Glimpsing IBM Watson's High Tech Analytics In Silicon Valley

Silicon Valley types want me hanging out at their business events. One such event last week brought me down to one of the Valley's private venues for an IBM Watson presentation. I'm not the target clientele for this Big Data analytics solution but I had to check things out. There was no suitable on-location backdrop for my badge selfie, so I had to take the photo below at an undisclosed location.

I signed up to hear their two tracks on procurement intelligence and trade-off analytics after the main pitch. IBM people get the API economy. I heard them pitch their API developer ecosystem at Oracle OpenWorld 2015, and now it's good to see the Watson engine in action. The Alchemy Language API looks like an incredible business intelligence (BI) tool. The "news explorer" live link diagram showing connected news stories would be excellent for PR or marketing people, or for open-source intelligence (OSINT) practitioners.

The main pitch dude's recommended reading list included a book on machine learning, but I couldn't write down the author's name from where I sat. Amazon lists plenty of machine learning best-sellers, so my local library must have one. I did capture Pedro Domingos' The Master Algorithm and Provost/Fawcett's Data Science for Business from his list, unless I copied the titles incorrectly. I have so many books to read already that adding these will push the completion of my business reading list well into 2016. That's what it takes to demonstrate thought leadership, and that's why I get invited to these events.

One IBM guy introduced his "Cognitive Computing Index" describing multiple ways for human operators to educate maturing AI systems. IBM suggests Watson's clients iterate revisions every 90 days for whatever they have the system compute. Iterative approaches to refining BI output are supposed to maximize the BI's monetary value, and seat count users should see this value in their commission revenue.

The trade-off analytics session demonstrated Watson's Pareto optimization, graphical outputs, and social media stream matching. The recommended pathway records are a useful audit trail for some data miner to explore. I bet that data mining the faulty pathways will reveal how the top 20% of data scientists in an enterprise are making 80% of the correct decisions. That would be some useful Pareto optimization when performance bonus allocation time comes around.

The procurement intelligence session was all about making purchasing people into knowledge workers. I remember how I did purchasing as a junior supply officer in the US Army back in the late 1990s. I searched the Web for three different vendors and picked the one with the lowest price. It was too easy and probably sub-optimal. The difference today is that Watson is supposed to make research on prices, vendor choices, and spending history a Big Data effort. If AI truly integrates internal and external data feeds as advertised, then it's a bona fide ERP revolution. If users comprehend Watson's word clouds, heat maps, and visualizations, then it's also a knowledge management (KM) solution.

I keep hearing Silicon Valley people talk about how they increasingly prefer workflow ERP solutions over managing legacy files. I told several IBM reps at this event that they will have to integrate workflow data signatures into the internal feeds Watson ingests if they want to stay relevant. It will still be a challenge for developers to build APIs that handle unstructured data, especially if the enterprise has no data warehouse or data lake aggregating external data feeds. The best developers will figure it out. I would figure it out but I'd rather fiddle with financial applications. Watson and other AIs are supposed to be the "easy button" for data transformation once operators are comfortable educating the systems. The AI revolution means everyone becomes an amateur data scientist.

S&P Ratings Business Under Fire From Australia

S&P just can't catch a break.  They came under the US Department of Justice's scrutiny while competitor Moody's escaped attention.  Now an Australian judge has found S&P liable for the ratings it gave to poor investments.  More rulings like this in the developed economies will make rating complex securities almost impossible due to liability exposure.  I say "almost" because a ratings agency will have to place such severe limits on its assessments as to make a firm opinion meaningless.

The old principle of caveat emptor is slowly succumbing to a culture of settling scores.  Investments are risky and more complicated investments carry more risk.  Whatever hides inside all of the moving parts of derivatives can blow up the whole instrument.  Sophisticated investors should know this but they feign ignorance when they think litigation can compensate them for bad judgment.  The courts should be a remedy for fraud, not stupidity.

I have no sympathy for investment banks who knowingly package garbage into an security and misrepresent it as a good deal.  Those people are liars and phonies.  The prevalence of such behavior on Wall Street's sell side should be sufficient warning to institutional investors that complex derivatives are at best unnecessary and at worst a disaster waiting to happen.  Ratings have always been mere icing on the cake.  The cake itself has always been of questionable nutritional value.

McGraw Hill Financial (MHFI) doesn't have to throw away S&P just yet.  The unit's index services are a very important brand in the financial sector.  Capital IQ is indispensable to countless traders and analysts, until of course something with deeper Big Data analytics comes along.  Potentially mortal wounds to credit rating services don't have to destroy an entire enterprise.

Nota bene:  Alfidi Capital does not rate derivatives.  If the Alfidi Capital Blog or research reports describe a stock, bond, or other security, such a description is always in the context of what I do with my own money.  In other words, my opinions are only useful for my own decisions and not for anyone else's situation.  

Mobile App Marketing at APPNATION V 2013

I've got app fever after scoring a full conference pass to last week's APPNATION V Conference at Moscone West in San Francisco.  The rapid adoption of mobile devices is spurring a whole new subset of marketing campaign marketplaces, metrics, and agencies.  This ecosystem is going to completely replace traditional marketing in less than a decade.  I had to get my fill.

The opening keynote with Facebook introduced the build / discover / monetize paradigm for apps.  Getting 5% of app users to pay is considered the threshold for success but high customer acquisition costs defeat monetization.  There seem to be two broad categories for app adoption:  viral growth apps (with a large installed base monetized through ads) and utility apps built to be sold.  The Facebook rep pushed their Parse feature to build an app's back-end infrastructure and their Login feature for tracking and sharing data on users.  Okay, I think I get the difference; Parse is for hosting and Login is for app management.  Facebook's research shows that the typical app users has maybe 40 of the things and uses each between one and ten times.  Most users don't enable push notification but Facebook's autofill button must be enabled for apps to use its functions.  Hear that, developers.  Enable your one-click adoption functions early in the app's development cycle.

The state of the "app nation" according to Flurry is that 2.3T app events occur each month, and not just downloads.  Smartphones and tablets are really the first wearables because everyone keeps them on hand constantly.  I was astounded to see the figure that apps represent 87% of people's online time on their mobile devices and traditional web browsing is the other 13%.  Apps are thus the intermediary in the typical mobile user's online experience.  The browser wars are over, and the browsers lost.  I learned a new KPI:  monthly active users (MAU), which doesn't just pertain to games.  App marketers should know that what Nielsen is to broadcast media, comScore is to software and apps.  Some apps now have more subscribers than telecom carriers and are deploying virtual storefronts.  This is why Flurry and other players are counting on in-app messaging to be a dominant communication channel and promotional channel for goods.  They also think that the app communication sector will separate into distinct channels for messaging, visual media sharing, and news feeds.  I take that to mean that an all-in-one app just won't appeal to users.  I'd like to see how the mobile sector handles a mature market like South Korea now that it has reached saturation with devices.

I actually learned something from Bloomberg's talk with the dude from Hotel Tonight even though I don't follow the travel and leisure sector.  Holiday periods provide different use cases on shopping and the effectiveness of promotions.  I think Hotel Tonight's heavily manual method for vetting hotels is not as scalable as it could be if they would automate their process.  I think that's why they'll have a tough time competing against Priceline.  I'm getting the impression that how developers stack their ecosystem with incentives determines who participates.  Gamification incentives like star ratings and free channel exposure incentivize hotels to use this particular app.  I don't use Google Wallet or PayPal but their ability to store credit card data reduces friction in online transactions.  I'll bet hotels like that when they participate in an app marketplace.  Here's one more trick I'd like app developers to try.  Apps need geolocation tools to optimize omnichannel promotions.  Heat maps will reveal heavy users' geographic regions and Big Data can correlate usage instances with travel patterns to see if heavy users are frequent travelers.

LifeStreet Media took more than their allotted ten minutes to make their case for in-app advertising.  Freemium distribution plus in-app ads equals revenue for app developers who don't have paid subscription models.  Real-time bidding (RTB) ad servers allow advertisers to bid on individual impressions rather than a certain demographic's content adjacency.  Advertisers can combine RTB servers with mediators who optimize ad inventory to experiment with different campaigns.  A mediation layer monitors different ads' monetization.

I caught enough of Mojiva's monetization workshop to see how a second tier of ad networks can increase fill rates.  Do a Google search of "mobile ad network" to see these things proliferate.  I'd like to know how advertisers segment their ad inventory.  Is it by geography?  Or by some other user demographic?  I guess the seasonality use cases the Hotel Tonight guy mentioned are a factor.

The "Candace and Vijay Show" featured two mobile PR legends imparting wisdom on building a PR campaign.  They defied the conventional wisdom I've heard about journalists by saying app developers should appeal directly to reporters in their verticals for coverage.  Uh, ooookaaaay, but a lot of these developers in attendance are building apps just for the sake of building apps.  They'll be disappointed to find the "app beat" isn't covered by many journalists.  The PR gurus also recommended getting traction on app blogs first before going to mainstream media.  I had sticker shock when they threw down the cost of professional PR representation.  PR agencies run from $10-15K/month and limited-run consultants cost maybe $5K/month.  There's no way even VC-backed app startups should have to pay out that kind of dough!  The best PR is free.  The one best thing I got out of their well-attended show was that getting a journalist to agree to do a story involves iterative negotiations.  The process starts with an initial pitch of a PR statement in an email that's six sentences or less, with a time window for response setting a preemptive close.  Keep dripping those journalists for a feature, because the major app stores won't risk embarrassment by featuring an app that has zero traction from other media.

I didn't have time to watch the MEF trade association present their policy initiatives in a separate seminar.  Their white papers and market surveys look like good background research for app developers just beginning to craft their go-to-market strategies.  MEF's free App Privacy tool is a privacy policy generator that creates code for developers to embed in an app.  Filling out a short form has never been so easy.

Google Analytics' power session was a revelation.  Every app download can trigger tracking information thanks to conversion tags from Google Tag Manager.  See folks, that's why that embedded privacy policy from MEF's generator will be so important if you want to get full benefit from metrics with no legal problems.  Analytics and Tag Manager can track an app user across the entire Google experience to see how they found the app.  See folks, there's no privacy left for any of us anyway.  Going online means everyone can see everything you do.  I am very impressed that Tag Manager allows developers to store conversion-based rule sets and triage app bugs.  That's a cloud-enabled business rule engine in decision management.  Google Analytics Academy is a MOOC blessing for developers (and also us bloggers) who want max benefits from the platform.  I need to read up on the Google Analytics Blog to see the tips I've been missing.

I got one big thing from the "masters of monetization" panel.  If your app drives installation of other apps you can capture 80% of the ad dollars spent across those apps.  This is why WeChat and LINE are so effective.  They enable app ecosystem replication.  Their other points also made sense as effective tools.  App developers who can show use case data to advertisers will justify higher eCPM but I think they'll have to adapt the formats they use to present it if RTB is the marketplace norm.  App developers prove their worth with data on click-through rates, targeted user segments, and geographic data on adoption.  The ability to segment a user base is IMHO an underestimated advantage if business users don't like seeing ads in their apps.  Optimization eventually reaches a point of diminishing returns.  Advertisers use frequency caps to avoid oversaturating a target demographic.

A real live VC from Signia Venture Partners was among the panelists who spoke on getting an app discovered.  This panel was almost a counterpart to the "Candace and Vijay Show" because they mentioned the utility of getting validation from thought leaders before getting you app into an app store.  VC-backed developers have a special advantage because their VCs can asked the major app stores to feature the app.  Candace and Vijay would call that "hacking your pedigree."  I cannot overemphasize the importance of showing KPIs like LTV and CPA to VCs!  Those VCs review broad averages for app LTV and CPA to determine whether app companies can be profitably acquired.  The wide availability of good user acquisition (UA) tools and ad tools means app developers can use video ads, promoted tweets, and other integrated attribution layers that will help avoid duplicative ad buys.  I just don't think having someone wear an orange elephant suit on the trade show floor will accelerate discovery but that's what one exhibitor did.  I didn't stop by their booth to find out if it was working.

There was a brief unscheduled session on getting an app funded that featured Sergey Brin's brother.  The dude is in some startup but I didn't hear a whole lot that isn't already widely known about raising capital.  He did speak highly about getting on the top crowdfunding platforms, which I track for my own private use.  He said AngelList and Y Combinator matter to VCs, so the traditional VC reluctance to endorse new ways of funding and training startups is totally breaking down.  Brin the Younger pretty much endorsed the marketing mix 4Ps without explicitly saying so.  Everything old is new again.

Wing Venture Capital was present for the panel on mobile platform wars.  It's clear to me that platform-agnostic technology and HTML5 compliance will have clear enterprise adoption advantages.  Startups should take note of that before starting.  They echoed things I heard at mobile conferences two months ago when they said CIOs can easily accept apps that comply with BYOD policies.  Hybridizing business and personal use of tech makes enterprise adoption easier.  I learned a new word combo:  "Stackable / glanceable" means tech that describes anticipatory content pushed to users.

Mobile Monday Silicon Valley offered up a master class on app marketing strategies from 148Apps.  The apps with the most successful conversion rates solve problems unique to mobile, such as combining geolocation with micro-tasking.  Automatic sharing and friend referrals are an underutilized but worthwhile app marketing feature, worth building into an app at the start.  Public events like this trade show also work in getting attention.  The panelists liked BuzzFeed's viral creation ability but I have no experience with that platform.  They also endorsed A/B testing of app icons and buttons, and I've heard others say that Twitter is a good way to do this with a large enough follower base.  They hinted that a red button is the best conversion-generating color.  Lots of analytics tools are best and Google, Flurry, and others have stuff we can all use for free.  The panel likes the concept of a video showing a user interacting with the app if it concludes with a strong call to action.  The panel shared one very disquieting insight; they noted that LTV is approaching CPA across the mobile app universe.  This implies the aggregate of apps will soon be unprofitable and app developers will have serious difficulty succeeding with apps as stand-alone business models.

One unannounced panel on transforming an app into a full startup was a good segue from that last panel, and noted attorney Roger Royse served as moderator.  Someone mentioned a local social event series but I'm not sure whether they meant Startup Monthly or Startup Socials SF.  Well, you can never have too many networks.  The typical formula of product/market fit and entrepreneurial passion apply to this kind of transformation.  I say it also takes a scalable model that evolves into a Buffett-style durable competitive advantage.  That means a combo of market leadership (in a market of any size), high barriers to entry, and high switching costs.  One panelist mentioned BJ Fogg's habit-forming design work as worth a look, so I'll have to read up on his Fogg Method.  I was pleased that the panel admonished entrepreneurs to compare the ROI of their startup to the ROI of their present occupation and the opportunity cost of not working.  They drove home once again the importance of the LTV/CPA comparison.  That is so crucial that many of the experts here mentioned it.

I missed the morning sessions on the second day due to a previous commitment.  Chartboost's session on day two was all about a good user experience.  Here come their top tricks.  Upgrades that add clicks, increase hurdles, or degrade the user experience will ruin your app store rating from users.  Do A/B testing by changing the title of the app's listing in the store.  Measure your retention numbers and get your sector's benchmarks.  Only launch a 100% complete app.  Users won't tolerate an app that's 80% functional.  Don't think you can iterate such a partially completed app because you'll never get adoption.

Roger Royse returned to moderate another monetization session.  I should not have been surprised to learn that most apps use standardized design patterns, but I'd sure like to see some knowledge center where these common patterns are categorized.  Maybe the Fogg Method will shed some light there.  The panelists didn't come right out and finger IoT but they did see a big opportunity for apps that can talk to remote devices.  Oh yeah, graphics and photos make an app look good even with a simple layout.  All I can say about that is that humans are so visually oriented that playing to the lowest common denominator of perception always works.

One roundtable on disruptors who transform enterprises revealed the perils of in-house apps.  Many apps contain rich content that can impede server performance, and the enterprise IT architecture must account for how fast apps will load without a full download for each use.  IMHO that just means enterprises need to move at least their densest apps to a public cloud, but I'm not a CIO.  I don't yet buy the roundtable's argument that mobile will be the dominant platform in enterprises.  Maybe those enterprises whose workforce is highly mobile, with lots of field service reps heading out to remote sites, will need it to be dominant.  A lot of people are still going to be chained to desktops for a while.

Google's AdMob reps came out to give their session on how their platform will master localization.  They define internationalization quite differently from localization.  "Internationalization" is a code base designed to be adaptable to many markets.  "Localization" are language and details configured for a specific market.  They also gave us a hint as to why LTV is converging toward CPA and making the app sector unprofitable.  It's the 85% of in-app user spending for things other than the app's features and services that degrades the LTV of those users.  The placement of ads matters.  Users don't mind in-app ads as long as they disrupt the user experience.  That means ads should go in between instances of use (like game levels) that don't interfere with content.  I can't believe that so many app developers don't design in their monetization strategy from the get-go, but some people at this conference needed the reminder.  Google's Cloud Platform is ready and waiting for developers to jump in.

The final session of the conference from Fiksu covered marketing campaign planning.  Fiksu presented data showing clear seasonal patterns for app store adoption regardless of sales volume.  The cost to acquire loyal users (multi/serial downloaders) rises during holidays and drops off in January.  Fiksu's Indexes, ebooks, and other resources cover this phenomenon in more detail.  They left us with three conclusions for planning a campaign.  First, plan ahead for seasonal ad spending around holidays and sports seasons.  Second, link the ad campaign's goal to strategic KPIs:  increase total users, decrease CPA, etc.  BTW, the app's rank in a store is a poor goal because it's too volatile over the short term.  Finally, choose a strategy of either volume or value.  Volume means driving large numbers of downloads.  Value means a high LTV for repeat users.  Either strategy requires placing a dollar value on customer acquisition costs.  Fiksu thinks January and February are the cheapest months to buy ads, and that app developers must time their app submissions around the "freezes" when the app store's display is unchanged.

Allow me to wrap this up by linking those two Fiksu campaign strategies back to Facebook's opening description of two product types.  The positioning of your app and its product/market fit will IMHO determine the campaign strategy you select.  Viral growth apps for a large base are monetized with ads; these demand a strategy of volume to drive large numbers of downloads.  Utility apps built to be sold are monetized with paid subscriptions, paid upgrades, paid unlockable features, and the like.  These demand a value strategy seeking high LTV customers.  Alfidi Capital has thus spoken and rendered profound wisdom.

Wow, that was an action-packed conference.  They said up front that it was oriented for a business audience over a technical one and they weren't kidding.  There are enough links to free resources to equip an entire academic major in marketing.  App developers need to take campaign design, ad spending, and analytics very seriously.  Those who don't are just playing in a sandbox and throwing away VC money.  Like Donald Trump says, it's just business.  

Social Intelligence Data Analytics Real-Time Big Data Marketing Disruption with Domain-Specific Use Cases

I attended a Meetup last week at Hacker Dojo in Mountain View.  I gave this article the longest title in Alfidi Capital's history because that's the only way I could incorporate the concepts announced in both the Social Media Marketing Monitoring Engagement SF group and the Frontier Big Data Cloud Weekly SFBay group.  The more Meetups I attend, the more content I absorb.  This stuff crosses so many boundaries that only a handful of geniuses such as Yours Truly can track it all.

Chase McMichael from InfiniGraph gave his talk on how marketing intelligence can find use case data in social media.  Check out his slides on SlideShare.  He observed that data harvested from outside your own organization isn't "clean."  Dirty data contains noise from inaccurate or irrelevant sources that IT managers must sort through a labor-intensive data cleansing process.  Chase thinks large enterprises lack integrated plans for aggregating and sorting Big Data.  His ideal business intelligence plan proceeds from reporting to analysis to monitoring with predictive analytics as the top layer.

Let's pause to reflect on Big Data integration.  My use of Google Analytics and Webmaster Tools pales in comparison to the challenges larger enterprises face.  IMHO DevOps people must increasingly design apps with data collection goals and analytics in mind.  If this emphasis isn't embedded early in the software development cycle, the finished app won't produce clean Big Data that feeds an ERP.  Big Data demands a redesign of CRM/ERP integration.  CRM will generate data marketers can use to adjust Customer Development, and that CustDev must drive the predictive analytics that will become ERP resource use forecasts.

I'm getting ahead of myself here, so let's get back to Chase's wisdom from his seminar.  He noted that Apple's app strategy gave its iPhone the edge over BlackBerry because the app ecosystem generated demand for the device's adoption.  I kept thinking about how companies track these ecosystems to look for demand trends.  Chase thinks that content marketing depends on trend discovery, a Big Data problem.  I now think that scoring an ecosystem's engagement with your enterprise's content is a key part of a Big Data strategy.  The real-time tracking aspect gets hard for companies that aren't savvy in social media.  User interactions in social media leave data fingerprints and DevOps people should build algorithms that track them as use cases.

Chase argues that Big Data analytics mixes anthropology, social science, and statistics.  His preferred "real-time marketing" reacts with relevant messaging to data derived from demand.  Now I see his earlier point about the ineffectiveness of labor-intensive IT methods.  Machine intelligence creates scalable data that does not require intensive human labor in its production.  The automation of Big Data's demand-driven response is what makes it scalable.

I grok this concept of automated data-driven marketing responses but I also heard similar things back in the late 1990s when I was on active duty in the US Army.  Military technology developers were excited about fielding the "Force XXI" suite of systems that were going to automate away all battlefield confusion.  These systems are now mature but the fog of war is still present on the battlefield.  Even the RAND corporation was skeptical about Force XXI's assumptions while the program was at its highest visibility.  Uncertainty in a competitive environment never disappears because executing a strategy requires human judgment to ensure the OODA Loop has the correct orientation and that progress tracks the correct milestones.

Big Data has a role to play in automating the marketing data collection feed into analytics, just as it can play on the battlefield in automating intelligence data collection.  IMHO, Big Data poses a knowledge management taxonomy challenge to large organizations once they've sorted the data streams.  My point is that humans cannot ever be fully removed from the OODA Loop, especially those C-suite executives who are responsible for ensuring the Orientation correctly reflects the enterprise's strategic posture.  Humans designing these systems can't just turn everything over to Hadoop and hope for the best.  The end result in a private enterprise would be like the US's McNamara Line in Vietnam where dirty data corrupted automated decision-making.  The human beings working on corporate CRM / Big Data / ERP integration task forces need clear guidance from the C-suite on a KM taxonomy that will prioritize the types of data that get automated.  The strategic guidance should also name the program managers or geographic region managers who will own parts of the automated decision-making cycle.

Chase finished with some hints on Googling ".xls" and other terms to see just how much Big Data that corporations have released is already in the public domain.  He thinks RSS feeds are a universal standard for content publishing (with visual content being especially powerful), and we should use them to collect content for scoring.  He also thinks we should check out InfiniGraph's SMO portal.  I think we should all check out IIA's Analytics 3.0 while C-suite folks should attend the Chief Analytics Officer Summit.  Those resources will given enterprises a start on developing guidance for DevOps and KM integrators as they start automating data-driven marketing responses.

Well done, Chase.  You got me thinking.  BTW, these Meetups make it cool to be a technology geek.  I saw several very attractive women at this Meetup, including one hot blonde Russian chick named Olga.  Hey Olga, send me your phone number and let's talk about Big Data at your place.