Predictive Marketing Vendors Look Beyond Lead Scores

It’s clear that 2015 has been the breakout year for predictive analytics in marketing, with at least $242 million in new funding, compared with $376 million in all prior years combined.

But is it possible that predictive is already approaching commodity status? You might think so based on the emergence of open source machine learning like H20 and Google’s announcement today that is it releasing a open source version of its TensorFlow artificial intelligence engine.

Maybe I shouldn't be surprised that predictive marketing vendors seem to have anticipated this.  They are, after all, experts at seeing the future. At least, recent announcements make clear that they’re all looking to move past simple model building.  I wrote last month about Everstring’s expansion to the world of intent data and account based marketing. The past week brought three more announcements about predictive vendors expanding beyond lead scoring.

Radius kicked off the sequence on November 3 with its announcement of Performance, a dashboard that gives conversion reports on performance of Radius-sourced prospects. What’s significant here is less the reporting than that Radius is moving beyond analytics to give clients lists of potential customers. In particular, its finding new market segments that clients might enter – something different from simply scoring leads that clients present to it or even from finding individual prospects that look like current customers. This isn’t a new service for Radius but it’s one that only some of the other predictive modeling vendors provide.

Radius also recently announced a very nice free offering, the CMO Insights Report.  Companies willing to share their Salesforce CRM data can get a report assessing the quality of their CRM records, listing the top five data elements that identify high-value prospects, and suggesting five market segments they might pursue. This is based on combining the CRM data with Radius’ own massive database of information about businesses. It takes zero effort on the marketer’s part and the answer comes back in 24 hours. Needless to say, it’s a great way for Radius to show off its highly automated model building and the extent of its data. I imagine that some companies will be reluctant to sign into Salesforce via the Radius Web site, but if you can get over that hurdle, it’s worth a look.

Infer upped the ante on November 5 with its Prospect Management Platform. This also extends beyond lead scoring to provide access to Infer’s own data about businesses (which it had previously kept to itself) and do several types of artificial intelligence-based recommendations. Like Radius, Infer works by importing the client's CRM data and enhancing it with Infer's information.  The system also has connectors to import data from marketing automation and Google Analytics.  It then finds prospect segments with above-average sales results, segments that are receiving too much or too little attention from the sales team, and segments with significant changes in other key performance indicators.

Like the Pirate Code, Infer's recommendations are more guidelines than actual rules: it’s up to users to review the findings and decide what, if anything, to do with them. Users who create segments can then have the system automatically track movement of individuals into and out of segments and define actions to take when this occurs. The actions can include sending an alert or creating a task in the CRM system, assigning the lead to a nurture campaign in marketing automation, or using an API to trigger another external action. Infer plans to also recommend the best offer for each group, although this is not in the last week’s release – which is available today to current clients and will be opened to non-customers in early 2016. That last option is an interesting extension in itself, meaning Infer could be used by marketers who have no interest in lead scoring.

Mintigo’s news came today. It included some nice enhancements including a new user interface, account-based lead scores, and lists of high-potential net new accounts. But the really exciting bit was preannouncement of Predictive Campaigns, which is just entering private beta.  This is Mintigo’s attempt to build an automated campaign engine that picks the best treatment for each customer in each situation.

I've written about this sort of thing many times, as recently as this July and as far back as 2013. Mintigo’s approach is to first instrument the client’s marketing efforts across all channels to track promotion response; then run automated a/b tests to see how each offer performs in different channels for different prospects; use the results to build automated, self-adjusting predictive response models; and then set up a process to automatically select the best offer, channel, and message timing for each customer, execute it, wait for response, and repeat the cycle. Execution happens by setting up separate marketing automation campaigns for the different offers.  These campaigns execute Mintigo’s instructions for the right channel and timing for each prospect, capture the response, and alert Mintigo to start again. The initial deployment is limited to Oracle Eloqua, which had the best APIs for the purpose, although Mintigo plans to add other marketing automation partners in the future.

Conceptually, this is exactly the model I have proposed of “do the right thing, wait, and do the right thing again”. Mintigo’s actual implementation is considerably messier than that, but such is the price of working in the real world. There are still nuances to work out, such as optimizing for long-term value rather than immediate response, incorporating multi-step campaigns, finding efficient testing strategies, automating offer creation. And of course this is just a pre-beta announcement. But, it’s still exciting to see progress past the traditional limits of predefined campaign flows. And, like the other developments this week, it’s a move well beyond basic lead scoring.

Predictive Analytics: Should Automated Content Selection Work by Segment or Individual?

Two vendors made the same point with me this week, which is reason enough for a blog post in mid-July. The point was the difference between basing content selection on individuals and on segments. I have never considered the distinction to be especially important, since segment membership is determined by individual behaviors and individual-level decisions are guided by behavior patterns of groups. But the two vendors in question (Evergage and Jetlore) and another I spoke with earlier (Sailthru) were downright religious about the superiority of their approach (individual-level selections in every case). So I thought it worth some discussion.

First, let’s clarify the topic. The distinction these vendors were making is between selecting content separately for each individual and selecting the same content for all members of a segment. Of course, customers are assigned to segments based on their individual behavior and other attributes, but once someone is in a segment, the segment-based system ignores individual differences. Among segment-based systems, collaborative filtering uses product selections almost exclusively: this is the classic “customers who looked at this product also considered these products” approach, which doesn’t take into account other aspects of the customer’s history. Other methods build segments based on customer life stage, demographics, and similar broad attributes. It’s possible to build segments based on very detailed behavioral differences, but that’s likely to create too many segments to be practical.

At an operational level, the individual-level systems use automated analytics to rank all possible content choices for each individual using that individual’s data. Segment-based systems either use rules to select content for each segment or use automated analytics to rank content choices for the segment as a whole. The individual-level approach makes the most sense when there are many content choices to consider, as with retail merchandise or entertainment (books, music, movies, etc.). Those are cases where getting precisely the right content in front of the customer is much more effective than offering everyone the most commonly-selected items. Retail and entertainment marketers also usually have detailed history which supports accurate predictions of what the customer will want. Segment-based systems work best when only a few choices are available.  This means a separate segment can be created for each item or, more realistically, segments come first and items are created to serve them.

So does the entire debate really come down to using individual-level systems when there are lots of choices and segment-based systems when there are only a few? Not really: collaborative filtering can also handle massive numbers of options with great accuracy. The difference is that collaborative filtering doesn’t really consider much beyond a particular product choice, while a sophisticated individual-level system will consider other factors including the current context and the customer’s history. Done correctly, this should yield more appropriate selections. On the other hand, individual-level approaches require more data and more complex analytics, so there will be cases where a segment-based method is ultimately more appropriate.

Moreover, the two approaches are as much complementary as competitive. A segment can indicate customer state, such as just-acquired, satisfied, or churn-risk, which constrains the contents considered for offer by the individual-level system.* Or a segment-level system could chose the type of message to send but let the individual-level system pick the specific contents. Dynamic content within email campaigns often works exactly this way.

In fact, I’d argue that state-based segmentation is essential for individual-level optimization because states provide a framework to organize the masses of detailed customer data. Without tagging the customer’s current state during each event, it would be very difficult for even the most sophisticated analytical system to see the larger arc of the customer life cycle or to understand the relationship between specific offers and long-term outcomes.

All this has practical implications for marketers considering these systems.

- for individual-level systems, make sure they can look beyond predicting the highest immediate response rate to measuring impact on long-term objectives such as conversion or lifetime value.

- for segment-level systems, make sure they can take into account the customer’s past behaviors and attributes, not just the products they have recently purchased or considered

- for all types of systems, assess how they track and guide the customer through the stages of her long-term journey

You'll want to consider other differences between content selection systems, such as number of items they can manage, how quickly they return selections, how they incorporate items without a sales history, and what data they consider in their analysis. Just remember that making selections isn’t an end in itself: you want to make choices that will create the greatest long-term value. To do that, it’s not a choice between individual and segment level analysis. You need both.

____________________________________________________________________________
* See my June 25 post for a more detailed discussion of state-based campaigns.

Will Machines Replace Marketers? Artificial Intelligence Isn't Ready Yet But Watch Your Back

Anyone who has chatted with me in recent months knows that I’ve added the impending domination of humans by intelligent machines to my usual list of obsessions. This most definitely applies to marketing, where I found many artificial intelligence-based solutions once I began looking for them. After accumulating a list, I’ve decided it’s time to pull together an overview of the topic.

My thesis was that AI-based systems already exist for most tasks that marketers perform, but are not yet connected into a single robo-marketer that (or is it who?) could do the job from start to finish. To test this, I listed the tasks that go into building a marketing program and matched these against my list of AI-based products.

Quite to my surprise, the machines haven’t risen so far after all. Of the three broad tasks I defined – planning, content creation, and execution – only content creation is served by what I consider to be strong AI* solutions. Some AI options are available for execution, but most are conventional predictive modeling products that I don’t count as strong AI because they still require humans to deploy their results. Marketing planning, which includes the all-important task of campaign design, is almost wholly untouched by AI.

Let’s look at each task in turn.

Planning

The lack of AI-driven planning solutions is especially surprising since so many planning tasks lend themselves to an AI approach. These tasks include market analysis (identifying potential buyers for a product, defining the needs and interests of potential buyers, identifying competitors, calculating potential market size and adoption rates), selecting marketing strategy, and selecting tactics (which can be defined as campaigns, experiences, or – if you’re cool enough – stages in the customer journey).

It seems well within the capabilities of current technology to find people who indicate a specific need, based on their Web searches or social comments, and then to understand who those people are in terms of demographics, behaviors, and other attributes. But the closest I could find were a couple of products that build profiles of groups the marketer defines in advance, such as brandAnalyzer by brandAnalyzer from Global Science Research and Empirical Insights from SG360. The one system that does look like strong AI is Bottlenose.  It performs the relatively common function of identifying trends but uses enough advanced technology, including natural language processing, topic discovery, and sentiment analysis, to impress me.

Similarly, competitor analysis should be well within the capabilities of companies like Radius, Everstring and Growth Intelligence, which already ingest the contents of corporate Web sites to understand each company’s business. But those vendors focus on finding prospects for B2B sellers. If any of them offers a competitor identification service, I’m not aware of it.

Business and marketing strategies are often compared to chess, a game that AI systems can famously play better than humans. I think the analogy is sound: like chess, business and marketing strategy involves a relatively limited number of moves with easily predicted short term consequences and large databases of past competitions which computers can study to predict longer term results. Strategic planning can also be supported by a rich treasury of simulation and optimization techniques that are very familiar to AI developers. Yet the closest I can find to strategic planning is optimization of media plans by media mix model vendors like MMA , Analytic Partners , Nielsen  and IRI.  But that is so tactical I classify it as part of execution. Otherwise, I haven’t seen anyone use AI to recommend a marketing strategy.

Nor has anyone really promised to design marketing campaigns using an AI system. This is another area that seems a natural application: computers can certainly use past results to predict the short- and long-term results of individual messages and, with a bit less certainty, of streams of messages. But while there are plenty of systems that predict the “next best message” or recommend what content the user is most likely to select, no one seems to have taken the obvious next step of using AI to design the best message sequence and refine it over time through automated testing. The closest I’ve seen are Amplero and Insightpool.  But they both start with individual data, so I'll discuss them in the section on execution.

Content Creation

Ironically, the most common reaction when I bring up machine-based marketing seems to be “well, they'll never write copy”. In fact, writing is one task where machines have already demonstrated huge success. General purpose writing programs including Wordsmith from Automated Insights , Quill from Narrative Insights and Arria NLG already write over one billion newspaper articles and reports each year, specializing in data-rich topics like sports and financial reports. Wordsmith also writes other sorts of reports, including summaries of marketing campaign performance.

Still closer to home for marketers, InboundWriter and Acrolinx score marketing content for effectiveness before it is released.

Most impressive of all from an AI perspective, Persado and Captora actually create content on their own. Persado does this by selecting and then testing content derived from a huge database of marketing language, classified by emotional, descriptive, and formatting categories. It works across email, landing pages, text messages, social posts, Facebook ads, app notifications, and other media. Captora automatically analyzes the topics and performance of content from the client and its competitors, finds opportunities for new campaigns, and creates appropriate landing pages to attract search traffic.

I consider Persado and Captora to be true AI-based marketing because they can actually replace work done by humans. But, as both vendors would probably rush to point out, what they really do is expand the volume of work that gets done, enabling marketers to execute hundreds of campaigns with the same human effort as it previously took to do dozens. So they are less about reducing the number of marketers than expanding marketer productivity.

Execution

While planning and content are arguably the most strategically important tasks that marketers do, there’s no question that they spend most of their time on execution. This is especially true since my definition of execution includes measurement and optimization because all three are so closely intertwined.

I broadly divide execution into audience development, message selection, and attribution. Audiences include paid media (purchased ads and lists), earned media (public relations and social influencers), and owned media (company Web sites and email lists, in-store promotion, call centers, etc.). Message selection includes content recommendations and personalization. Attribution includes everything that measures the results of marketing efforts – although, in practice, much message selection also relies on simple attribution to improve selection results.

Each execution category is served by advanced technology. Paid audiences are built with predictive modeling for list selection, programmatic media buys for advertising, and automated content analysis to understand intent. Earned audiences are built through influencer identification and predictions of who will cover which stories. Owned audiences are refined through more predictive models and behavior analysis. Message selection also relies on advanced analytics to recommend the right content for each individual and to find the best-performing messages for groups. Likewise, attribution systems apply sophisticated methods to isolate the incremental impact of individual marketing actions on long-term results. This long-term perspective is what distinguishes attribution from the measurement built into message selection systems, which instead chase immediate results such as email click-through or Web page conversion.

These technologies are certainly impressive, but few of them actually remove marketers from the process – which you’ll recall is my definition of marketing AI.  Predictive model scores, for example, are usually plugged into marketer-created rules that decide who receives which treatments. Even the recommendation engines rarely do more than predict which messages an individual is most likely to select.  Human-built rules still determine which messages are available and when messages will be presented.

There’s a lot of gray in this picture. Model scores and recommendation engines often replace complex segmentation rules even though some other rules remain. They may not replace marketers altogether but they do enable marketers to run larger numbers of more refined programs.  And they're a supporting technology for true AI marketing systems even if they are not AI themselves.

On the other hand, I think programmatic media buying does rise to the level of true AI. Again, the critical distinction is whether they replace human marketers – and I think that media buyers are pretty much not needed to execute programmatic programs. Obviously a human still has to set up the programs and provide the creative, but the programmatic systems then make complex judgements on their own.  Are these “judgements” significantly more advanced than the “judgements” that go into a lead scoring predictive model or personalized content recommendation? I’m not really sure. Maybe I’m misled by the fact that “media buyer” is an established profession while “lead scorer” or “content personalizer” are not job titles that people had before computers were available.

There are a few execution products that approach the border of true AI and may actually cross it. These include:

- Amplero, a newly released system that uses tree analysis to find very small market segments and then identifies the best content for each segment. What separates it from other personalization tools is that it optimizes against long-term value, such as revenue over the 14 days following each message, and its decisions take into account messages previously presented. I’d definitely consider Amplero true AI if it could plan sequences of messages, which would be pretty much the same as building multi-step campaigns. The system doesn’t do this yet but the vendor tells me they’re working on it.

- Insightpool, which identifies social media influencers, predicts how likely they are to take a user-specified action, and then recommends multi-step campaigns to encourage that result. Influencer identification and activity prediction are impressive but not unique; what makes me classify Insightpool as AI is its ability to select campaigns. This is something that you’d ordinarily expect a human marketer to do, even if the marketer was working with lists that the other functions had prepared.

- OneSpot converts existing pieces of content into multiple ad formats to permit reuse, classifies them (automatically, so near as I can tell) by purpose, and then delivers them to precisely targeted or retargeted individuals through programmatic ad exchanges in the optimal sequence to meet long-term goals. The reformatting, classification, and sequencing all strike me as things that humans would otherwise do manually, and of course I’ve already argued that programmatic media buying itself qualifies as marketing AI.

Final Thoughts

I know this post is too long to be effective but I wanted all to get this information down in one place. Artificial intelligence is an important topic in our general society and seems to attracting increased attention, even though Google Trends suggests otherwise. Marketers in particular are thinking about it as they adjust to rapidly changing technologies that increasingly rely on predictive analytics and other automation for effective management.

Given the hype that accompanies pretty much every new technical development, it’s helpful to see that AI-based marketing isn’t as far along as one might expect. But don't take that as a reason to relax: while it’s not time to panic, it’s definitely time to prepare. AI marketing systems already present some significant opportunities and their scope can only grow – perhaps exponentially as key techniques become more widely distributed. Now is the time to start building a realistic understanding of how these systems work, what they can and can’t do, and how they’ll fit into your future.


_________________________________________________________________________
* “strong AI” is used by AI experts to mean systems match or exceed human intelligence. I’m using it in that roughly sense, although with the more specific meaning of “systems that perform tasks that otherwise require human marketers”.

Everstring Offers Fast, Flexible, Account-based Predictive Models for B2B Sales and Marketing

Remember how much simpler life was back in 2010? Among our quaint notions, we thought that B2B companies couldn’t build predictive models because they didn’t have enough data about their customers and prospects. The Internet has changed that, providing oceans of relevant detail from company Web sites, social media, job boards, and other sources. Today, at least a dozen vendors are offering predictive models for B2B lead scoring, sales intelligence, and customer success management.

Many of the original scoring vendors specialized in a single application.  But today, most are broadening their products to serve multiple purposes. This lets the vendor charge more to each client and blocks out potential competitors. Marketers also benefit since they only have to buy, learn, and integrate a single system.

Everstring is relative newcomer to the B2B predictive modeling arena, founded in 2012 but only seriously entering the market after it received $12 million in Series A funding last August. The late start has let it adopt a broader scope from the beginning, offering both lead scoring and new prospect identification. The company plans to extend its offerings later this year to include real-time treatment recommendations.

But what really sets Everstring apart are two other factors: it works at the account rather than individual level, and it builds models really, really fast – as in, six minutes for a new model once data connections are in place. The two factors are related: Everstring can work quickly because it only imports a client’s account list and sales activities, saving complicated data mapping and analysis, and because it has preclassified its master database of six million U.S. businesses into clusters based on similarities in products, technologies used, hiring patterns, news events, social data, and other factors. This means that building a new model only requires using activity history to identify the client’s responsive accounts and finding which segments have the highest concentrations of those accounts.

That’s pretty light work compared with loading individual level data and identifying which attributes are most predictive for each client’s business. Matching against six million companies rather than 100 million individuals speeds things up too. The approach also lets clients score anonymous leads if IP address or similar information can identify their company. Models for different products can be based built by selecting only accounts that purchased that product.

Once a model is built, Everstring can score any new leads by just by identifying the segment their company belongs to and applying that segment’s score. Lists of new prospects require simply taking names from the highest-scoring segments.

Sounds pretty simple, eh? That’s because I’ve over-simplified. The data gathering and actual math are actually quite complicated.  Beyond that, Everstring does more than provide segment scores, which  measure the fit between a new account and the client’s previously responsive prospects. Specifically, it also measures purchase intent by based on more than 1 billion clicks per day on third party Web sites and emails. And it measures engagement by analyzing visitor behaviors on the client’s own Web site, gathered through a tracking pixel, plus other data imported from marketing automation. The combination of fit, intent, and engagement will guide the real-time treatment recommendations and can support additional scoring applications.  Fit scores alone are much more limited..

So, how do you deploy all this? Everstring has standard integrations with Salesforce.com, Marketo and Oracle Eloqua, which can send data for the initial model building and score new accounts as they are added to those systems. A real-time API can integrate with other CRM and marketing automation systems.

Pricing for Everstring is based on the types of models and volume. Lead scoring usually runs from $60,000 to $100,000 per year. New prospect names is additional. Pricing for real-time message selection isn’t yet set. The system currently has about 25 clients, nearly all added since last August.