How to build tender alerts that actually hit the mark

A good tender alert should make Doffin work easier, not more tiring. (In Norway specifically, Doffin is the national tender database where contracting authorities publish their notices.) Even so, we see many suppliers give up on their alert setup after a few weeks, not because it does not work, but because it alerts on too much. The alerts fill the inbox, and the relevant opportunities drown in the noise.

We have seen many setups over the years, both the ones that work well and the ones that tip over. In this article we share what we see working for the suppliers who have built an alert that is actually useful.

What a good tender alert should actually do for you

In contrast to tender search, where you actively go in and look for relevant notices when you have time, the alert is a passive tool that delivers what you have asked for. It is worth being clear about what you want out of it before you start configuring anything.

A good alert usually combines three or four filters: CPV codes and keywords to catch the relevant notices, geography where relevant, and a relevance score based on the company profile. It helps to understand how these filters interact: CPV codes and keywords widen the net, because a notice is alerted if it matches either one of your codes or one of your keywords. Geography and the relevance score do the opposite and narrow things down. It is the interplay between what widens and what narrows that produces accuracy, not any single filter on its own.

Start with the CPV codes, but do not stop there

CPV codes (called "Procurement codes" in Cobrief) are the foundation in most alert setups, and there is a reason for that. They are standardised, and contracting authorities use them on every contract notice. But they are also coarser than you think.

What we see in the suppliers we talk to is that they start with one or two codes they "think" fit, and then get surprised by what falls outside. There are around ten thousand CPV codes to choose from, and the same service can sit under several quite different branches. An IT security supplier can find their hits both under consulting services (CPV division 72) and business services (division 79).

It is worth spending some time browsing the CPV tree. Look at what sits next to the codes you first thought of, and test whether some of those also produce relevant hits.

Keywords and free text are good precision tools

If CPV codes give you breadth, keywords give you precision. This is where you can catch the specific projects that have not necessarily been given an obvious code.

For example: if you work in "smart building" or "building automation", you can catch notices that are CPV-tagged as pure construction contracts, but where the content is actually about automation. There is nothing wrong with the buyer's coding, but the keyword often hits more precisely than the code alone.

It is worth knowing how keywords actually work, because it surprises many people. Each keyword is evaluated on its own, and a notice matches if the text contains at least one of your words. So you cannot require that both "automation" and "smart" appear in the same notice by adding them as two keywords: instead you get alerts about everything that mentions one or the other. That also means more keywords widen the alert, they do not narrow it. If you want to be more precise, the trick is to use more specific words or phrases, for example "building automation" as a single keyword rather than "smart" and "automation" separately (a keyword made up of several words matches when those words sit next to each other). The narrowing itself comes from geography and the relevance score, which we get to below.

Geography: how much or little should the alert filter?

For many suppliers, the region matters. If you have to deliver physically on site, it makes little sense to get alerts about regions you do not operate in. Then it is fine to set a geographic filter from the start.

There is one point here that is easy to misunderstand: the geographic filter in Cobrief looks at the place of performance, that is, where the work will actually be delivered, not where the buyer has its office address. A buyer in Oslo can publish a contract to be delivered in Tromsø, and vice versa. You are choosing the place of delivery, not the buyer's location.

For others, geography is almost irrelevant. Digital services, consultant assistance, and national framework agreements should not be filtered out. Such notices are often not tied to a specific place, so you have to remember to include the "Not location bound" option in the geographic filter, otherwise they disappear. This is one of the most common pitfalls we see: a company narrows down to its own region, forgets the location-less notices, and only discovers months later that it missed a national framework agreement.

A good starting point is to be generous with geography at the start, and tighten it after you have seen what kind of hits you actually get.

What we see as common pitfalls

When we go through alert setups together with suppliers, the same things come up again.

The first is too many CPV codes. When the list becomes very long, the alert becomes so broad that it loses its purpose. You get alerts about everything in the same general category, and the important hits drown.

The second is the opposite: only one or two codes. It feels safe, but it almost always means you miss real opportunities that are coded slightly differently than you assumed.

The third is that the filters are never updated. A supplier who has launched a new service, or expanded into a new geographic area, often has an alert that still reflects where they were quite a while back.

And the fourth, perhaps the most underrated: they never test the filters before they start. It is worth looking at the recent stretch of notices that would have matched, before you let the alert run. Then you quickly see whether the filters are too broad, too narrow, or just a little off.

How alerts get better over time

You will never get the filters perfect on the first attempt, and you should not expect to. What we see working best is treating the alert as something that is fine-tuned over time.

A simple move is to briefly note which hits you open and spend time on, and which you discard immediately. After a period of actual use, you have a small data set that tells you where the filters hit well and where they miss. Then the adjustments become concrete.

For those who want an easier path, the tender alerts in Cobrief are built around a company profile. They give each tender a relevance score based on how well it fits your profile, that is, the CPV codes, keywords, and buyers you have prioritised. The score can be used in two ways: to sort tenders with the most relevant on top, and to set a lower threshold so that tenders below a certain score are not alerted at all. That lets you dampen the noise without having to change the codes and words themselves.

Similarly, our tender search feature has the same filters as the alert, so you can test setups against historical data before you put the alert into production.

Closing thoughts

A well-built tender alert is one of the few investments in tender work that pays back every single week for a long time. It just takes a bit of work to set up properly, and a bit of fine-tuning along the way. By the way: at Cobrief, tender alerts are completely free. You can read more about how Doffin fits into the wider procurement landscape in our article on Doffin.

We at Cobrief are happy to sit down with you and go through your setup. Just reach out if you have questions.