Optimove Helps Optimize Customer Retention (And, Yes, It's a Customer Data Platform)

As I wrote last week, it sometimes seems that every system I look at these days is a Customer Data Platform. Of course, this is partly because I’m choosing to look at that type of system, and partly because CDP vendors are reaching out to me. But I do believe another reason is that CDPs are an idea whose time has come: I’ve recently seen at least three CDPs that are just emerging from stealth or beta mode. All were developed because someone else recognized the huge unmet need for getting  better customer data to marketers.

One of the vendors that contacted me was Optimove, a Tel Aviv-based firm that calls itself a “retention automation platform” but definitely fits the CDP criteria. This means that Optimove is a marketer-controlled system that loads data from multiple source systems, puts it in a marketing-friendly format, and makes it available to external marketing execution systems.

Like many CPDs, Optimove also includes a campaign engine that pushes specific marketing actions to the external systems. Optimove’s approach is unusual in basing its campaign interface on a calendar that lays out the campaign schedule for each user-defined customer segment. This makes it easier for marketers to build a comprehensive contact strategy from multiple campaigns.

The campaigns themselves each have their own schedule, allowing them to run once, daily, weekly, or monthly. Users can also limit the number of messages sent to each customer by assigning an exclusion period to each campaign. Other campaigns can be instructed to respect or ignore these exclusion periods, ensuring that high priority messages are delivered in all circumstances. Each campaign triggers a single action, which can be directed to email, banner ads, direct mail, Facebook custom audiences, in-app pop-ups, SMS, app message boards, call centers, or other channels. The connections may be through file transfers or APIs.

Optimove's campaign interface is unusual, but the system is even more unusual in taking performance measurement very seriously. Its standard campaign setup requires users to assign a success measure and to either set aside a control group or set up a multi-way split of alternative treatments. This enables standard reports, including the campaign calendar itself, to show the incremental value provided by each campaign – the critical information needed for long-term optimization. By contrast, most marketing systems make success targets and testing optional if they support them at all. Users can also see a history of all campaign results for a given segment, making it even easier to identify the most productive programs.

The campaign segments themselves, which Optimove calls target groups, are built by accessing data that Optimove has loaded from the client’s data warehouse and operational systems. Optimove has standard data models for different industries, reflecting its current customer base: online gaming (bingo, casinos, poker, sports betting, etc.), foreign exchange trading, and ecommerce. The system assumes the data has already been coded with customer IDs, which something that makes reasonable sense given the focus on retention rather than acquisition. 

Data is typically loaded daily or weekly. After each load, customers are assigned to life stages (typically, new customers, active customers, and churned customers) and to multiple segments based on behaviors and attributes, such as location, product preferences, and spending levels. The system then uses the life stages and segment attributes to assign customers to "microsegments" that cluster analysis has found will behave similarly. It’s important to understand microsegments represent a current customer state that will change over time: that is, each customer belongs to different microsegments at different stages in her life cycle.

Optimove calculates the probability of moving from one microsegment to the next and uses this to predict how a given group of customers will behave in the future.  This is the basis for its lifetime value and churn predictions – key metrics in system reports. This type of forecasting is something else that really should be done by every marketing system, but rarely is. Optimove also provides cohort analysis reports, comparing performance of customers who joined during different time periods. This is yet another important type of information that is not always available.

Optimove does have some limitations. I was surprised there are no standard reports to highlight attributes that separate responders from non-responders within a promotion audience: this is pretty common information that helps marketers to refine their segmentations and better understand what is driving results. Nor does the system current recommend the best action to take with individual or a group. Both features are being worked on for future release.

Optimove was founded in 2009 and currently has about 70 clients, mostly in Europe. It has a few U.S. customers and is looking to expand in this market. Pricing is usually based on the number of customers and begins around $2,500 per month.

Adometry Combines Attribution with Optimization

So…my last two posts on attribution systems (MMA and VisualIQ ) were among the least popular ever, right down there with Marketing Lessons from Chernobyl (which, let’s face it, was in pretty poor taste). But vox populi isn’t always vox Dei, eh? I think it’s an important topic, so here we go again.

The lucky recipient of that less-than-stirring introduction is Adometry, which in no way deserves any disrespect. From humble beginnings in click fraud prevention, they have grown in recent years to be one of the leaders in algorithmic response attribution. Their latest expansion moves them beyond digital channels to offline media including direct mail, television, and print. They have also moved from attributing past results to using predictive models to optimize current and future campaigns. Impressive.

The core of Adometry’s attribution methodology is to compile the sequence of marketing messages seen by each individual, and then compare results of individuals whose sequence differs by only one message. Any difference in results is then attributed to that message. This is conceptually simple, but requires clever treatments to handle low volumes for specific sequences and to isolate the impact of attributes such as placement, time slot, creative, and list segment. Adometry also lets users model against multiple events in the customer life cycle, such as sign-ups, first purchase, and repeat purchase. It calls these all conversions, which I personally found a bit confusing but suppose would quickly get used to.

The system also classifies each conversion as attributable, multi-touch, and multi-channel, depending on whether it was linked to at least one message (attributable), to multiple messages (multi-touch) and to messages in multiple channels (multi-channel). For each category, it shows the conversion count and revenue: so, for example, you see the number and revenue for multi-touch repeat purchases. That’s a lot of information to digest, but does give a great deal of insight into the effect of different promotions and channels on different parts of the business. This encourages marketers to look beyond any single measure, such as cost per order, that tells only a small part of the business story.

The system’s optimization process begins with the attribution analysis, but then adds auto-generated predictive models to estimate the impact of future ad plans, including interactions across channels. Users can enter scenarios with budgets for multiple channels and campaigns, and then apply other constraints such as limits on the change in spending per channel. They also define output measures for the system to optimize against: like other optimization systems, Adometry can only optimize against a single measure, but this can be a composite of several items. For each scenario, the system will determine the optimal budget allocation and show the expected results across each output measure. Users can modify the recommended plan and have the system re-forecast the results. The final plan can be output to a spreadsheet for further editing. Adometry can also be connected directly to ad buying platforms, including systems for real time bidding on individual impressions.   The company says optimization typically yields a 20% to 40% improvement in ad-to-sales ratios.

The database of marketing messages per individual can be used for other types of analysis. These include reach and frequency reports, which show the number of individuals reached in total, reached in each channel, and reached exclusively for each channel. The reports count impressions as well as individuals; show how many people were reached in each combination of channels; show the number of people with each number of impressions (one, two, three, etc.); and show the current member count in each funnel stage.

Adometry’s data comes primarily from tags embedded in advertisements, emails, and other online messages, which drop cookies to identify who sees which message. The system can also draw data from Web server logs or third party tags. Adometry can further enrich its database by appending external information about individuals, using both online and offline sources. This lets it profile the audiences associated with different events, channels, campaigns, and other attributes. Optimization models can use data that can’t be tied to specific individuals, such as weather, economic conditions,  and mass media like television and print. The system can also verify which ads were actually seen by individuals, providing more precise inputs to the attribution calculations.

Pricing for Adometry is based on the number of channels and volume of data. It starts around $100,000 per year for the smallest clients with enough volume to use the system effectively (about 30 to 50 million impressions per month). Currently, more than 50 companies use Adometry’s attribution services.

Marketing Lessons from Chernobyl

I’ll be speaking about optimization this Wednesday at the Online Marketing Summit conference in Santa Clara, CA. Since I’m very comfortable with the actual topic, most of my prep time has been spent looking for pictures for my slides.

One discovery was the image above, which shows is how I think most people imagine optimization: a team of dead-serious revenue engineers carefully tweaking dials and watching gauges until they find the perfect balance among alternative marketing investments. That the real world isn’t quite so rigorous is a sad truth I’ll cover during the conference.

But this picture isn’t just any power plant. It’s the control room at the Chernobyl nuclear reactor which disastrously exploded in 1986. Look closely, and what do you notice?

Yes, those hats. Apparently the Chernobyl plant was being run by pastry chefs. That explains so much.

My theory is this: the Soviets had a little-known tradition that translates roughly as “switch jobs with your friends day”. The year of the accident, a team of bakers decided to change places with their buddies in the Chernobyl control room. The nuclear engineers spent the day calculating the volume of pie tins and optimizing heat convection in the baking ovens. Meanwhile, the pastry chefs were decorating fuel rods with icing and asking, “What if we replace the reactor coolant with meringue?”

This did not end well.

Well, maybe that didn’t happen. But my imaginary pastry chefs sound a lot like stereotypical marketers: experts in a subjective field where decisions are based on taste, feel, and appearance, and progress comes through intuitive experimentation. Those methods work well in the kitchen, but can’t be safely transferred to a nuclear reactor. Nor do they work for marketing optimization.

Like reactor management, marketing optimization programs need to be based on deep knowledge of the underlying process. They rely on precise tracking mechanisms that support long-term monitoring of detailed results. They need to be run by marketing equivalent of nuclear engineers, not pastry chefs.

This doesn’t mean that data geeks should take over marketing. Chances are, things weren’t going very well in the Chernobyl bakery that day, either. The city needed both bakers and scientists. But having them wasn’t enough: they needed each in the right place. Marketing departments are the same.