Informatica’s Franz Aman Puts Big Data To Work

“We all have to step up and use data to drive decisions, not opinion,” said Franz Aman, senior VP of marketing at Informatica.

Informatica’s Franz Aman Puts Big Data To Work

There’s a special B2B marketing team at Informatica, the Silicon Valley-based data-management solution provider. Led by Franz Aman, senior VP of marketing, this team has taken on the integration of CRM, marketing automation, and analytics systems, bringing the data from all three into a central “data lake,” as Aman likes to call it. They then slapped data visualization on top—and all sorts of interesting things began to happen. recently caught up with Aman to discuss what this data-marketing program has allowed them to do that’s different from their traditional marketing. You’ve undertaken a hugely ambitious, data-driven marketing strategy. Why did you feel the need for such a big change?

Aman: Two equal parts of conviction and inevitability—the conviction that it is just not interesting to debate opinions about marketing; arguing about what works and doesn’t work; finger-pointing between marketing and sales; struggling to defend budgets, etc. If it’s all about opinions, everyone will prefer their own. Don’t you?

The moment you have data, the discussion changes. It becomes constructive. Evidence-based. I wanted that.

Inevitability because of the trends we’re all staring at: more of the buying experience going digital every day; more channels, systems, and apps involved; less patience on the customer side and zero tolerance for a crappy customer experience.

Marketing is becoming more technical, and there is no way to avoid it. We all have to step up and use data to drive decisions, not opinion. Data is no longer a by-product of a customer interaction. It lets us understand their interests and intentions. As marketers, we’d better listen. Everything seems to be calling itself “big data” these days. What makes your approach a big data marketing program?

Aman: It is certainly not the volume of data we have; in big data terms, we’re working with tiny to minuscule volumes of data. But marketing data is fairly diverse and often not very structured. Big data technology—Hadoop, big data management tools, etc. —deals well with data of all shapes and stripes.

A big data approach also gives marketers self-service access to data like never before. What used to be a three-month IT project can now be done on the spot.

So it’s not the sheer volume of data. It’s the diversity of the data and the way you work with it. In your blog series, “Naked Marketing,” you talk about a 60-day sprint to get your big data marketing engine up and running. What happened in those 60 days?

Aman: We bought ourselves some Web service space and horsepower, installed Hadoop, used some of our own big data management tools, then poured data from a number of sources into one place and put visualization tools on top of it. For the first time, we had all the data we cared about in one place. Now we can see which Web pages are viewed across a customer company, who the individuals are in our database, etc.

Of course, doing this in 60 days would have been impossible if we hadn’t had our act together—keeping our data in good shape, using tagging and campaign- and source-codes, creating a breadcrumb trail across all marketing systems so that we could connect the dots and get end-to-end line of sight. Without that, it would have been impossible to tell which marketing channels and programs drive which leads converting into revenue. In addition to technology, it was essential having a solid marketing operations team to operationalize all that.

Without all the data focus beforehand, we could have never gotten this project over the line in just 60 days. What is a “data lake,” and how is it different from traditional marketing data warehouses?

Aman: I think of warehouses as highly structured, California Closet-style installations, and data lakes are magic closets you can dump anything into without ever worrying about which drawer or section something goes into.

Of course, having a nice walk-in closet with California Closet niftiness will always be well-organized, and you know just where to find those socks—it will be very easy to count all the red ones and size up the number of pairs of shoes.

But … every time you put something in, you have to adhere to the organizing principles you decided on when you built it. If you don’t, the value of the organizing principle goes to zero. Same in a data warehouse, where every bit of data is loaded into dimensions and fields as carefully crafted as a California Closet solution.

That is a “schema on write” model—deciding on your data structures before you build—and it works like a charm when you need penny-perfect answers; think financial roll-ups etc. If you want to store something new and different that doesn’t fit the current schema, you’ll have to call in your closet maker. Think of a three-month IT project to remodel a dimension in your warehouse.

The “magic closet” data lake, on the other hand, helpfully accepts anything you want to throw in it, regardless of shape and size. It has tremendous processing power to find anything you ask it to find—using Hadoop and cheap, redundant compute capacity.

This is “schema on read,” or making sense of data at the time of retrieval. It works beautifully for unstructured and unexpected data. But if you need to know the exact number of red socks you have, you’ll have to ask by using the correct magic spell. Otherwise you may get part of the answer.

But where “ish” is OK, having access to the data and being able to ask new and different questions at all times is a wonderful thing—no IT cabinet maker required! This model lends itself really, really well to solving marketing questions because, let’s be honest, we marketers come up with new questions faster than anyone could organize data for. What can you do now that you couldn’t do before you implemented the big data marketing program?

Aman: Having all the data we need in one place allows us to connect all the dots across the various marketing apps we’re using.

Every marketing application today has some reporting capability. That’s great to optimize whatever the app does, like optimize landing pages, email open rates, etc. But seeing beyond the trees and understanding the forest requires more: an end-to-end line of sight; knowing which channels and programs drive the most net-new names, pipeline, revenue, etc.

Once you have this amazing view, it becomes possible to take revenue or pipeline and divide it up among the marketing channels, programs, or offers. To look across multiple touches for individuals connected to an opportunity using a statistical model, instead of simplistic, last-touch attribution, we can analyze the value of programs, offers, etc., based on multitouch attribution.

One of the most important benefits is the agility and flexibility to ask new and different questions without needing an IT project. What used to take months can be done in hours. Data self-service is a wonderful thing. Granted, you need the right tooling to get data in and out.

Looking at all marketing interactions by account also allows for segmentation and targeting for account-based marketing (ABM), now all the rage in B2B marketing. Has your approach changed the relationship between marketing and sales?

Aman: Yes. Sales uses the data lake to do a lot of important things: mine prospect and customer interactions; find additional contacts in accounts they should pay attention to; discover customer interest in specific products; see the potential for upsell and growing deals; identify the members of buying teams in accounts so we can influence all the right people. … Seeing the pull and the interest spreading quickly across the sales team by word of mouth tells us that we’re really on to something. How do you figure out which tactics are working? What’s your approach to attribution modeling?

Aman: This is such an old question, but most marketers still don’t have definitive answers for it. Often teams will stare at last-touch pipeline and revenue reports and cut the wrong programs because they “don’t drive enough pipeline.” Last-touch reports will always disadvantage programs at the top of the funnel, although they may add tons of net-new names to the marketing database and therefore be the source of “fresh blood”—new logos and leads that over time will turn into the real deal.

We have experimented with different algorithms for attribution, everything from linear—not really realistic as not all touches are created equal—to blended time decay and position-based—older is less important and key touches, like the first touch that converts someone into the database, can be valued more highly—and survival models—coming to us from medical survival models applied to marketing tactics providing statistically representative answers. This is where data scientists can add a lot of real value. Do you need the IT department to be on board for this, or can a marketing team do it on their own?

Aman: I would strongly recommend that all marketers work with IT and not around them. Ultimately, there will come a point when marketers will want access to enterprise data that only IT has the keys to. Yes, it has become easier to get started quickly, even as a business-only initiative, but in the end data lakes should be company resources and managed as such. IT is great at that. What are the obstacles for any marketer embarking on the road to big data marketing?

Aman: First, marketing is becoming more technical by the moment, and the skill challenges grow more pressing. You’ll need people who have a passion for data and know how to work with it, make sure you can trust it—don’t forget QA—who can interpret it because they understand the business enough to detect when data doesn’t make sense before it costs anyone credibility.

Second, without creating the breadcrumb trails of campaign codes, tagging, lead IDs, etc.—and operationalizing all that—it will be impossible to bring the data together even if you get them into a data lake because you couldn’t “join” the data sets from different applications using a common ID or mapping of IDs. Where will you go next with this?

Aman: Since all of our data is coming together in the marketing data lake, it’s our best source for seeing behavior across marketing channels.

To create the best customer experience possible, we will drive segmentation and targeting from the data lake, so we get consistency across all channels. Nurturing across email, social, on the Web, and in retargeting should all be consistent rather than system by system, app by app. I’m expecting both a better prospect/customer experience and a lift in conversion rates. Any last words of advice for senior marketers who like the sound of all this but aren’t sure they have the people or resources for it?

Aman: Make sure you have the foundation in place, meaning you get your data act together. Otherwise, there is no point going there. Make sure you get big data management tools and data lake self-service tools that lower the skill-set bar.

Leverage IT and potentially other functions to pool resources. A data lake can easily support more than one function, and there’s typically benefit from having more complete data across all customer touch points across their entire life cycle.

Take some risk and have fun. You can do it.