Developing an attribution model from Day 1

  Written by Louis Gudema

Dave KrupinskiI recently interviewed Care.com CTO Dave Krupinski about how they developed a marketing attribution model and the value that it’s had for their company.

This was originally posted on the IDG Connect blog.

Louis: Please describe what is Care.com? What does it do?

Dave: Care.com is the largest online marketplace for finding care. We help families find child care, senior care, special needs care, tutoring, housekeeping, pet care and other care services. And we help caregivers find jobs and manage their careers. In addition to helping families find caregivers and caregivers find jobs, we also help families manage the care once they find it. For example, we provide them with tools and services to manage the financial relationship that they have with the caregiver, including tools for managing payroll and payroll taxes, as well as tools for making one-time payments from their desktop computer or mobile devices for Saturday night babysitting, one time or pet sitting — things like that.

Louis: And you’re one of the co-founders. Why was this of interest to you?

Dave: We saw that there was a clear need. There were four of us that started the company. I think we all had personal examples of the need for care in our own lives. I had two young daughters that we were looking for a caregiver for. There wasn’t a service like Care.com around really. And we envisioned it to be a comprehensive service, not just for childcare but for all types of care that a family might need. So we started it with four services: child care, senior care, tutoring, and pet care. Each of the four founders had a need for at least two of the four services at the time. And the four founders all came from a very mission-oriented company, so we were all thinking along the lines of providing services for families around a real pain point. Previously we worked for Upromise, which helped families save for college, which was a real pain point for families. So it was in line with where we had been. It was also a market that we already knew a lot about in terms of the audience that we were speaking to because it turns out that the main decision-maker when it comes to college savings is the female parent, the mom in the household. And overwhelmingly the Care.com user base is female. So it was a demographic that we were familiar with.

Louis: So is this just in the U.S.? Or how many countries are you in?

Dave: We’re now in 16 countries and supporting 7 languages.

Louis: How long ago did you start looking at attribution? What drew you to this as a question?

Dave: So, you’re talking about the attribution of acquisition efforts to various marketing channels. I remember that before we even launched Care.com we were building the architecture to be able to track our online acquisition activities and attribute members appropriate to those activities. So it really went into our technical platform from the earliest stage of the platform development.

Louis: What does that mean, that it went into the platform?

Dave: That the structure and facilities for capturing and tracking that information were part of the very early architecture of our platform. When I think about attribution there are a number of different ways to manage, or track, your attribution. When you have traffic coming to your website from sources that you control such as online display advertising, search engine marketing, where you control the ways in which the users come to your website, you can also track how they come to your website and capture information associated with those acquisition channels. But you need to put in place the right logic and capabilities in your platform for tracking that information and storing that information and reporting on that information. And that kind of functionality was put in from the very start.

Louis: So that’s different from the information that you would get from, say, weblogs or some other data?

Dave: You could implement this entirely using weblogs if you wanted to, and the right parsing of those weblogs and the reporting of data from those weblogs, but we put in place a higher level structure so we didn’t just have to parse the data from the raw weblogs that made it easier for us to track and report on it. It’s the same information that you can get from weblogs, assuming that you’ve also managed the information that you’ve put into the URLs that drive traffic to your site from wherever you’ve placed your marketing assets.

Louis: So you have something like UTM codes in your URLs?

Dave: Something like that, yes.

Louis: Did you develop personas first? You said “mom” is the big person in the family. But did you have different personas and were you finding different conversion paths for different personas?

Dave: We do have different personas. We think about things like, Is someone coming to our site a planner and thinking about things that they might need 3 or 4 months down the road, or are they coming to the site with an urgent care need because their full time caregiver just called up and said they can’t make it tomorrow because they’re sick. Those are very different use cases. And those use cases also cut across the different services that we offer. Because someone looking for child care and someone looking for senior care are going to be different. And they’re going to come to our site with a very different frame of mind. For child care, someone’s coming to the site with the frame of mind that maybe they’re thinking of having their first child and they’re very excited. It’s generally a positive frame of mind. And someone who is coming to our site for senior care, it’s often not a positive frame of mind. It’s often a crisis, it’s often a situation that’s getting worse and may persist for a long time to come. So we think a lot about personas. And we also think about, “Is someone looking for a full-time caregiver or a one-time caregiver?” And you can see on our home page today, when someone’s going through the initial enrollment process, “When do you need care? Is this something you need in the next week or the next month? Or immediately?” And we try to tailor the experience based on the type of care they need and how quickly they need it.

Louis: And what digital and physical channels do you use to market Care.com?

Dave: We’ve probably experimented with just about every channel that you can think of. Some channels we use more than others. Right now television is one channel that we use quite a bit. And then we work through a lot of online channels: display advertising, search engine marketing – and others.

Louis: Do you use magazine ads? Or direct mail?

Dave: We do. But to a lesser extent.

Louis: How were you able to track the activity across the physical channels and digital channels?

Dave: The biggest challenge, and the one that we spent the most time on, is television. It tends to be the case that anyone who does television works with an agency that buys the media, places the ads, they also have their own attribution models. So the agency that we work with, they ran attribution models against our ad placements. And those models tend to look at things like, based on weblog data and sign-up data, what is the immediate change to the baseline that can attributed to when a spot ran in a given geography. So, we’ll literally look at the activity to the website, both in terms of traffic and sign-ups, and we’ll correlate that with the exact time that different TV ads are run. There’s no perfect attribution model. So at the same time that the agency is running its own attribution model we’re also running our own internal attribution model. We compare the two models. And we have analytics professionals on staff at Care.com that approach TV attribution in probably a similar way to how the agency approaches it, but not in exactly the same way. So we do our own modeling.

Then we look at, for example, to parse out the effect of seasonality versus the effect of advertising. Like many companies, Care.com tends to do more advertising in high seasons. So when we see a spike in activity both in terms of traffic and sign-ups during those more active seasons, what it due to the seasonality effect or the effect of advertising? And being able to answer that question was one that took a lot of work on our side and a lot of different modeling approaches to fit the data into assumptions that we were making. And being able to go back over history and testing those assumptions against what actually happened ultimately allowed us to fine tune an attribution model. But it takes a fair amount of data and a fair amount of time over which you have that data to really build an attribution model.

Louis: What about other channels? Or newer channels?

Dave: I would say that it’s always evolving. The techniques and approaches that we may have had two or three years ago may be obsolete now because there are different marketing channels that may become more relevant. I think our attribution models maybe haven’t changed much but the types of marketing opportunities that are available to companies continue to evolve and continue to get better. For example, we have a channel like Facebook which wasn’t around when we started. And over the past few years lots of companies have tried to provide a solution for advertising on Facebook which provides an attractive ROI. Companies are still trying to figure that out. You always have to have some of your time and energy put into experimental and new channels that are being introduced into the market all the time.

Louis: What about mobile? Have you been able to track mobile well?

Dave: We are able to track mobile well. And we can distinguish traffic coming to our desktop site versus our mobile site. Mobile has been a big challenge for many companies. For the vast majority of companies conversion on mobile devices is lower than conversion on the desktop. And that is just another channel through which you now have to think about optimization and how you drive qualified users to your mobile channel. We see a high correlation between increases in mobile traffic and television advertising. Mobile is the second screen that’s there when you’re watching TV and seeing the ads. People are checking you out on their mobile device while watching TV.

Louis: What about issues around privacy in browsers, in the EU. You’re not allowed to do the same sort of things in the EU as you can in the US. What if someone has disabled cookies? Did those affect your ability to develop models?

Dave: In the EU, especially in the UK and The Netherlands, we need to provide explicit disclosures and enable the user to opt-in to the use of cookies. There’s a certain percentage of people who will not opt-in to the use of cookies, but I think that the vast majority do. And if someone doesn’t then we’ll just work with the data that we have.

Louis: How does brand equity fit into attribution modeling, and can you calculate the value of your brand?

Dave: Good question. We know our brand is highly valuable and probably provides a longer tail to our marketing activities because it’s an easy brand to remember.

Louis: Do you do much brand marketing, or do you do mostly lead generation-style direct marketing?

Dave: If you look at our TV ads, I think you’ll see that they’re a good combination of brand marketing and direct response. Yes, we do invest in brand marketing and our TV ads are a good example of that.

Louis: What about the on-site experience? Was site optimization part of this? Looking at landing page optimization or…

Dave: Very, very much so. We think about the end-to-end experience, not just the marketing activity but what happens when the user comes to our website. And we do a lot of landing page optimization and flow optimization to ensure that we’re getting the highest yield out of the visitors that we’re driving to the site.

Louis: So, were there industry standard or package software tools that you used for this, or have you developed your own tools?

Dave: We’ve developed our own tools across the board.

Louis: And I’ve heard you mention the value of the predictive value of the attribution model, that that’s how you can really tell if your model is correct. So you tested that with past data?

Dave: We tested it with past data, exactly. We went back and ran all of our models against the full history of data that we have. And then we also use it to project into the future.

Louis: So what have been the results? I don’t know if you can share any sort of percentage improvement in marketing results or anything of that sort?

Dave: It’s not strictly about an improvement in marketing results. It’s about visibility and transparency into where the business is going. So that we can more accurately plan the business, knowing that if we spend a certain amount, knowing what the return is on our spend is highly valuable to us from the perspective of being able to manage our business, plan for growth, and support that growth from a staffing standpoint. Having an accurate attribution model improves our ability to predict and our ability to manage.

This and 9 other interviews with sales and marketing leaders are available in this free ebook:

download sales and marketing best practices ebook

One thought on “Developing an attribution model from Day 1

  1. Pingback: CFOs and CMOs actually have much that they can work on together - revenue + associates

Comments are closed.