What a 312-keyword tender alert really finds

Most tender alert services in the UK work in pretty similar way. You give them a list of keywords and if you're lucky CPV codes, and they send you every new notice that contains keywords and cpvs based on an algorithm from the 80s called BM25. A UK contractor that is thinking of moving to Open Opportunities shared the list it uses with its current service. We tested that list against a year of UK tenders, then replaced it with Activity Searches here's what we found.

What the list contained

The list had 312 keywords and 136 CPV codes. CPV codes are the standard category codes that public buyers attach to their notices. Together, the keywords and codes described 16 types of work. They ranged from tree surgery and ecology surveys to drainage, coastal defences, EV charging and solar panels.

Where keywords go wrong

A keyword search finds every notice that contains the word, whatever the word means in that notice. 62 of the 312 keywords have more than one meaning. We ran some of them through Open Opportunities keyword search, on UK tender notices published between 12 November 2025 and 12 November 2026, and read the first page of results for each.

Keyword	UK tender notices found in the year	Relevant among the first results	What else it found
lock	493	4 of 19	Door locks, locksmiths, laboratory equipment
island	307	2 of 25	Cross-border research, Falkland Islands and Jersey contracts
rail	463	0 of 15 were lineside work	Rail haulage, rail replacement buses, rail training
Leeds	157	1 of 22	Leeds Council, University, Hospital and more showed up

The supplier meant "lock" as in canal lock. Most buyers who use the word mean a door lock. "Leeds" was meant to be LEED, the building standard, so as written it matched notices from Leeds City Council. Between them, "lock" and "island" matched 800 UK tenders in the year.

The CPV codes had similar problems. 21 of them are codes for goods, so they match tenders to buy gritters, salt and electric vehicles. The supplier sells the work itself. 13 codes are very broad. One of them, 71000000, covers all architecture, engineering and inspection services. It covered 4,891 of the 35,715 UK tenders published in the year, which is about one tender in seven.

Some keywords work well. "EV" and "tree" both returned mostly relevant tenders. A long list only needs a few ambiguous words to produce a large number of irrelevant alerts.

The tenders that keywords miss

Buyers describe the same work in many different ways, and a keyword list can only hold the words its author thought of. Activity Search found relevant tenders whose titles contain none of the supplier's 312 keywords. The supplier's list included "Japanese knotweed" but not "rhododendron". The last two titles in the table are reference numbers, so no keyword could match them. A keyword search would find these tenders only if one of the keywords appeared somewhere else in the notice.

What we did instead

1. We grouped the keywords and CPV codes by the type of work they describe.
2. For each type of work, we wrote a short, plain description of what the supplier does.
3. Where one description covered several types of work, we split it into separate descriptions. Solar panels, heat pumps and LED lighting became three searches. This gave 21 searches in total.
4. We ran each description in Activity Search. Activity Search reads each notice, works out what the buyer wants done, and matches it to a description of what a supplier does.
5. Each search covered UK buyers and notices published between 12 November 2025 and 12 November 2026. That includes tenders, contract awards and early market engagement notices.
6. For each search, our team read the top 30 results and counted how many were for the work described.

What we found

In 19 of the 21 searches, at least 24 of the top 30 results were for the work described. 12 searches returned 30 relevant tenders out of 30.

The keyword tests above used the same UK tenders over the same 12 months, and also counted how many of the first results were relevant. "Lock" gave 4 relevant tenders in its first 19 results, and "island" gave 2 in its first 25. The keyword tests counted tender notices only, while the activity searches also included contract awards and market engagement notices.

Writing one description per type of work made a clear difference. A winter maintenance description that also mentioned gritting vehicles returned 13 gritting contracts in its top 30, and the rest were purchases of gritters and salt. A description of the gritting work alone returned 25. A single description of solar panels, heat pumps and LED lighting returned only solar contracts. Separate descriptions returned 26 relevant heat pump and building energy contracts and 25 relevant LED lighting contracts.

This review measured how relevant the top results were. It did not measure how many relevant tenders each approach finds in total. Our published benchmark measured that across five national portals: keyword search found about 30% of relevant notices, and Activity Search found about 95%.

What to check if you use keyword alerts

1. Search for your shortest and most general keywords on their own, and read the first page of results.
2. List the words buyers use for your work that are not on your list, starting with the titles of contracts you have won.
3. Check whether any of your CPV codes describe goods when you sell services.
4. Check whether any of your CPV codes cover a whole division, such as 71000000 or 90000000.

If you would like us to run the same review on your own keyword list, book a demo and bring the list with you.