Thumbtack Engineering Blog - Latest Comments

Re: A Data-Driven Approach to Improving our Customer-Professional Matching

Jim Griffin — Mon, 11 Jul 2016 10:57:50 -0000

As a vendor, it has never been the number of quotes we receive, but always about the quality of the leads. If a request has more than 1-I don't know, it's an automatic rejection because they are doing an impulse request & have not given any thought to the parameters of their need. Thumbtack has failed miserably in reducing stupid requests or requests from stupid people (hard to ascertain which we are dealing with).

As a vendor, I have repeatedly attempted to get actionable information generated by the client questionnaire process. Relying on emails & direct contact with Thumbtack reps who "take my suggestions under advisement" has been totally ineffective. We expect a 2-8% response rate on requests generated by on-line referral services. Thumbtack's response rate is currently under .01% for those leads which we view as rational requests. If we were to include those leads we receive which are filled with "I don't know" responses or those requiring vehicles not in our fleet, the response rate drops to .0001%. Hardly worth our time to even view.

No one seems to understand, nor care, there is no "1 size fits all" questionnaire that will supply the correct information needed by vendors in different industries, in order to deliver a viable quote.

The transportation industry requires several specifics in order to determine the viability of a request.
1. A beginning location & an ending location
Inadequate addresses means we are unable to figure transit time which impacts the quote.
2. A specific time frame with a minimum time requirement
Inadequate info time-wise means we are unable to quote. Every vehicle in our fleet has a minimum time frame. For someone to request a 1 or 2 hour trip is an automatic rejection & tells us they are ignorant or cost sensitive, thus not worth quoting.
3. A specific head count (or range) in order to quote
Vehicles in a fleet & federal regulations determine viable client requests. Unless a head count is delivered in a specific way, we are prevented from delivering a quotation. Multiple times, Thumbtack repeatedly has been urged to change categories: 1-4 people, 5-9 people, 10-14 people, 15-24, 24+
These categories actually match the industries fleet offerings & would guarantee we don't quote on requests we are not legally equipped to service.
4. A means of clarifying inadequate info without being charged for a lead.
5. Thumbtack needs to reject insufficient or inadequate responses, forcing requestors to answer all the questions.
5. Supplying contact info for the individual in order to follow up.

Until the majority of these points are addressed by Thumbtack, vendors will continue to receive & reject requests. We will not pay for leads which don't fulfill the requirements we need for viable information & will continue to "Trash" those crappy leads.

Re: When Randomization Is Not Enough: Improving Sample Balance in Online A/B Tests

vladmalik — Fri, 26 Jun 2015 13:41:58 -0000

In a nutshell, are you proposing that: (1) You have prior information about how various segments of your visitors perform on a given metric. (2) You run an A/B test normally. (3) Check to see if the Control and Test groups are randomized with respect to some key segments. (4) If there are imbalances in the randomization, you adjust for it. Do you propose to adjust results of the test using the information you have about the baseline performance of the segments? Is this what you mean by "when each user interacts repeatedly with our site, we can proactively seek to balance out users based on historical data." Why "repeatedly"? (In the context of a conversion, I assume a user interacts just once, and the result is a pass or fail)

In your simulations, I'm not clear on when "rerandomization" happens. Do you mean that you kept testing for some specific randomization scheme and eventually found one that resulted in balance ? Is that the "seed" you're referring to? Then I take it you retested it 100 times and the "after" charts are the result of that? Or did you instead run 100 tests normally and then adjust each of them based on historical data about their sub-segments, and that's what we're seeing?

PS. Might help to label the X axis on the charts as "confidence interval" or "effect size". Not obvious at a glance.

Thanks.

Re: Interviewing for Female Friendliness: Tips from an Engineer

Katie Thomas — Wed, 05 Nov 2014 16:39:24 -0000

Thank you so much!

Re: Interviewing for Female Friendliness: Tips from an Engineer

Jonah Price — Wed, 05 Nov 2014 13:37:25 -0000

Very Insightful, Katie! I'll be incorporating these questions into my own interview process next time I'm on the market. This is the kind of place I'd like to work.

Re: Interviewing for Female Friendliness: Tips from an Engineer

Christine Negroni — Wed, 05 Nov 2014 01:45:53 -0000

I think this post is full of thoughtful and creative ideas. Looks like Thumbtack made a great hire when they hired you!

Re: Thumbtack Engineering

Robert Zaremba — Sat, 18 Oct 2014 18:02:34 -0000

go-vendor doesn't resolve dependency versioning problem (godep, gom and many other are much better in this field)

Re: Working at Thumbtack, the New Guy's Perspective

Omar Reeyaz Thanawalla — Thu, 03 Jul 2014 02:29:05 -0000

Wow this was an awesome read. Thanks Tommy!

Re: Thumbtack Engineering

paulyoder — Tue, 01 Jul 2014 10:42:43 -0000

Building an Angular directive from scratch is also one of the best ways to truly understand how directives work.

Re: Thumbtack Engineering

Alexander Kojevnikov — Mon, 30 Jun 2014 12:52:33 -0000

That makes sense, thank you Carl!

Re: Thumbtack Engineering

Carlana — Mon, 30 Jun 2014 08:38:54 -0000

The reason that the first argument in Go’s range syntax is the index is that for _, b := range array makes a copy of b. This makes sense if you think about it, because all variables in Go are memory slots with copies of values in them, but when you already have b in an array, it doesn’t really make sense to copy it out of the array unless it’s trivially small (pointer or single simple value).

Re: Thumbtack Engineering

Paul Ferguson — Sun, 29 Jun 2014 11:25:00 -0000

Cheers for sharing, been looking for just such an autocompleat. Interesting to see how others are solving these thing in Angular as it's still new to me.

Re: Thumbtack Engineering

Alexander Kojevnikov — Fri, 27 Jun 2014 17:12:22 -0000

Thank you Tom! Regarding slice bounds, it's just slightly more verbose when you want want to limit your slice to max N elements. Instead of `array = array[:N]` you now need to write an `if`. Not a deal breaker though...

Re: Thumbtack Engineering

Tom Fuller — Fri, 27 Jun 2014 16:53:00 -0000

Good read. This iterates over array values, ignoring the indices:
for _, b := range array

I too would like a ternary operator. For going off the end of an array, though, I prefer a panic since that's usually a bug that could lead to other problems.

Re: Thumbtack Engineering

lookfirst — Tue, 24 Jun 2014 18:47:08 -0000

I have a new moto for your company:

"not invented here" -- it's alive and kicking at thumbtack!

Re: Thumbtack Engineering

Katie Thomas — Tue, 24 Jun 2014 14:51:38 -0000

Hey lookfirst, thanks for reading!

We want to minimize our dependencies on third party libraries and we already use AngularJS. We couldn't find any good implementations in Angular at the time, so we wrote our own.

Re: Thumbtack Engineering

Katie Thomas — Tue, 24 Jun 2014 14:50:38 -0000

Hey Jakub,

That's a good point! Smarty was originally intended to be used
internally, and we don't use the smarty templates for php. The shared
code is prefixed with angular (angular-smarty) which helps differentiate
it in external code. If anyone wanted to use it and change the name
it would be easy for them to do so.

Katie

Re: Thumbtack Engineering

codedungeon — Fri, 20 Jun 2014 18:15:21 -0000

Agreed, when I first saw it, that's what I thought of first.

Re: Thumbtack Engineering

Jakub T. Jankiewicz — Wed, 18 Jun 2014 04:07:58 -0000

Smarty is probably not a good name, there are already smarty templates for php.

Re: Thumbtack Engineering

lookfirst — Tue, 17 Jun 2014 19:17:47 -0000

Shrug, why not just wrap twitter typeahead and save yourself some coding time? It's a trivial amount of code to do so.

Re: Thumbtack Engineering

Terrence Andrew Davis — Mon, 05 May 2014 23:31:29 -0000

My compiler compiles 50,000 lines in half a second. My AOT code is 50,000 lines and my JIT code is 50,000 lines. I compile everything, all the time. I have a script that checks document jump-to-links that takes a long time to run.

Re: Food rules for startups: eight delicious ways to build a better company

mjankowski_XXX — Sun, 28 Jul 2013 17:05:59 -0000

I used to cook in our startup a while ago. The CEO didn't approve of that :-)

Re: Food rules for startups: eight delicious ways to build a better company

Kay Richardson — Wed, 05 Jun 2013 06:28:34 -0000

This was a great article and an innovate way for companies to build better relationship with their team and not just an "employee" relationship. Most companies who are not creating the opportunity for communication will lose out on the insight of employees to present new and innovative ideas that can improve sales, branding, products and productivity in the I. Which results in more revenue. What is your company really doing to welcome innovation and creativity in your organization? I've worked with companies who have missed the boat to obtain additional revenue streams because the idea wasn't presented by management. The competition was 10 steps ahead which resulted playing catch and being a recognized brand in the community who sought their services. Don't miss out on the forward thinkers or the team members who are not management to add value.
Great innovators are always thinking when you're not.

Re: The usability experts were right: 5 changes we made to increase conversion by 50%

Matt Sharper — Thu, 15 Nov 2012 17:57:21 -0000

awesome blog post

Re: A primer on Python decorators – Thumbtack Engineering

swvist — Fri, 15 Jun 2012 12:01:00 -0000

Very lucid explanation! Thanks :)

Re: A primer on Python decorators – Thumbtack Engineering

Mike McFarlane — Wed, 06 Jun 2012 07:02:22 -0000

Hi, just started learning Python at Audacity (so much fun) and found this article in Hacker Monthly. It's well above my 'pay grade' but I thought I would try some of the code anyway. Thanks for writing this article.