We have rolled back a feature for the first time, and that's alright

We have expressed time after time that our approach to product delivery is all about splitting the big tasks into smaller features and ship value continuously. Likewise, we are always fighting against the feeling that the code that is just about to be deployed must be improved a bit more before reaching our community. I'm sure many of you can relate to that.

Despite all that, we try to ship minimum lovable products that will be further expanded thanks to our community's use and feedback. We sometimes still deploy too early. And, sometimes, too late :-).

But, we had never had the impression that we could not easily fix something that was in production... Until now. One of our features has been too problematic to keep it live, and this post aims to tell you what happened and what we learned from it.

First shot, first miss

Converting your Manfred profile in the SSOT (Single Source of Truth) for all your professional data is not an easy task. It is obvious that we are late into the market in terms of being an app where you can host and display your professional data. For that reason, it seems reasonable to focus on joining forces with some players of the market, when possible.

We feel that to provide our Manfredites with as much value as possible, we need to integrate Manfred with other big job platforms that have been running for years.

LinkedIn was the obvious first choice, as we are aware that a most of the Recruiting sector uses the data and features offered by the American corporation. A few tech companies detected business opportunities there, and they developed solutions to address many of them. However, as we have done many times in Manfred, our approach was different to the already-existing one in the recruiting sector.

As a summary, we were building an importer to allow users to get their LinkedIn data into their Manfred profile in just one click. To do so, we tested several external services, and selected one that showed good results while we kept an eye on the others, so we could fill any mistakes made by our chosen one. After a few weeks in production and many more running tests, we have found some flaws that explained the limitations of the final product.

What happened? Why have we rolled back our LinkedIn importer?

• These scrapers could only work with public profiles, and many members of our community do not even know if their LinkedIn profile is public. We found difficulties explaining this to manfredites and circumventing this issue.
• Whenever a profile was available in English, these tools selected it and imported the information, even though it was not the main language of the profile. As the language preference could not be chosen when importing the information, some profiles resulted in a weird mix of different languages.
• Anti-scraping measures by LinkedIn should be taken into account too. Many of the big job platforms we are trying to integrate with invest lots of money in keeping their users' data controlled. Despite us having some precautionary measures, there is still room for improvement in this matter.
• Timeouts were also a problem. We set a 2-minute timeout on our side, but many of the tools used required more time, around 3 minutes. Either because of a long queue or because these scrapers could not fully access the information, many manfredites we left unassisted.

All of this resulted in a feature that, as the days passed, went below 50% success rate, while our testing numbers were much higher. We could not put our Manfred community through a product that works just 50% of the time. Therefore, we have rolled back this feature, and are changing our approach.

We have missed our first shot, this only means we are a step closer to the final goal.

What to do with LinkedIn Importer now?

This is a core feature in our business, so we'll treat it as such. We are determined to help our manfredites manage their own professional data easily; and to avoid becoming hostages of certain platforms.

We are creating our own team and putting integrations among the top priorities of our long-term roadmap. For now, we will apply what we have learned and coded in the last weeks into our own LinkedIn Importer. So we are not starting from scratch, and we are aiming high.

As it has always been in Manfred, our community is one of our strengths, so we asked for help. And that call has been answered. Some brave members will help us test our LinkedIn Importer and free your data.