Search is a Team Sport
I failed doing search alone. But reflection revealed why!
Learning the Hard Way
In the very beginning of my Search career, I was given a task to revamp the Autocomplete for an e-commerce company. They way it was approached was that one of the “Product Manager” created a google doc listing down a bunch of things that were broken. It was essentially a list of bugs. After creating that doc, he checked out mentally from the problem and restricted his role to just QA whether all the bugs were fixed. There was no attempt to make a mental model for the bigger picture. I was wearing many hats — search and relevance engineer, data engineer, pipeline engineer and so on. This is how a job in a startup looks like but I did not get too far. In retrospect, the reason was that search as a problem needs to stay in your brain’s RAM. You need to hold a big mental model in your head to make things progress in a coherent manner. Sometimes, it’s that big that it needs multiple people to remember various aspects of Search. It’s nothing but horizontal scaling on human brains. People thinking about different aspects of search can then debate and science the shit out of the problem at hand.
After the above experience, I was paired up with a different PM who’s only job was Search. Search CTR was his only Key Performance Metric. He was good at communication skills and a good team worker. Another young and smart engineer joined us; she was willing to take any challenge, learn anything and fill any gaps. CTO of the company personally took an interest in the problem and made this his (one of the) highest priority for a few months. Business people were anyways emotional and were ranting about how the search was underperforming and bubbled up the use cases that they were most unhappy about. We had enough motivation and resources to get things working. We were given support from Devops and Data Team. If you believe me, the speed at which Search improved just skyrocketed. We made frequent releases and pushed CTR upwards by a single digit every month for a year.
Qualitatively speaking, autocomplete and Search results look beautiful now!
In retrospect, these were the people and teams who played a direct or indirect role in helping us develop search at a Blazing Fast Speed.
Data Engineering Team
- Provided Historical Data. Used for feedback loops.
- Provided Search Performance Metrics => Continuous improvement.
- A/B Testing & Lean Development
Relevance Engineer
- Design Analyzer Chains & tune other params in ES
- Bring an engineering perspective to what is feasible using Elasticsearch or Solr
Product Manager
- Brings Business perspective to the table. Aligns search strategy to business Strategy.
Machine Learning Engineer
- To bring smartness to search by Leveraging ML techniques
- Query Understanding, Named Entity Recognition, Learning to Rank etc.
Engineering Manager
- Search is never perfect; it’s a moving target. Someone with mental maturity to decide when to stop. Someone who could embrace lean methodologies and execute iterative development of Search and as result to be able to ship regularly in small batches.
Apps and Front-end
Android, iPhone, website, mobile website etc made the changes in UI to make search and autocomplete a First-class Citizen of the whole ecosystem.
Whole company
Different people in company will have a different expectation from search and can provide very interesting Vantage points. When so many eyes are looking at things, no bug can remain hidden for long.
Conclusion
Search is Team Sport. Don’t do it alone. Assemble a force and then launch the invasion!
