Friday, February 24, 2017

Why MarTech Fails: A Data-Driven Answer

Do you suffer from Martech Fatigue Syndrome (MFS)?  Symptoms include obsessive concern with the number of martech vendors, anxiety at the prospect of evaluating new systems, and fear of missing out on an important new capability.  Severe cases have reported hallucinations of vendor logos covering vast surfaces and nightmares of being buried under a collapsed martech stack.  MFS is rarely life-threatening but can disrupt the quality of your every day marketing.  If you or someone you know shows symptoms of MFS, please call Scott Brinker immediately.

So far as I know, Martech Fatigue Syndrome is not yet a real thing.  But I've definitely sensed a certain weariness in recent discussions of marketing technology.  The initial excitement about new opportunities has become exhaustion as marketers realize they need to keep making investments even though they're not using their existing systems to the fullest.  See, for example, this Kitewheel study, which found 72% of agencies use less than 40% of their tools every week.

So what’s the problem? Have marketers simply purchased the wrong technologies – after all, they’re new at the system buying business and martech is filled with bright and shiny distractions. Or are they buying the right systems only to find that other roadblocks get in the way of success?

Many people have asked similar questions. So many, in fact, that I've found a half-dozen surveys in the past two months touched on the topic.  You can see the questions and their answers at the bottom of this post.

But each survey asks different questions and gets slightly different answers.  To look for over-all patterns, I’ve combined the answers on the following table, putting similar items on the same row and keeping related ones nearby.  I’ve grouped the answers into general topic areas: organization, management support, marketing strategy, data management, delivery systems, and external factors. High-rated issues within each survey are shaded orange and low-rated issues are shaded green.  In other words, orange cells are the biggest problems, green cells are the smallest, and white cells are in between.

The reason for all that careful arrangement is to see any clusters in the answers. Sure enough, some do emerge: the biggest problems are concentrated in organizational issues (lots of orange).  The one exception that people think their own skills are perfectly adequate. Of course.

The management support area is mostly neutral except for a slash of orange for Return on Investment.  That make sense: measuring ROI is always a challenge for marketers.  To be clear, the answers are referring to the ROI of marketing programs in general, not martech investments in particular.  If anything, the surprise is that related items like management support and budget are less of a roadblock.

Marketing strategy isn’t a major problem in most cases, with just one survey as an exception. As with skills, this basically means that marketers are confident they know how to do their jobs.

The next two items, data management and delivery systems, are where technology comes in. There’s more green here than orange, confirming our hunch that access to adequate technology isn’t marketers’ main problem.

The final group, external factors, is no problem at all.

As a final bit of analysis, I've normalized all the answers to create a combined ranking for each category, splitting out ROI and internal skills since they are so different from everything else.  Apologies in advance to any real statistician who is horrified at the procedural flaws in this approach.   The rankings do seem to come out about right, and putting it all into one graph meets the goldfish attention span test.

Bottom line: measuring ROI and organizational roadblocks are the biggest reasons marketers fall to get value from technology. Finding good technology and knowing what to do with it are less of a concern.  These answers aren't unexpected but now you have a data-driven answer next time anybody asks.


Survey Details:

Retail Touchpoints for Magnetic
Barriers to cross-channel experience:
not enough data for full profile
internal organizational silos
don’t know what types of messages resonate best
struggle to get right message to the right person
delivery systems are not integrated
struggle to integrate first and third party data
don’t know how to use our data to create a better experience

Winterberry Group for IAB Data Center of Excellence
Obstacles to value from data-driven marketing:

year ago

difficulty proving ROI
lack of internal experience
insufficient technology
siloed organization
inadequate first party data
tests have failed
lack of leadership support
competitive pressures
lack of guidance from agency/service partners
inadequate third party data
little demand from customers

Forbes Insights for Aprimo 
Agile marketing challenges
proving ROI
employees not empowered
can’t connect agile marketing to business outcomes
hierarchical organization
lack of management vision
lack right technology
lack IT talent
lack marketing talent
difficulty choosing right third party

Kantar Vermeer for American Marketing Association 
Not confident the organization’s marketing team

year ago

has right operating model (people/structure/processes/tools)
understands ROI of efforts
has clear strategy
is investing in right customers
is doing the right things
has clear brand position
has right capabilities

Obstacles to 1-to1 personalization
organizational constraints block personal accountability
automating decisions at scale
assemble real time customer view with full context
understand buyer behavior in context
creating compelling offers and content
integrating third party data
data quality
understanding who to personalize when in which channel
sustainable data architecture

Econsultancy for Adobe 
Most difficult customer experience components to master
journey design

Thursday, February 16, 2017

Zaius Offers Mid-Market Customer Data Platform Plus Analytics and Campaigns

It wasn’t until the end of a long demonstration that I finally understood what Zaius is. Which is pretty ironic, since they’re an almost perfect example of a Customer Data Platform – that is, a system that assembles customer data from multiple systems and makes it available for marketing and analytics. If anyone should recognize a CDP when they see one, it’s me. Come to think of it, if anyone is going to call something a CDP even when it isn't, that’s probably me, too.

So what fooled me about Zaius? It’s probably that most of their clients are mid-sized ecommerce companies, and the systems I’ve recently seen for ecommerce marketers have focused on personalized messaging and optimization. Zaius seemed to fall into those categories since much of our discussion focused on building marketing campaigns and doing attribution. I probably wasn’t helped by Zaius’ Web site, which calls it a “B2C CRM” and then lists single customer view, real-time marketing automation, and cross-channel attribution as its main features.  Single customer view is clearly CDP territory, but the marketing automation and attribution are not. In fact, CRM and marketing automation are feeder systems to CDPs, so you could argue it’s logically impossible for the same system to be both.

None of which really matters, I guess.  Let’s forget about labels and look at what Zaius does.

Turns out, the primary thing that Zaius does is to build that unified customer database. It has connectors to gather data from Shopify and Magento ecommerce systems; Salesforce ExactTarget, Oracle Responsys, IBM Silverpop, MailChimp, and SendGrid email services;  and the Segment, Tealium and Google tag managers.*  More prebuilt connectors are on the way. In the meantime, Zaius can capture data from Web sites through Javascript tags, from mobile apps through a System Development Kit, and from pretty much anything through APIs and batch uploads. The system loads data into a structured schema, which must be updated to accommodate new fields or objects.   Non-technical users can add custom fields on their own, but Zaius staff must add a new object. The system will reject records that have unexpected or invalid data and notify users of the problem. Zaius doesn’t automatically apply address standardization or other data transformations, although the vendor can create custom adapters to do some of that.

Once data is loaded, Zaius does deterministic identity resolution, which means it will chain together data using any identifier known to be associated with an individual. (For example, if a phone number and email address have been associated with the same person, any new record with either that phone number or email address will be linked to that person). It builds profiles of anonymous identifiers, such as cookies, and will link them to known individuals if they are later associated with a personal identifier. The system will merge identities if it discovers a connection, but it doesn’t do probabilistic matching across devices, fuzzy matching of similar postal addresses, or householding.

The data loading process also includes sessionization, which associates events that occurred around the same time. For example, multiple Web page views during a single visit would be a session. Zaius assigns events to sessions after they are linked to unified identities, so one session can include interactions across several channels. This might help users find customers who called on the phone after having trouble placing a Web order.

Zaius gives users tools to analyze the data it has captured, to create and export segments, and to run outbound marketing campaigns. Analytics include dashboards, attribution reports, and funnel analyses that track customers through a purchase process. Because Zaius is unifying data from multiple sources, its analyses can span events that happened in different systems. This means a funnel report could include an outbound email, a Web visit from a link in the email, and an ecommerce purchase during that visit. Neither the email or ecommerce system alone could track this entire path.

The system can report on customers at different stages in the life cycle, giving a useful overview of the user's business. It also lets users see tactical metrics, such as number of new customers acquired in the past month, and then drill down to see which campaigns produced those customers. Users who want to explore still further can look as deep as the specific events within an individual customer’s history. Security features can limit users to specified subsets of data, such as particular Web sites or product groups.

Segmentation in Zaius can draw on any data in the system. The system provides a form-based segment builder that can create complex expressions. These can be saved and used within other segment definitions. Users can export segments to other systems, including a two-way audience synchronization with Facebook and Google. In addition, a real-time API lets external systems query Zaius directly to find individual customer profiles. Segment exports and API access are what qualify Zaius as a CPD.

Segments can in turn be used in marketing campaigns.  These are built with templates that let users specify the channel (email, push, or SMS) and delivery type (once, recurring, continuous, or event triggered). Users can create email messages using a drag-and-drop interface that supports advanced personalization, such as selecting the top products in a customer’s most commonly purchased category. Personalization variables can be built with a scripting language or by inserting pre-built objects. Testing features let users define a test duration, evaluation criterion, and content versions. Users can set aside a portion of the audience to automatically receive the winning version when the test is complete. Zaius lacks more advanced optimization such as multi-variate tests that automatically create different combinations of features, finding segments within the audience that respond best to different versions, or predictive modeling.  Zaius sends email and SMS lists to external vendors for delivery. It uses Amazon SNS to send push messages. The vendor plans to add direct mail and browser push channels in the future.

Zaius was launched in 2014, with an original focus on providing a unified customer view and analytics. Its initial clients were large enterprises but most sales are now to mid-market firms with at least 100,000 contacts. Pricing is based on volume and starts at $1,000 per month.

*Segment and Tealium are CDPs themselves, but let’s not confuse things even more.

Friday, February 10, 2017

LeadGenius Adds a Dash of Artificial Intelligence to Account Based Marketing

You may have noticed that I’m writing a little less about artificial intelligence than I had been. It may be that Skynet has imprisoned the real David Raab to block him from issuing dire warnings about its imminent threat to humanity and replaced him with a less alarmist simulation. You can’t actually prove that isn’t happening. But the David Raab, or Raab-bot, writing this will tell you it’s because he’s concluded that AI is destined to become so pervasive that it doesn’t make sense to treat it as a distinct topic. It will simply be embedded in everything and so should be evaluated as part of whatever it belongs to.

LeadGenius is a good example. The company is in the business of assembling B2B marketing lists – an industry dating back centuries to city directories and beyond. But LeadGenius was founded in 2011 to commercialize university research into combining AI with human inputs. It has since expanded from list gathering to all stages in the Account Based Marketing process, sprinkling in dashes of artificial intelligence at every step along the way.

Let’s look at those stages, using the four step structure of the Raab Guide to ABM Vendors.

1. Identify target accounts. This includes assembling data on potential accounts and selecting the right targets. Like many data gatherers, LeadGenius uses a combination of Web and other sources to build company and contact lists. Nearly every vendor who does this applies some form of natural language processing to extract information from unstructured sources. LeadGenius does this too. But it goes further by using artificial intelligence to identify records with questionably accurate information. It then sends these to humans for direct verification by telephone. The company guarantees 99% accuracy in its data, which is significantly better than most competitors can offer. AI's contribution here is to let LeadGenius call only the companies that need human contact, reducing over-all effort substantially. 

To find the right targets, LeadGenius loads a client's current CRM lists. It analyzes these for accuracy and completeness, providing users with reports that highlight problem areas. There’s probably some AI at work in that analysis. LeadGenius then identifies major file segments within the customer base and finds similar companies in the broader universe, estimating potential buyers and revenue by segment. Somewhat surprisingly , LeadGenius doesn’t create predictive lead scores, having found its more useful to prioritize prospects based on company attributes like size and industry. LeadGenius does use artificial intelligence, or at least its country cousin “fuzzy logic”, to map business titles into buyer roles, taking into account how different terms are used at different size companies to describe the same role.

2. Plan interactions. LeadGenius has a basic email campaign capability, including segment definition, email templates with personalization variables, and email sequences. There don’t seem to be any particular AI features here, although we’ll see in a moment that email does play a key role in LeadGenius’ AI utilization.

3. Execute interactions. LeadGenius sends emails through corporate or individual salespeople’s email accounts. It captures replies and uses AI-based natural language processing to classify them, distinguishing answers that indicate interest from out-of-office messages and clear rejections. Hot leads are pushed back to salespeople’s inboxes.  All response classifications are added to the database where they can be used in future selections. So AI does indirectly drive interaction flows. Response data can also be posted to or Marketos, with additional integrations planned for the near future. Messages through other channels would have to be executed through marketing automation or CRM.

4. Analyze results. LeadGenius has the usual email and campaign reporting, enhanced with the AI-based response classifications.

As promised, you see a bit of AI magic at each of these process stages (assuming you count the AI-based email response as enabling the interaction planning). Certainly there’s room for more features and more AI use. But it’s already enough to illustrate how AI will add power throughout the ABM cycle.

LeadGenius is used primarily by large enterprises selling to small businesses. Those are the firms that can most benefit from its comprehensive data, market analyses, and prospect lists. Pricing is tailored to each client. The company has more than 150 B2B clients.

Thursday, February 02, 2017

Quaero AdVantage CDP Bridges Identified and Anonymous Data

It’s a common pattern: several vendors proudly roll out new products they developed in secret, only to find they’re all very similar. The amazing coincidence isn’t really so amazing: everyone sees the same problems and has the same technologies available to solve them. So they come up with similar solutions.

Simultaneous rollout.  I've had this picture in my head for years.  Apologies to Dr. Seuss.

We’ve seen some of that in the Customer Data Platform industry, but there’s a twist. Many CDPs evolved from older systems and inherited some of their ancestors’ characteristics. One of those lineages goes back to marketing databases from simpler days, when postal mail and email were the main channels. The big challenges for those systems were loading complex data structures (addresses, transactions, message history, etc.), cleaning that data, and identifying records that belonged to the same individual. In that world, there was no such thing as an anonymous customer and most data was neatly structured. As I say, a simpler time.

Quaero’s AdVantage is a good example of a system with deep roots in the old methods – but updated to handle modern challenges. Quaero itself was founded back in 1999 as a marketing services provider (meaning they built custom marketing databases and attached tools like the Unica campaign manager). It was purchased in 2008 by CSG International, a telecom customer communications specialist, and repurchased by the original management in 2014. By then, the managers had already started work on a next-generation platform designed to handle both traditional and online data, using relational databases for one and a NoSQL system (in this case, Hadoop) for the other. The company has recently introduced this to the market as AdVantage.

The split architecture of AdVantage is actually pretty common among CDPs, since anonymous and identified customer data are often kept separate for privacy reasons. It’s also common to hold all the raw data in a NoSQL data lake and extract it to a relational database where it's refined and restructured for analysis. AdVantage does that too. It’s a bit less common for vendors to be so open about these details; Quaero management's transparency is probably another result of their maturity.

What’s truly unusual is the sophistication of AdVantage’s data processing itself. After nearly two decades of wrestling with customer identities, Quaero has mastered tricks that many newer vendors have yet to see.* More concretely, the system provides over 1,000 prebuilt “workflows” that perform tasks within data staging, loading, cleaning, transformation, aggregation, scoring, and measurement. These can be configured to specific situations, giving users a great deal of power without writing actual queries or scripts. Workflows can also be strung together to create larger flows, which AdVantage visualizes nicely.  This lets users trace exactly how the system got to its results. Configuring the workflows is still definitely technical work, which is either done by IT staff or the Quaero services team. But AdVantage makes it more efficient than hand coding and vastly more accessible to anyone other than the original coder.

Another important feature is that AdVantage flows work with metadata, meaning they are not mapped directly to the underlying data stores. This means an implementation can move to different platforms without losing most of the work. That makes it easier to adopt new technologies and to convert to more powerful platforms if a system outgrows its original installation.

AdVantage’s features for working with identified customers are especially mature, handling different kinds of “fuzzy” name and address matching as well as creating a “golden record” of best values from all sources. It also has strong features for unifying anonymous inputs, which it supplements with device matching services such as Tapad and Oracle Crosswise. AdVantage creates separate customer IDs for identified and anonymous profiles, and then, when possible, links them with a master ID

Once the data is assembled, AdVantage makes it available to marketing users and applications such as business intelligence tools and campaign managers. AdVantage provides its own interactive reports and segmentation interface. But most users will attach their own tools such as Tableau or Looker. AdVantage also connects to execution tools such as email engines. These tools can directly access both the relational and NoSQL data stores. Quaero has built standard connectors to common products, both to load data and access it. It builds new connectors as clients need them.

Existing AdVantage installations are hosted on Amazon Web Services, although a client could also run it on-premise if desired. Pricing is based on factors including number of sources, data volume, users, and applications. An average installation runs $15,000 to $25,000 per month although some are lower and higher. Quaero provides services with the product to help clients get set up properly and make changes over time.

*Some others have, especially those with a similar background.

Monday, January 23, 2017

#FlipMyFunnel Launches Account Based Marketing University

Not the ABMU mascot
It sometimes seems that Account Based Marketing is really Marketing Automation 2.0, in that many people leading the charge to ABM were also involved in launching B2B marketing automation ten years or so ago. How none of us has aged a day is a mystery we shan't discuss.

One of the lessons learned during the growth of marketing automation was how important it is to get marketers trained in new techniques. The ABM community addressed this early with thought leadership from the ABM Consortium and now more extensively with the ABM University, a project launched earlier this month by the high-energy folks at #FlipMyFunnel.

ABMU offers 250 online lessons taught by more than 40 thought leaders, including Yours Truly (although in fact I haven’t done anything yet). There will be tests and a certificate of completion. Introductory price for the course is $500, planned to go up to $1,000. (Don’t be confused by the Sign Up for Free button on the Web page. They’re trying to get rid of it. !#$@#$ martech.)

Professors of #ABMU include:
Craig Rosenberg, Co-founder and Chief Analyst at Topo, Inc.
Christopher Long, Director of Marketing Operations at WP Engine
Maria Pergolino, SVP of Marketing, Global Marketing at Apttus
Matt Senatore, Service Director, Account-Based Marketing at SiriusDecisions
Koka Sexton, Founder at Social Selling Labs
Julia Stead, Director of Demand Generation at Invoca
Justin Gray, CEO at LeadMD, Inc.
Matt Heinz, President at Heinz Marketing
David Raab, Owner at Raab Associates
Tyler Lessard, CMO at Vidyard
Jill Rowley, Queen of #SocialSelling
Lincoln Murphy, Growth Architect at Sixteen Ventures

So far there is no ABMU fight song or mascot, although I have pointed out to them that Aardvark costumes are widely available at reasonable prices.

Wednesday, January 18, 2017

Customer Data Platform Industry Profile: A Look Inside the Numbers

My snarky twin at the Customer Data Platform Institute just published a new report on the CDP industry. Since few industry vendors release financial or business details, the report relies on public sources including Owler for revenue estimates, Crunchbase for funding history, and LinkedIn for employee counts. Most vendors did provide client counts, and several privately shared other information where the public data was clearly wrong. You can download the report here. I'll wait while you do that.  (Sound of fingers tapping.)

Okay, you've downloaded it, right?  Good.

As you see, the report only presents figures for the industry as a whole. We feel those are reasonably accurate but that data for individual vendors are too unreliable to show separately. That may sound illogical but bear in mind that figures for the larger vendors are more reliable, so many errors that are significant for individual small vendors don’t materially change the total. Also remember that some vendors provided information in confidence and we made estimates of our own for some others.

I do feel I can safely publish statistics for three groups within the industry.  This gives some additional insight without exposing any proprietary or misleading vendor data.  The groups are based on each vendor's original business.  They are:

  • Tag managers. This may seem an unlikely starting point, but it actually makes sense.  Tag management was originally about collecting data once (when a Web page loaded) and then sharing it with other systems that would otherwise have their own tags. This gave the Web site owner more control over what went where and reduced the number of tags on each page.  The data sharing was similar to what happens in integration platforms/data hubs  like Jitterbit and Zapier. So tag managers were always about data distribution.  To become true CDPs, the tag vendors had to ingest data from additional sources and send the data to a persistent database. Ingesting new sources can be challenging but vendors could grow incrementally by choosing which sources to accept.  Feeding a persistent data is basically just adding a new destination for data sharing.  So the transition to CDP offered a reasonable path to escape being a commodity tag manager.

  • Campaign managers. I’m using this term loosely to include companies that offered any sort of marketing message selection. It includes systems that do email, Web site messages, mobile app messages, and omnichannel campaigns. These vendors all started out as CDPs in the sense that they always built unified customer databases. Among other things, this meant that most included reasonably robust cross-channel identity resolution. These vendors didn’t necessarily start by sharing their database with other systems.  But they do it now or I wouldn't consider them a CDP.

  • Data assembly systems. This is a bit of a catch-all category but almost every system in this group was designed primarily to create a customer database that would be accessible to other systems.  Intended uses included analytics, marketing execution, or both. (I say "almost" because the group includes two systems that built databases primarily to support their own attribution services.) There’s more variety within this group than the other two.  But many vendors provide advanced identity resolution and all are strong at providing external access.
Here are key statistics for each group.

original purpose:



2016 revenue

customer count

revenue / customer

employee count

revenue / employee

Tag Management


$356 million

$118  million





Campaign Management


$106 million

$108 million





Data Assembly
$246 million
$100 million

Here are some observations:

  • Revenue is split about evenly among the three groups. That’s a bit surprising because tag management is an older and more established category than the others, so you might have expected it to have more revenue.  Vendors in the other categories do tend to be newer, smaller, and growing more quickly.

  • Tag management vendors have many more customers and earn less revenue per customer. This largely reflects the original tag management products, which are sold to many non-enterprise customers. But the tag management vendors also have hundreds of enterprise clients.  Many of those clients are building the large-scale customer databases we expect to call a CDP.  The tag management group also includes a couple of vendors who specialize in building CDPs for smaller companies. These are not as expensive as the enterprise installations.  The campaign management vendors average about $100,000 per customer, which is what you'd expect for an enterprise CDP.  Revenue per customer is just $34,000 for the data assembly vendors, but that's largely due to one vendor with 2,000 clients.  Without that vendor, revenue per customer figures for the data assembly group would be $88,000. Backing out non-CDP clients is why the report puts the number of CDP customers for the entire industry at 2,500.
  • Revenue per employee is generally in line with what we expect to see at growing Software as a Service companies. The standout here is the campaign management group, which has an impressively high figure of $207,000 per employee.  This suggests the campaign managers have a high value-added business.  The relatively low amount of external funding is more evidence that campaign managers throw off considerable cash from their own operations. The much lower revenue per employee for the data assembly companies, $108,000, is more typical of new SaaS ventures.  Indeed, several of these are just starting to earn revenue from their first clients.  These data assembly companies have attracted considerably more funding than the campaign managers, giving them a cushion to invest in growth. (If you’re wondering about that company with 2,000 clients, its revenue per employee is similar to others in its group.)

There are other nuances to consider in assessing these figures. For example, several vendors do business through agencies, which makes it harder to count clients and to compare revenue per client.  But the over-all picture that emerges is a healthy industry that is already attracting substantial revenue and funding.

The report projects a 50% annual growth rate, which yields an estimated $1 billion revenue for 2019.  The projection is based on public and private reported growth rates, which actually averaged much higher than 50% on a revenue-weighted basis.  The report used 50% to be conservative.  While past performance doesn't guarantee future growth, I think CDP revenues will if anything accelerate because most marketers still don't realize what a CDP can do for them.  As more of them get the message, CDP adoption should skyrocket. So the future is bright indeed.

Monday, January 09, 2017

Artificial Intelligence, Virtual Reality, and Government Control: Perfect World or Perfect Storm?

If it weren’t the print edition, I would have sworn today’s New York Times business section had been personalized for me: there were articles on self-driving cars, virtual reality, and how “Data Could Be the Next Tech Hot Button”. That precisely matches my current set of obsessions. It’s especially apt because the article on data makes a point that’s been much on my mind: government regulation may be the only factor that prevents AI-powered virtual reality from taking over the world, and governments may feel impelled to create such regulation in self-defense of their authority. The Times didn’t make that connection among its three articles.  But the fact that all three were top of mind for its editors and, presumably, readers was enough to illustrate their importance.

I’m doubly glad that these articles appeared together because they reinforced my intent to revisit these issues in a more concise fashion than my rambling post on RoseColoredGlasses.Me. I suspect thread of that post got lost in self-indulgent exposition. Succinctly, the key points were:

- Virtual reality and augmented reality will increasing create divergent “personal realities” that distance people from each other and the real world.

- The artificial intelligence needed to manage personal reality be beyond human control.

- Governments may recognize the dangers and step in to prevent them. 

Maybe these points sound simplistic when stated so plainly. I’m taking that risk because I want to be clear.  But depth may add credibility. So let me expand on each point just a bit.

- Personal reality. I covered this pretty well in the original post and current concerns about “fake news” and “fact bubbles” make it pretty familiar anyway.  One point that I think does need more discussion is how companies like Facebook, Google, Apple, and Amazon have a natural tendency to take over more and more of each consumer’s experience.  It's a sort of “individual network effect” where the more data one entity has about an individual, the better job they can do giving that person the consistent experience they want.  This in turn makes it easier to convince individuals to give those companies control over still more experiences and data. I’ll stress again that no coercion is involved; the companies will just be giving people what they want. It’s pitifully easy to imagine a world where people live Apple or Facebook branded lives that are totally controlled by those organizations. The cheesy science fictions stories pretty much write themselves (or the computers can write them for us).  Unrelated observation: it's weird the discussions which Descartes and others had about the nature of reality – which sound so silly to modern ears – are suddenly very practical concerns.

- Artificial intelligence. Many people are skeptical that AI can really take control of our lives. For example, they’ll argue that machines will always need people to design, build, and repair them. But self-programming computers are here or very close (it depends on definitions), and essential machines will be designed to be self-repairing and self-improving.  Note that machines taking control doesn't require malevolent artificial intelligence, or artificial consciousness of any sort. Machines will take control simply because people let them make choices they can’t predict or understand. The problem is that unintended consequences are inevitable and for the first – and quite possibly the last – time in history, there will be no natural constraints to limit the impact of those consequences. Random example: maybe the machines will gently deter humans from breeding, something that could maximize the happiness of everyone alive while still eliminating the human race. Oops. 

- Government intervention. Will governments decide that some shared reality is needed for their countries to function properly?  How closely will they require personal reality to match actual reality (if they even admit such a thing exists)?  Will they allow private business to manage the personal reality of their citizens? Will they limit how much personal reality can be delivered by artificial intelligence? These issues all relate to questions of control. Although there’s an interesting theory* that the Internet has made it impossible for any authority to maintain itself, I think that governments will ultimately impose whatever constraints they need to survive on individuals, companies, and the Internet. This probably means governments will enforce some shared reality, although it surely won't match actual facts in every detail.  It’s less certain that  governments will control artificial intelligence, simply because the benefits of letting AI run things are probably irresistible despite the known dangers.

So, is the choice between having your reality managed by an authoritarian government or by an AI? Let's hope not.  I prefer a world where people control their own lives and base them on actual reality.  That’s still possible but it will take coordinated hard work to make it happen.

*For example, Martin Gurri’s The Revolt of the Public