Half of next year’s tenders are already predictable — we counted
We tracked 444,534 buyer-and-category pairs through six years of contracting. Between 54% and 62% of them buy the same category again the following year. Public procurement is a calendar, not a lottery.
Ask a bid manager where next year’s pipeline will come from and you will hear “we watch the portals.” Watching the portals means learning about a tender when its clock starts. But how much of public buying is actually new — and how much is last year’s buying, repeated?
We took every combination of buyer and purchase category (CPV division) in our corpus and asked a blunt question: if a buyer bought a category in year N, did they publish in the same category in year N+1?
| Cohort | Buyer×category pairs | Bought again next year |
|---|---|---|
| 2019 | 51,026 | 54% |
| 2020 | 56,947 | 57% |
| 2021 | 91,115 | 62% |
| 2022 | 140,170 | 54% |
| 2023 | 185,493 | 56% |
| 2024 | 444,534 | 56% |
The practical consequence is large. If roughly every second tender is a repeat, then roughly every second tender can be prepared for before it is published— by reading the buyer’s history instead of refreshing a portal. The incumbent is doing exactly that, without calling it anything: they simply know when the contract they hold expires.
The rule of thumb we would give any bid team: list the buyers who bought your category in the last eighteen months. Half of that list is your real pipeline for next year — with names, dates, and last-time prices attached.
Data & method
Pairs = normalized buyer name × 2-digit CPV division with ≥1 notice in the cohort year, from our buyer-activity aggregate (2019–2025). Repetition = ≥1 notice by the same pair in the following calendar year. Rates are lower bounds: buyer-name variants split pairs, and categories bought on multi-year cycles (e.g. equipment every 3 years) count as non-repeats in this design.