conversation, a referral, or being the agency already in their inbox when they are ready.
06 / Finding 4
Real negatives
Cyberattacks.
5 of the 28 Study A buyers had been hacked at some point. None in the year before they bought: 2 before that year, 3 after the deal.
Regulator penalties.
We found none in any buying year. The 3 dated enforcement events we found all fell outside it. No complete register exists, so this is not proof of none.
Compliance deadlines.
Every one of the 28 buyers had a binding deadline inside its buying year. A filter that keeps everyone sorts no one. The count measured our deadline list, not the buyers.
Mergers and acquisitions.
In Study B, 0 of 20 brands showed a sale process before hiring the firm. 3 changed hands or entity after it. The books were outsourced first.
07 / Honest notes
What we got wrong along the way
- The purchase date was often the go-live. 10 of the 28 dates in Study A were launch dates, not signing dates. The "buying year" was months the agency was already building, so the project showed up as its own cause. Anchor on the decision, or exclude the last 3 months before a go-live.
- Some windows contained the purchase itself. In 5 Study A windows the announcement of the deal was one of the events. Exclude anything that is the purchase.
- Website dates lie. One newsroom had 19 of its 34 posts on a single modified date, from a site migration. That put zero of them in the window. Use the date printed on the page, never the file date.
- We started without the buyer's identity. The first pass of Study B had no legal name, domain or accounts for 28 of 30 deals, and every check after it was wrong. We rebuilt the deal list first.
- The answer sat in the wrong check. In one Study A deal, the check built to find a systems rollout said no. The rollout role was sitting in the acquisitions check's events, 11 months out. Read every event for a buyer as one timeline, not check by check.
Limits
- There is no group of non-buyers. We compared each buyer with itself a year earlier. That separates timing from company type. It does not tell you how often companies that never bought show the same event. Nothing here is a lift or a cause.
- The samples are small (19 to 28 purchases per market) and the four markets are very different. A pattern that holds in three of them is a lead to test in yours, not a law.
- Unknowns are left out, not counted as no. Where we could not see a year for a company, it leaves that year's denominator.
- The rule for "rose" is ours. A 15-point gap is a judgement, and in Study A we report the counts without it.
08 / Takeaway
The year-before test
You can run this on your own customers with a spreadsheet and an afternoon per ten deals. No data vendor needed. It works the same on a client's past deals if you are an agency or a market-entry consultant choosing which signals to sell.
- 01
List 20 to 30 of your recent purchases. Use the date they signed, not the date they went live. Write down which kind of date you have.
- 02
For each one, mark two windows: the 12 months before signing, and the 12 months before that.
- 03
Pick the signals you use or are being sold. For each buyer and each window, write fired, did not fire, or could not see. Link the page that proves it.
- 04
Drop anything that is the purchase itself, and anything inside 3 months of a go-live.
- 05
Count each signal in both windows, using only the buyers you could see in that window.
- 06
If it fires as often the year before, it is a "who". Use it to pick accounts, not to time outreach.
- 07
If it rises by 15 points or more, it is a candidate "when". Check the order and the lag before you build a campaign on it.
Want this run on your own deals?
We check your past buyers against their year before, then find the companies at the same point now.