What AI can and can’t do
Ontology meaning used in this book is a set of concepts and categories in a subject area or domain that shows their properties and the relations between them. There is no AI without IA, where IA is information architecture – another term for ontology. The book describes a bottom-up, data- and content-centric approach to creating ontologies.
We are at an inflection point in human history. AI will change everything really soon. But meanwhile my customers want lower prices while venture-funded startups are trying to steal those customers from me.
We know how we want our companies to work. Enterprises ought to be customer-focused, responsive and digital. The elements that are required to make the shiny new technologies live up to their promises require hard work that is not sexy and shiny.
Main reasons for failed AI initiatives:
- Technology vendors have sold »aspirational capabilities« – functionality that was not yet in the current software.
- Overestimation of what was truly »out-of-the-box« functionality.
- Over ambitious »moonshots« programs that were central to major digital transformation efforts but unattainable in practice, or existing organizational processes incompatible with new AI approaches.
AI starts with IA. An ontology reveals what is going on inside your business – it’s the DNA of the enterprise. An ontology is consistent representation of data and data relationships that can inform and power AI technologies. In different contexts, it can include or becomes expressed as any of the following: a data model, a content model, an information model, a data/content/information architecture, master data or metadata.
While customer-centricity is ostensibly a major focus of most enterprises today, in many cases the understanding of customer experience is immature, or is siloed in parts of the organization that “own” one aspect of customer services or customer support.
High-fidelity customers journeys are the key to successful AI initiatives such as personalization and campaign optimization. It is not enough to be customer-centric. It must also operate quickly. Getting products and solutions to market faster requires faster decision-making, faster feedback, faster iterations and faster experimentation – and each of these requires faster information flow.
Three most common problems that stop corporate organisms from competing effectively in the economic ecosystem:
- Friction and Siloed Communication
- Incompatible Data and Language
- Junk Data
Tagging is a central part of what makes ontologies able to speed the performance of enterprises. Important information needs to be flagged, tagged and held up as meaningful. Tagging or separation of signal from noise, can happen at multiple levels. Not having the right tags causes meaning to be lost – the noise drowns out the signal.
There are three areas in which your company needs to rethink its resources, focus and attention so it can exploit the power of AI:
- Data – this includes how data is architected, managed, curated and applied.
- Technology – appropriate technologies must scale processes by relying on detailed enterprise knowledge.
- Operationalization – this includes rethinking how the organization delivers value through end-to-end digital processes.
Many of the AI technologies on the market are actually Band-Aid solutions that try to make up for our sins in data and content curation.
The ontology becomes a reference point to inform where information structures need to be harmonized and where there can be (intentional) differences.
The different parts of the organization – internal systems and processes – move at different speeds and change at different rates. ERP systems are relatively stable and do not change frequently. At the other end of the spectrum are social media applications and programs that change extremely rapidly. Business always changes at a faster pace than the internal IT organization can support. Even if IT could keep up with the best-if-breed tools for the business, people would not have the capacity to absorb change that quickly.
Decisions in each of the three areas – data, technology and operationalization – need to be grounded in the organization’s strategy for how it will serve the customer and must be based on data-driven approach not trial and error.
Ontologies capture more than data representation. They represent consistent fundamental organizing principles. In addition to allowing systems to talk to one another, they also increase speed of information flow, they enable data, content and information to be integrated, they streamline end-to-end processes, they permit the aggregation of data sources and they stimulate faster adaptability in a rapidly changing ecosystem. They allow for the contextualization of insights, the cataloguing of metrics and the creation of feedback mechanisms for continual improvements.
Building the ontology
By providing a structured way to interpret corporate data, an ontology does for knowledge what the chart of accounts does for financial information. Once you create the framework for the ontology, you can get more from your current investments in technology and apply emerging artificial intelligence techniques to drive your business.
The term ontology refers to a domain of knowledge and the relationship among different concepts. Ontologies are built up from taxonomies. A taxonomy is a clearly defined hierarchical structure for categorizing information. Taxonomies and ontologies have different purposes and uses. Taxonomies describe each concept and the ontology consist of all the taxonomies and all the relationships among them.
Ontologies begin as a holistic understanding of the language of the business and the customer and are then designed into processes, applications, navigational structures, content, data models and the relationships between concepts.
The ontology becomes a reference point for the organization that provides consistent naming, structures and standards for applications and new technologies. It becomes source of truth.
Before we can apply AI, we need two things: a core understanding of the business problems we are attempting to solve and mechanism for identifying the important signals contained in data.
If an ontology is not developed within an application-centric framework, it will become an academic exercise. There are two ways to begin building an ontology: a “user-and problem-centric” approach, which focuses on the users and their challenges and a “data-and content-centric” approach which focuses on organizing the data.
The user-and problem-centric approach includes nine steps:
- Observe and gather data (pain) points
- Summarize into themes
- Translate themes into conceptual solutions
- Develop scenarios that comprise solutions
- Identify audience whom the scenarios affect
- Articulate tasks that audiences execute in scenarios
- Build detailed use cases around tasks and audiences
- Identify content needed by audiences in specific use cases
- Develop organizing principles for data and content
Two principles of classifying document/data/content/information (metadata descriptors of the content):
- Is-ness – what it is? (description)
- About-ness – what is it about? (classification)
The challenge with data- and content-centric approaches is that they lose understanding of the user’s goals. The challenge with a user- and problem-centric approach is to prevent it from becoming a laundry list of user challenges.
Every information system should begin with consistent language, concept relationships, terminology and organizing principles. The ontology supplies these.
Well-designed customer experience accounts for the customer’s needs and all possible events the customer may experience. Enterprises interacting with customers pursue two conflicting goals: agility and scalability. Agile execution is crucial as the pace of business and markets accelerate. Scalability conflicts with agility in that it typically entails coordination of many tasks and processes that cannot be completed quickly. Agility is based on reacting and adapting. Scaling requires consistency and standardization. It is only through the use of an ontology that the two goals are not at odds with each other.
Many customer journey maps are based on what the organization thinks its customer experience rather than what they actually experience. Preparing for AI-powered customer experience demands a new kind of journey map: a high-fidelity journey map.
A high-fidelity customers journeys are representation of the customer’s needs in data terms. To complete the high-fidelity journey map, you must validate the journeys through primary research. When high-fidelity journey maps are combined with customer attribute models (descriptors that represent customer’s interests, needs, tasks, objectives, role, history, propensity to buy,), these representations tell our AI technologies how to enhance weak signals like a simple keyword search with more contextual clues that piece together what the customer really needs.
To build a roadmap for making technology improvements that will optimize that experience use the six steps:
- Understand and map the customer lifecycle.
- Define customer engagements strategy at each step of that lifecycle.
- Survey and assess existing tools and approaches.
- Assess the maturity of your supporting processes.
- Assess tools, technologies and internal processes with regards to engagement strategy and technology landscape.
- Develop the implementation roadmap based on enterprise maturity and high-value areas of opportunity.
Every customer model has static customer data like industry, roles, interests and dynamic customer data like campaign responses, click through or recent purchases. We also need to define stages of customer journey like aware, buy, use and renew.
Different customers will want different types of relationships. The nature of relationship and your organization’s unique personality in that relationship is part of your differentiation.
Three components: a content management system (CMS), an email marketing technology and a customer relationship management system (CRM); form the bare minimum and in some organizations, they may be all that is in place. You need to evaluate each tool according to how inherently important that class of tools is in a particular stage of the customer journey.
The result of well-deployed customer experience technology is the ability to provide the right information for the right customer at the right time to move them seamlessly through their journey.
Agile digital marketing is able to leverage data and content from diverse systems, processes and organizations in a coordinated and synchronized fashion. Many marketing organizations are still working in a fragmented way. Personalized content for a mobile campaign. Print campaign. Omnichannel marketing programs. Email marketing. Web commerce. Marketing assets and programs are now born digital.
Digital marketers need to become knowledge enablers, champions of data quality and digital architects. The ontology is very much the responsibility of the senior leaders in digital marketing.
Ecommerce can create the foundation for metrics-driven governance, the decision-making playbook that is the cornerstone of a data-driven organization.
Search is now a conversation as well. Search terms are often short, ambiguous and an approximation of the searcher’s real information need. Site navigation is also important. Both processes improve incrementally based on context.
Two of the most common mechanisms for presenting product and content that best serve a particular customer: predictive offers and shopping basket analysis. The fuel for a customized ecommerce experience comes from two kinds of data: product and customer data.
A primary hierarchy is called the product taxonomy.
A shallow understanding of the customer will impact the way that product data is architected and populated, resulting in missed opportunities.
PIM (product information management) systems hold information about products, including product relationships. Best practices for managing product relationships include centralizing product relationship data, capturing and documenting the definitions of relationship types for consistency, creating processes for ongoing maintenance and governance, integrating relationship identification into the item onboarding process and reporting regularly on product relationship by type.
We need to model our users through use of personas. Personas are iconic representations of the different types of customers that we serve. One categorization is first-time buyers and repeat customer. Developing a persona involves the creation of what are variously called customer stories, user stories, scenarios or use cases. Whatever you call them – audiences, personas or customer types – designers use this classification to make taxonomy and customer experience decisions.
We have Explicit and Implicit Customer Data. Explicit, objective or applied metadata are dynamic and faster changing. Implicit, subjective or derived metadata are static and slower changing. In order for the CDP (customer data platform) to function, the different systems have to consistently define attributes of the customer and their interaction using identical or analogous terminology.
The definitions of customer characteristics constitute an attribute model. The model has to be defined to account for all customer data and how it flows among systems. The CDP may also cause the campaign management system to acknowledge that the target is a current customer. Mapping information flow among technologies is a critical element of CDP deployment.
Automation is not magic. It starts with a complete understanding of the path humans follow and the ways a site can react.
Once you understand what drives messaging and design unique engagement components for it, you can experiment. After breaking the messages up into reusable building blocks, the blocks need to be structured and organized through the ontology.
Marketers need to develop a messaging architecture that the AI can optimize across these audiences. Even if the organization lacks the processes to develop a messaging and component architecture, there are many reasons to design potentially personalized content based on a large number of descriptors. If segment data is detailed, you can perform analytics to identify underserved markets and customers.
Machine learning algorithms perform best with a lot of variables. The more customer features we define up front, the better the algorithm can perform.
Complex ecommerce functionality includes elements of product information management, digital asset management, customer information management, content management, marketing and promotions management, email management, social media marketing, shopping cart functionality, payment and order management integration, data quality and governance and more.
The critical ecommerce functions include the following:
- Product launch and item onboarding
- Product information management
- Product configuration
- Commerce and order management
- Payments and pricing
- Data quality
Time to go from one maturity level of ecommerce to another is for large enterprises from six months to one year. Maturity checklist for Ecommerce:
- Supplier onboarding
- Product data onboarding
- Demand generation metrics
- Omnichannel experience
- Self-service metrics
- Content operations
- Personalization strategy
- Digital assets
- Content architecture
The status quo is dangerous and relationship-based selling is insufficient. Companies must invest in systems to support and surround their salespeople and optimize sales processes. Before a customer meets with you, they have used online resources to be more knowledgeable about your organization’s offering and options. This is true especially for the more complex sales that require human interaction. Consultative selling and complex, long-sales-cycle solutions require comprehensive supporting content and knowledge.
Because the entire value chain is defined by the movement of data from supplier to customer, any company’s digital transformation is a data transformation.
How AI can revolutionize sales? AI technologies can be used to research a prospect to determine if the fit the criteria of targets. Many companies use BANT and strategic fit.
Once a prospect meets two out of five BANTS criteria, the inside sales rep tags them as a marketing qualified lead (MQL). Sales moves them through the process. After three out of five is meet, it becomes a sales qualified lead.
To nurture leads efficiently, initial calling and outreach should start with consistent communications using proven templates and messages. The best lead-nurturing programs are also consistent in cadence and messaging.
AI and machine learning algorithms can score prospects based on buying signals and a propensity-to-buy model, allowing sales representatives to focus on the highest likelihood prospects.
One options of using AI for sales support is so called CPQ (a configure-price-quote). Customer focused CPQ helps buyers solve problems and build solutions rather than simply selecting products. It also helps generate leads and track lead conversion.
- Define CPQ business objective
- Define user profile
- Determine architecture
- Set configuration
- Specify pricing
- Specify quoting
- Analyze integration
- Build business case and budget
- Build strategy for change management and socialization
- Understand data challenge
For every 1 Euro spend on software, 3-4 Euro is spent on customization in ERP and CRM projects.
At the heart of all sales-related AI is ontology. One that’s intimately connected to the enterprise CRM’s application. The CRM ontology includes customer descriptors that enable targeting, selection of appropriate offers, bundles, messaging and channel strategy.
Future will belong to the companies with the fittest algorithm, data model and knowledge base, all driven by a corporate ontology, the core source of value in the knowledge architecture.
Complexity, consistency and structured information – these are components that artificial intelligence excels at.
Many organizations consider customer services to be a cost center, but numerous studies are showing the opposite is true. Misaligned incentives and metrics and lack of ownership or accountability get in the way of delivering better service for the greater good of the organization.
Virtual assistants are one technology that can improve customer service. It is faster than humans. They belong to a class of applications that leverage advanced techniques around query processing and task contextualization. Search is the most familiar of these applications. We can talk about basic search, knowledge portal, virtual agent or intelligent virtual assistant.
To make an AI application successful, you must understand the structure of the knowledge it seeks.
Workstreams to Prepare a Bot for Functionality:
- Process analysis
- Dialogue design & intent classification
- Content analysis, domain modeling & ontology design
- Componentization of knowledge content
- Training data corpus development
- Hybrid learning & continuous improvement model creation
It’s only in the last few decades that corporations have networked their internal systems together, with “knowledge workers” collaborating increasingly through technology.
Tools that are easy to deploy and meant to improve productivity have the unintended consequences of fragmenting knowledge and information, reducing productivity as the environment becomes more chaotic. When information is not organized in a meaningful way employees lose faith in the accuracy of the information sources and no longer trust that they have the most up-to-date or accurate information. When knowledge workers complain about “information overload” they are really talking about filter failure.
There is a better way to manage internal information: an integrated knowledge management system. There are three scales on which knowledge varies: by structure, by value and by context.
A collaboration site is a place to accomplish work rather that a place to access and reuse that work. Once the output is complete, those who produce it promote it to a repository and refine, review, tag and structure it for the various channels and audiences that will consume it.
Unfiltered, uncurated and unstructured content is easier to create, but it is less valuable.
Explicit and deliberate tagging makes it easier for people to find the structured, high-value content in document repositories. Much of machine learning concerns pattern recognition and the identification of what are called “features” which are ways of describing the information.
Knowledge can be tacit (based on experience) or explicit (captured and written down). A procedure is explicit, whereas an experience is tacit. Tacit knowledge sometimes does not make it out of the heads of engineers and into a form that is consumable by downstream systems. When no one captures this knowledge, the result is inefficiencies, duplicate effort and wasted time. So how does tacit knowledge get into the knowledge base?
AI can help with tagging content with metadata. In large collections a statistical classification approach can also be effective. The ontology is the source of predefined terminology. It helps improve data quality and consistency; it is the source of truth that developers, merchandisers, users and systems can look to for the most up-to-date reference data and lexicons. The ontology is the source of all the information architecture elements that comprise content models.
Employee journeys can be classified and tracked just as customer journeys can be. The availability of knowledge within the enterprise directly affects the customer experience. That’s why it matters.
Many organizations resort to “acts of heroics” to enable these processes, with employees, reacting to a series of crises, working late nights or weekends to accomplish their work. But acts of heroics do not scale. Inevitably the process breaks down and comes crashing to a halt with an even bigger crisis that people need to respond to.
The alignment of internal metrics and external experiences provides more fuel for knowledge management approaches to make workers more productive.
The key to building the ontology appropriately is to start with a narrow focus on the most important issues, questions and problems. This ensures that the ontology will be practical. It then expands over time to cover more use cases, actors and scenarios.
Ideally you can map data to processes, outcomes and business strategy.
Organizations need processes to improve content quality and eliminate ROT (redundant, out-of-date or trivial content).
Physical meets digital
For any company that makes physical products, ship things or maintain physical spaces like stores, schools, hospitals or hotels, the connection between data and the physical world is a crucial part of how AI can advance the enterprise.
Manufacturing is the application of knowledge to the material world. A product is the application of knowledge to matter through approaches and methods that have been developed and refined over time. Engineering is the application of knowledge – science, mathematics and empirical evidence – to the innovation, design, construction, operation and maintenance of everything. product innovation is data innovation. It is human creativity combined with expertise, analytics and know-how from across fields and industries.
Digital twin is a virtual representation of the product or process with its simulated performance and behaviors.
We can use AI for quality control or even making new molecule possible. The sequencing of the human genome has opened enormous opportunities.
AI applies through product lifecycle in stages like:
- Rework decision
- As-built records
- Product ideation
Product lifecycle management has been supplanted by product lifecycle intelligence (PLI), using analytics to improve results of innovation. Some scenarios like:
Many organizations with complex trading partner networks are building standards that enable transparency and traceability throughout the entire supply chain.
Machine learning and AI applications can make sense of resource management inputs and parameters and can help to identify anomalies.
The tracking that comes from instrumenting and tracking physical objects enable companies to monitor and improve their strategies in real time.
Strategy and governance
AI can actually help you manage your business? A technology vendor called Leaderspace, focused on business transformation, is attempting to put that theory into practice. AI is going to be ablet to tell us what’s wrong with business and how to make it run better and faster than ever before. It’s just a matter of applying the right technology, tools and intelligence to the problem.
AI-powered strategy combines human creativity with data insights and large-scale analysis, pattern identification and prediction enabled by AI-technologies. Human still need to decide what is important to other humans; human judgment, knowledge and experience are still critical. Human creativity, combined with the power of computer technology, continues to build more and more capabilities.
The IoT will enable companies to record, process and analyze information from our environment and act upon it. Organizations need to think creatively about what they would do with data about their products, potentially even expanding to new businesses outside their core.
In any enterprise, one thing needs to remain constant – a mechanism for allocating resources, solving problems and ensuring a return on investments. It can be applied to any IT, data, content, knowledge, personalization, search or customer experience program.
Effective AI governance depends on tools that include governing bodies, decision-making procedures, feedback metrics, a RACI chart, reporting tools and templates, an oversight committee and working groups and a training and change management program.
While governance is not exciting, sexy or terribly interesting to many, it is the glue that holds programs together and ensures that projects get appropriate resources and deliver an acceptable return on investment. It is used to hold groups and teams accountable and to monitor progress in areas that may be entirely new to enterprise. This approach, combined with the metrics framework tied to the customer lifecycle (or product or employee lifecycles), is the best mechanism an organization can adopt to ensure AI program success.
The term moonshot denotes a highly risky, expensive, game-changing project. There is no clearly defined path to the outcome and the means to get there may have to be invented. The only thing that is clear it the vision of outcome. There is and advantage of creating a vision unconstrained by the limitations of today’s processes and systems and to using that vision as catalyst for constrained planning. Vision without execution is delusion.
While IT organization can provide infrastructure and instrumentation, the business side needs to develop metrics and key performance indicators as well as remediation playbook based on its knowledge of customer needs and how to optimize the delivery of the desired customer experience.
IT needs to educate the business units about how to own their data. It should also provide the company with the correct initial design, tooling, instrumentation and infrastructure.
An AI roadmap is an approach to defining and managing the transformation that will accompany AI efforts. The roadmap begins with an assessment of the maturity of the business across multiple dimensions today and moves on to a definition of a future vision. Then the systemic gaps need to be identified and filled if possible. Orchestration is about maximizing organizational efficiency and achieving scalability for the new digital business model.
Organizations are measured on a quarterly basis, but journeys to new capabilities take years. Leadership changes. Businesses are reorganized. Stakeholders become fatigue. The journey requires resilience and endurance. It’s a marathon, not a sprint. The key to success on the journey is to define what is needed to support a future state, to benchmark where you stand today and to measure the level of maturity.
One goal of your digital transformation must be maturity in understanding and serving the customer throughout their journey.
Think about Maturity model for content optimization, Maturity model for customer information, Maturity model for product information, Maturity model for knowledge management, Maturity model for orchestration.
Why projects fail:
- The wrong approach to “moonshot” projects
- Unrealistic expectations and budget
- Overpromised functionality
- Incorrect resources devoted to initiative
- Overly broad scope
- Overly complex technology stack
- Inclusion of rebranded legacy products
- Lack of training data
One consequence of the lack of foundational planning is that organizations end up with too many disconnected initiatives, a condition known as “digital fragmentation”.
Here are some principles that can help with this transformation:
- Project a “must do or be left behind” attitude
- Start with measured risks
- Determine what is most important for the business in the short and longer term
- Pursue success rather than avoiding failure
- Set priorities based on multiple dimensions
- Conduct experimental projects out of the spotlight
- Create throwaway projects to acquire learning
- Agility is not an excuse for not planning
- Consider what can be developed iteratively
- Evaluate current state versus future state (and the maturity of supporting processes)
- Develop an orchestration model
- Implement metrics-driven governance
- Assess maturity
- Align technology
- Focus on doing what’s hard and essential versus what’s fun and interesting
- Build the right team
- Grow capabilities
- Plan a moonshot – versus risking everything on a moonshot
- Pursue organizational readiness
- Manage old-school mindsets
- Manage change
- Understand the cost of mistakes versus the cost of experiments
- Manage risk factors
- Build a test bed
- Investigate what’s practical versus what’s possible.
The future is out there and it will be different.