AI solutions aren’t only a tool approach; it’s about well-understood use cases and ways to measure their impact
14 hours ago
The world we’re shaping around AI includes various kinds of organizations: those constructing sophisticated AI technologies, others developing AI-based solutions, and eventually, organizations that aim to make use of AI for positive impact or to support their businesses. When initiatives to deploy AI within the last group happen, capability constructing and training are mostly oriented towards technical infrastructure, data ecosystems or technical skills; and although its significance is indisputable, we fail to grasp that usually, AI solutions aren’t only a tool approach; it’s about well-understood use cases and ways to measure their impact. This guide goals to be useful to anyone leading AI initiatives and to enrich any strategy aimed to boost innovation capabilities through AI.
“AI is just not about its capabilities and guarantees, but in addition about how its used…” (The age of AI: And our Human Future -Kissinger, Schmidt, Huttenlocher)
Every technique of innovation through artificial intelligence consists of two parts: capturing knowledge and utilizing knowledge. This guide goals to display the strong relationship between each and the five dimensions that compose them (Use Cases, Early Wins, People, Technology, and Governance). Although they will independently coexist, together, they will significantly improve the possibilities of identifying and deploying AI-based solutions to make a considerable impact.
I would love to make clear the intended scope of this guide. There’s a variety of good work on this topic by consulting firms (Deloitte, McKinsey, BCG, Gartner, to call a couple of) and corporations within the private sector or independent research ( Catalyst Fund,Profit.co,Dorien Herremans , to call a couple of). Subsequently, it’s not my intention to present one other bespoke conceptual framework or reinvent the wheel. The truth is, among the steps presented may sound very familiar to anyone leading an AI practice in a B2B tech consulting company. My intention is to maneuver away from the abstraction of a conceptual framework and try to operationalize a set of steps with some tools that may also help firms significantly improve their possibilities of identifying and deploying AI-based solutions to make a considerable impact.
It’s not an AI tool approach; it’s all about USE CASES. Which means to extend our success rate on our AI project, we must discover real problems that affect our end users or the corporate we’re working with. This really isn’t anything latest, as most frameworks around AI strategy emphasize the importance of identifying good business cases as a start line.
This part is what I call “capturing knowledge”, and although everyone recognizes it as a very important step, there may be little information in regards to the “How?” to do it. For this guide, I divide this capturing knowledge step into two dimensions: The identifying process and the prioritization process, which specifies parameters to assist select which use case may very well be more relevant to interact with, and achieve Early Wins.
Find out how to discover good opportunities to deploy AI?
01) Initiatives: What challenges does the industry you’re in face?
02) Use Cases: How is the corporate attempting to resolve such challenges?
03) Stakeholders: Which division/business unit does the challenge belong to? Who decides? Sponsors? Detractors?
04) Insights: With what insights in the corporate are the challenges identified? Where do they arrive from?
05) Data: What data do you’ve gotten available to resolve the challenge? Is it validated? Do you would like more data?
06) Tools: What tools (technology) does the corporate use to resolve the challenge?
Every development follows an adoption curve; technology moves faster than the capability of human beings to adopt it, and far faster than firms’ adaptation to this latest customer behavior. That is type of the essence of the “Collingridge Dilemma”, but it surely’s also relevant for understanding success in AI initiatives.
Trajectories differ amongst firms; large corporations could have more tolerance for research, testing, and failure at the start so as to achieve significant results or radical innovations around AI. Nevertheless, as mentioned before, many organizations are willing to make use of AI to support their businesses but face different dynamics, resembling limited budgets and fewer tolerance for waiting for results. But Early wins aren’t nearly profit or quick success, extracting some concepts from Kotter’s Change Management Framework, it’s about constructing momentum that excites people to pursue a standard vision, to do things they’ve never done before, and to encourage discovery.
Early wins and gains may be viewed from two different perspectives. From a business goal perspective, basic ideas have stayed the identical in firms over time, any project that generates increased sales or reduces costs is at all times an excellent fit. Subsequently, any AI initiative that demonstrates (meaning evidence with measurable data) the flexibility to drive efficiency, enable automation, or make predictions to speed up decision-making processes can be an excellent place to start out. From a Use Case perspective, it’s vital to contemplate that NOT the whole lot must be solved with AI, projects that may’t be addressed through traditional means, are data-rich, or involve large amounts of labor are more likely to be well received. Finally, don’t forget that early executive buy-in is very important; a strong endorsement may be the difference between reaching the deployment stage or getting stuck in the course of the journey.
“Wins are the molecules of results. They have to be recognized, collected, and communicated — early and sometimes — to trace progress and energize volunteers to persist”. (The 8 Steps for Leading Change-John Kotter)
Even though it may sound like a cliché, people and skills are vital, and almost every framework emphasizes it. Nevertheless, while the dimensions and expertise of the team will rely upon the dimensions of the corporate or budget, the rate at which the team identifies potential AI initiatives to pursue and deploy them will exclusively rely upon what I call “Continuous Learning”, inspired by the continuity concept behind practices like DevOps or MLOps and Peter Senge’s “The Learning Organization”. Let’s deconstruct the concept:
The Skills You Have Vs. the Skills You Need: The substitute intelligence field encompasses a various range of skills and roles. As you start to discover good AI initiatives, you might find situations where your team lacks the mandatory skills to execute them. Within the early stages of developing an AI Business Plan, deal with leveraging the talents that enable you to design, validate, and deploy “Early Wins”. Then, as these Early Wins are deployed, credibility is gained, and AI initiatives develop into tougher, transition to accumulate or develop more sophisticated skills.
Awareness and Rituals: AI initiatives are neither a one-shot game nor a one-man show. It’s about constantly feeding the team with ideas to guage and pursue; some will succeed, and a few may fail, but you would like a pipeline of ideas constantly flowing. Primarily, you must have AI Initiatives flowing through three varieties of stages: Planting (stage of research, where the corporate’s Pains/Gains that may very well be resolved with AI are discussed), Growth (initiatives approved to proceed to a design, testing, or validation process), and Harvest (initiatives deployed and able to scale or be replicated).
To ascertain a funnel of AI initiatives constantly flowing through each stage, include in your AI marketing strategy an assessment to discover:
01) How does the team capture AI initiatives?
02) How do the teams collaborate with other teams in the corporate, customers or end users to discover AI initiatives?
03) How are relevant initiatives prioritized? Who determines their relevance?
04) How are latest AI initiatives tested or validated? How is the acquired knowledge documented and shared?
The message is, Get out of the constructing! Organize every day meetings inside the team and workshops with other business units, schedule visits together with your customers and end users (to not sell them, but to grasp their business pains) and conduct workshops with them as well.
Keep in mind that a funnel of AI initiatives is sort of a muscle; a continuous learning culture isn’t in-built just someday. With that said, practices shouldn’t be done only once, but incessantly, so as to transform awareness and rituals into attitudes and beliefs. In the long term, attitudes and beliefs are those that encourage discovery and push you to develop latest capabilities and explore latest grounds where you haven’t yet applied AI. One thing is needless to say, in the event you don’t train the muscle incessantly, ideas will stop flowing.
In technology, budget will probably be a limitation but not a restriction. Fortunately, we live in exciting times in AI development, so for computing, modeling, testing, and deployment, you may profit from either the open-source ecosystem built around AI or the free tiers offered by some service providers (Google, AWS, Azure, IBM cloud, Oracle cloud). While these include restrictions, they may also help with the research, design, and testing stages, which we aim to speed up to validate good use cases for deployment.
So, what we aim to attain is convenience; either constructing something from scratch to have full control over the architecture or leveraging pre-developed use cases and consuming them as a service, either entirely or as a part of a mixed architecture. Inspired by the multi-step strategy playbook for managing digital disruption developed by the IMD-Digital Business Transformation Center, the next dimensions could make it easier to select the very best technology to start out with and methods to scale:
In case you lead a small AI business unit or constructing one which needs to attain “Erly Wins” quickly, perhaps your best choice can be to leverage the open-source ecosystem, pre-built models, or prebuilt solutions. Reasonably than aiming for radical disruption with AI, aim for incremental advantages through the use of what has already been proven and tested. This approach is quicker for validating ideas and designing, testing, and deploying AI initiatives, which is important within the early stages to construct confidence amongst your stakeholders for pursuing later disruptive challenges.
If there may be some flexibility in waiting for early successes (Early Wins), your best bet may very well be to start out from scratch slightly than using pre-existing solutions. While this approach can offer significant rewards in the long run, it also presents challenges by way of managing feasibility, time constraints, and value. Results can wait, but they have to be visible when the time comes.
Be mindful which you could also achieve “Early Wins” when constructing AI initiatives from scratch (It’s all about use cases). For instance, Python has a number of resources for developing supervised machine learning models, resembling forecasting time series or predicting the probability of events like customer purchases, bill payments, or churn models for customer retention (Take into consideration that implementing these models would require your team to be stronger in statistical inference and modeling slightly than technology). In case your AI initiative involves unstructured data like text or videos, tools like PyTorch or the Hugging Face community offer open-source models for projects requiring text processing or video and image recognition. (In case you’re interested, listed below are some examples: this one involves using Python for text processing and sentiment evaluation, while this one utilizes Hugging Face resources for video evaluation)
Finally, while carrying out your technology assessment to your AI Business Plan, there are two considerations you need to keep in mind:
01) Balance between Skills and Technology: The technical skills that your team currently has, in addition to those being developed, must align with the technology needed within the short term for Early Wins and the technology planned for future use. It goes each ways, in the event you intend to utilize a particular technology, be certain that your team has the suitable skills to administer it or the flexibility to learn it quickly. Technical skills may be present in the labor market (depending on the technical skills required, they might come at a price) or developed internally, but requires time, and time is a constraint when pursuing Early Wins.
02) Wide Funnel-Fast Testing: The balance between validated AI initiatives, People, and Technology should lead to a broad funnel of AI initiatives, big in opportunities and efficient in testing speed. The portfolio should constantly include a combination of AI initiatives: Incremental AI initiatives for early wins and a gentle stream of income, Evolutionary AI initiatives to duplicate successful deployments in other markets, and Disruptive AI initiatives to stay on the forefront and anticipate future trends. The portfolio-mix is determined by the tolerance for waiting for early wins. AI is just not a solitary endeavor; it involves managing a portfolio of AI initiatives. The bottom line is to repeatedly expand the funnel and shorten the testing process in order that AI initiatives may be tested and deployed quickly at low price.(For further insights on managing an innovation portfolio, consider reading this text)
A famous quote states, “Normal is an illusion; what’s normal for the spider is chaos for the fly”. Recent years have highlighted the fact that we live in uncertain and dynamic business environments, where resilience and the flexibility to adapt quickly are essential assets. Applied to AI initiatives, which means so as to deploy and scale rapidly, we’d like Machine Learning pipelines which might be efficient, support frequent execution, and are reliable. For data management, good work has been done with frameworks like DAMA-DMBOK and DataOps, and for AI, now we have MLOps.
MLOps: Deployment presents challenges resembling potential issues with the handoff between Data Scientists and IT teams, tracking of the event stage, and impacts produced by changes in data or model drifting. With MLOps, we gain several advantages. First, within the Development stage, it’s all about “experimentation” (data exploration, feature engineering, model prototyping, and validation) while keeping records and knowledge of this process. Second, in version controlling, we answer questions like who?, why?, and the way? for future compliance and reusable components (resembling a feature store for data). Lastly, in monitoring for changes in data or model deviations and ensuring model fairness. You’ll find a useful guide in Google’s Practitioners Guide to MLOps.
Ethical AI Governance: One other issue related to AI governance is compliance. While the controversy on regulatory frameworks is ongoing (with some in the ultimate stages of implementation), firms can begin with self-regulated frameworks for risk assessment, controlled testing environments, development protocols, and ongoing monitoring to attain ethical and responsible AI deployments. In this text, I share some ideas of a self-regulated framework based on the EU AI Act that may very well be useful to your AI marketing strategy. One other precious resource that is important to read for guidance is the Artificial Intelligence Risk Management Framework from the US National Institute of Standards and Technology (NIST).