The Fiverr 'Impressions' algorithm - analysis

[desmond_aubery]

The Fiverr 'Impressions' algorithm definitely needs a great deal of work. The profiles are spasmodic to say the very least.

A stable algorithm needs to manage inputs (internal drivers) and disturbances (external drivers) well.

Fiverr - this poor algorithm may be part of the reason for present system instability. Imagine this effect amplified across your large seller-buyer base. Ouch!!! 

[desmond_aubery]

AI assisted review of the above plots.

Unfortunately, I can't definitively analyze the stability of the algorithm from the plots you attached because the time scale on the horizontal axis is missing. This makes it difficult to assess the frequency of the fluctuations we see in the plots.

However, I can say that the plots do show significant volatility in impressions over time. There are periods of sharp increases in impressions followed by periods where impressions fall. This could indicate that the algorithm is sensitive to inputs and disturbances.

Here are some possible reasons for the volatility in impressions:

• Changes in user behavior: The way that users search for gigs on Fiverr may change over time. If the algorithm is not able to adapt to these changes, then it could lead to fluctuations in impressions.
• Changes to the Fiverr platform: Fiverr may make changes to its platform that affect how gigs are displayed in search results. These changes could also lead to fluctuations in impressions.
• Competition: The number of impressions that a gig receives will also depend on the competition from other gigs. If there is a sudden increase in the number of competing gigs, then this could lead to a decrease in impressions.

It is also important to note that not all fluctuations in impressions are necessarily a bad thing. Some volatility is to be expected on any platform that is constantly changing. However, if the fluctuations are too large, then it could make it difficult for sellers to plan their business.

If you are concerned about the stability of the algorithm, then you can try contacting Fiverr support for more information.

[emmaki]

It's because there are too many sellers and Fiverr has to give everyone an equal chance.. In theory, this sounds great, in practice, it's not really so great due to the high level of unqualified sellers on the platform who need no access. It also means that - for now - people who have been on the platform for years, even over a decade, are struggling to find buyers, even if they have 5-star profiles and a 10 success score. Of  course, the decline in buyer numbers (300k in the past year) may also explain this, but there are still 4m active buyers. The question really becomes "how many sellers are there?" - but Fiverr does not disclose those numbers. 

Either way, it doesn't really matter much how talented you are, how experienced you are - it's all very much the luck of the algorithmic draw. 

Not much about Fiverr makes sense at the moment. If you look at Contra, for example, access to their marketplace is regulated by either paying them or bringing in clients. This almost certainly stems the "wrong" sort of sellers from joining. 

I should add that I do support the "new seller boost" in general as new talent needs to be found in order to flourish. The problem now (I think) is that with AI, there are many sellers who are not particularly talented who are using AI to fake everything from their communication skills to the final delivery. Can AI detect that? Questionable. Even if it could, Fiverr's bizarre non-disclosure policy with regard to the use of AI means that the platform actively supports this, intentionally or not.

And then we return to the question of the buyer exodus....

It's an omnishambles. 

 

[desmond_aubery]

54 minutes ago, emmaki said:

It's because there are too many sellers and Fiverr has to give everyone an equal chance.. In theory, this sounds great, in practice, it's not really so great due to the high level of unqualified sellers on the platform who need no access. It also means that - for now - people who have been on the platform for years, even over a decade, are struggling to find buyers, even if they have 5-star profiles and a 10 success score. Of  course, the decline in buyer numbers (300k in the past year) may also explain this, but there are still 4m active buyers. The question really becomes "how many sellers are there?" - but Fiverr does not disclose those numbers. 

Either way, it doesn't really matter much how talented you are, how experienced you are - it's all very much the luck of the algorithmic draw. 

Not much about Fiverr makes sense at the moment. If you look at Contra, for example, access to their marketplace is regulated by either paying them or bringing in clients. This almost certainly stems the "wrong" sort of sellers from joining. 

I should add that I do support the "new seller boost" in general as new talent needs to be found in order to flourish. The problem now (I think) is that with AI, there are many sellers who are not particularly talented who are using AI to fake everything from their communication skills to the final delivery. Can AI detect that? Questionable. Even if it could, Fiverr's bizarre non-disclosure policy with regard to the use of AI means that the platform actively supports this, intentionally or not.

And then we return to the question of the buyer exodus....

It's an omnishambles. 

 

In my particular field (Thermal Engineering), you can arguably count the number of competitors on one hand in Fiverr space - so, market flooding is unlikely to apply.

The Fiverr algorithm should follow a more smooth trajectory, with slight, gradual swings over time due to seasonality, economy, world events - as 'disturbances'. Spasmodic swings seem to predict a poorly-tuned, or designed, algorithm.

If they are letting AI drive this for them - then - a huge ooooooops!!! AI will never have had sufficient training input/output data on the Fiverr platform to manage this adequately.

[donnovan86]

31 minutes ago, desmond_aubery said:

In my particular field (Thermal Engineering), you can arguably count the number of competitors on one hand in Fiverr space - so, market flooding is unlikely to apply.

No, but there can be AI as you said or just less demand for the time being. Or people searching on other platforms as well. 

[desmond_aubery]

14 minutes ago, donnovan86 said:

No, but there can be AI as you said or just less demand for the time being. Or people searching on other platforms as well. 

How would/could this explain the spasmodic (jerky) nature of the plots?

In a well-designed model, the swings would be far smoother.

Edited May 23 by desmond_aubery
Text added...

[emmaki]

Oh, the jerky graphs are the same for everyone. It's a pretty useless metric anyway as it only tells you how many times your gig was shown across the site to different people. At best it might be useful to identify if your gig "SEO" is working (or not). The algorithm is personalized and only showing your gig to people it thinks might be interested in your service. 

Nicheness may increase jerkiness, as not many people may be looking for your specific service everyday.

Mine is flatlining at the moment, but that's because I'm out of office 🙂

[desmond_aubery]

26 minutes ago, emmaki said:

Oh, the jerky graphs are the same for everyone. It's a pretty useless metric anyway as it only tells you how many times your gig was shown across the site to different people. At best it might be useful to identify if your gig "SEO" is working (or not). The algorithm is personalized and only showing your gig to people it thinks might be interested in your service. 

This appears to be a 'measure' of Fiverr marketing effectiveness for my individual Gigs. Spasmodic... 🤒