When Product Management meets Data Science

Lessons from a small NLP startup on applying product management principles to a data science based startup.

Transcript

1. When product management meets data science.
2. You fraud! You said you were from the ultracool Techstars?
3. What is Stitched? Stitched Timesheets Employee Lists Profiles & CVs Projects & Proposals
4. Break the problem down • What might be connected? • Machine Learning • Python • Suggestions • What is connected? • Neo4j • Java • Facts • Show the connections • All the Javascripts • Node + React • Simple Insight
5. Data Scientist vs Product Manager • Traditionally “Researcher” vs a “Project Manager” • The best product managers are really data driven • The best data scientists are really user centric
6. Inside the mind of a Product Manager Is that safe?
7. • Inspired: How To Create Products Customers Love
8. Jobs to be done • The Innovators Dilemma • A framework for making product decisions • https://www.intercom.com/books/jobs-to-be-done • A job that exists already • Your software takes this job and is 10x Better • Constantly ask yourself if your software is doing this job? • How easy can you make it? • Do you really need a neural net?
9. Evaluating what good looks like • They nail a ‘Job to be Done’ • This is why Siri SUCKS! • They are wonderfully simple • Nobody cares about how the sausage was made • There are no “options” or “configurations” • You can improve them without people noticing
10. Live Demo Is that wise?
11. A hybrid of tactics to do the job • Typeahead = Machine Learning + Graph • Search • Lots of blended techniques that have weight • Elastic Search • Time based Algo • TF/IDF • Closeness • Cached so very, very, fast! • Document Similarity • w2v results with graph results • Clustering of skills • kmeans cluster of w2v terms to autosuggest groups
12. word2Vec King - Man + Woman = ?????
13. word2Vec Doctor - Man + Woman = Nurse!?!?!
14. Is it ok to be “A little bit sexist?” Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings : https://arxiv.org/abs/1607.06520 Doctor - Man + Woman = Nurse!?!?!
15. NO! Is it ok to be “A little bit sexist?”
16. Things to remember • Take away a job to be done • But really study that job • Every ‘Feature” is really a roadmap • Choose the easiest vs the sexiest • Talk to a product manager or god forbid even the users • Brutally timebox experiments • Be prepared to explain your model - GDPR • Check you didn’t unleash the singularity • Remember you’re building a product not a research project • Take risks, but explain this carefully to investors!!!
17. Thanks, and questions? Stitched - Bristol @mrmattwright www.stitched.io