Ask HN: How to do online jobs better?
I have had a long-simmering desire to create a new kind of website for job-seekers. It is due largely to my experience of how demoralising and ineffective traditional job sites feel.
I really want to avoid the existing model which has job seekers trawling pages and pages of jobs and then shooting off tons of CVs and flooding the job-posters with irrelevant candidates.
I am hesitant to just slap on a 'social' type interface where you can connect using linkedin/facebook and refer friends etc but I have hit a mental brick wall about what other options I can consider.
Would love to hear from HN'ers some new angles or approaches to consider. Appreciate any and all input.
21 comments
[ 4.5 ms ] story [ 56.1 ms ] threadFor job-seekers, the marginal cost of one more application is too low, so companies get too many applications. With too many applicants, filtering becomes more random, so the reward for applying for a job you are only tenuously qualified for increases. It's a classic hawks vs doves with a big payoff for being a hawk amongst mostly doves
So, make it more expensive to apply for a job. Two ways I can think of doing this:
1. Charge applicants ten bucks to apply for a job. Have companies report if they were good enough but they found someone better. In that case, give them a refund. In all other cases, keep the money. (EDIT: This may possibly be illegal in some jurisdictions, but there are doubtless ways around that.)
2. Show companies how many jobs the applicant has applied for recently (not necessarily where they have applied, but you could perhaps positions/business areas) and/or restrict applicants to N applications per month. Aggressively detect and ban sock-puppetting — perhaps you can require verified ID of the applicant's name and get companies to check that is the name of the person applying.
You could also make things less stressful by operating on a say weekly or monthly cycle. Jobs are posted at the beginning of the month, applicants choose which N jobs they want to apply for, applications go out, new cycle starts. This avoids stampedes and requires applicants to be discerning.
Your core idea might be a good one, but companies have better ways of making applying more expensive. They have puzzles and automated assessments. Some places basically only hire people who can network in, others require a specific certification, and many bias towards people who have some public work.
Recruiters literally have no idea of what makes a decent candidate so they have to resort to simple (almost syntactic) measures - 6 years of C++, Oracle etc. where they have no idea of what those terms actually mean or how useful they are (what Oracle product - there are zillions of them).
I'd quite happily delegate filtering of candidates to people I trust and if they get paid for doing it then it is a win-win situation. If you had a site with effective workflow to manage each stage of the process then it could be pretty slick.
[NB I spent quite some time on this idea a while back but parked it to work on other ideas].
Opening those databases up to third parties might be a risk many of them are unwilling to take.
The cost of advertising (which is high) is amortized by the long term value created by the CV database. Giving third parties access to those CVs is something I would imagine most recruiters would be reluctant to do.
However one of the big issues with innovating in this space is that HR departments are generally quite conservative so it's hard to sell innovative products to them (look at the difficulty SnapTalent had with their initial product).
You'll probably have more luck selling to third party recruiters who are more risk tolerant.
There's about a dozen companies currently doing the whole social network recruiting thing, but none of them has serious traction, and I'm not personally convinced it's a model that will work.
I'm working on this problem by building a niche job site, it's not hugely innovative technically, but it should help with the relevance problem and also be traditional enough to appeal to HR departments.
One of the other ideas I had in this space was building a platform for matching recruiters to candidates, so recruiters could send targeted messages to candidates matching a set criteria (+using game mechanic techniques to stop spamming). If you want to talk more about this idea feel free to drop me a message.
One model might be a curated niche of job postings for whatever niche you would understand the most. Then you need a way to reduce the talent pool the HR department sees.
One way to do this is give the job applicant a way to do a 'quick pitch' and the HR department can click a button called 'Tell me more' and the jobs process can go from there if both parties are interested. Maybe you could display stats as a way to incentive certain actions. If a job applicant has applied to 300 job offers with a cut and past and has only got 3 tell me mores, that would show to to the HR applicant. If however, a job applicant had a 98% tell more more rating on 30 job applications that means he's probably customizing it and the HR department can tell he probably has a pretty good EQ.
Should be fairly easy to solve this. You can only apply for jobs through a form that appears once you vote up. Once you apply your vote is fixed.
The main problem with job sites is matching jobs to applicants. As far as I know noone has gone beyond simple keyword searching. If you had an actual userbase you could make use of collaborative filtering techniques (like Amazons recommendations) to provide a tailored service to each user. CVs, personal blogs and github accounts give you enough information to seed each users recommendations even before they start rating jobs.
What sort of collaborative filtering techniques? How would that work? I'm intrigued! Would love to chat further via email.
I used some pretty simple techniques: one bayesian filter to filter jobs from other posts in mailing lists etc, one bayesian filter to score jobs based on keywords. Both filters were trained by feedback from the console ui. The main problem is excluding site-specific keywords that distort the scoring (eg if a site with mostly crappy jobs includes its own name in the listing then even the good jobs will score low by association). A lot of job sites have manky markup so I also had a different scraping script for each site to extract text. All in all its only a couple of hours work. I've been thinking recently about extending it and adding a simple web ui, since finding freelance work is pretty time consuming.
> What sort of collaborative filtering techniques?
I didn't have any specific in mind but there are plenty of good machine learning books that cover different tecniques. If you don't already have a background in maths then 'Programming Collective Intelligence' is a good book to start with. 'The Elements of Statistical Learning' goes into a lot more detail but requires some basic maths.
> Would love to chat further via email.
Email is in my profile.
a) Ads posted by Recruitment Agents giving me no idea who the actual employer might be.
b) Ads with no sensible salary hint.
On (b), I don't mind there being a wide band (e.g. 25-60K, depending on experience), as at least that is a hint. What I really hate is "£Competitive" or "£Neg+Bens" and other such non-information.