Like every new technology trend, AI is perceived as an expert topic. It seems widely accepted, that every business problem needs research and ends in its own AI solution.
In a previous article, we introduced the concept of “AI Cooking” to emphasize that corporate adoption does not mean to embrace research and scientific work at the same time. “AI Cooking” do not deny that AI use cases exist, where research is important to enter unknown regions. It is an approach for smaller companies to respond to common and recurring AI business cases to mitigate their risk to fall behind.
This article introduces the main features of this future AI working horse for business, derived from recommendations of renowned analysts. We talk about:
- Reusable ingredients for AI solutions
- Fast implementation through code-free orchestration
- AI recipes for rapid solving of recurring and similar problems
- “Predict as you train” paradigm for seamless operationalization
- Case management for a 360° view on AI solutions
- Recipes as an innovative approach to AI white boxing
- Rapid solution deployment through dual use platforms
The next and last article about reusable AI will present the world’s first AI business platform for “AI Cooking”, made to work hand in hand with a reliable AI business process.
Future AI Business Platform
Renowned analysts talk to a wide variety of enterprises all over the world to retrieve a clear picture what business expects and needs when it comes to AI. So why not take a look at what they recommended in 2019 already.
It is much like a feature list of a next-generation AI business platform with a huge potential to become a working horse for fast corporate AI adoption.
Attention moves from research and scientific work to business topics reusability and operationalization. They are key to reduce costs, increase quality, and minimize time-to-value.
Other topics are building blocks and orchestration: With the clear message that the future of AI in business will be code-free and the composition of customized AI workflows will become an everyday business experience.
Renowned analysts describe the next-generation AI working horse as a horizontal businessplatform: A single-stop to address a wide variety of use cases and an end-to-end view on AI projects, from business problem to AI-driven workflows.
To summarize: It is a recommendation to complement todays’ research-centric platforms by a new and business-ready AI platform to answer business questions.
In the remaining part of this article, we pick up these recommendations, and show how to implement them in terms of the “AI Cooking” metaphor.
Reusable Building Blocks: AI Ingredients
The foundation of future AI business platforms.
Let’s start with an example use case: Imagine your customers complain about low marketing conversion rates, poor customer satisfaction and increasing churn rates. And ask you to help with a data-driven solution to create fine-grained customer segments for more personalized marketing messages.
From a high-level view, the marketing solution looks like this: Customer data come from one or more data sources, move through various transformation stages, enter an algorithm to group them by similarity and end up as computed segments in a required data destination.
What are my reusable building blocks?
IT solution providers know that smaller companies usually do not manage their own proprietary data sources. They tend to leverage existing cloud services. Think of HubSpot, Salesforce, Snowflake and many others, including 2nd and 3rd party data marketplaces.
Why should we develop software to connect to these common data sources over and over again? These sources come with structured (cloud) interfaces.
So, build data connectors once, make them configurable and reuse many times. Reusable operators
Existing algorithms are good enough. It may be surprising but there are not so many of them. The way how e.g. fine-grained personas are derived, is most often the same:
Customers are characterized by a set of attributes and a mathematical algorithm computes more or less distinct groups of similar customers. Prior to data-driven solutions, often age and gender were selected to manually derive personas.
Data-driven segmentation is more fine-grained, but it is definitely no magic sauce that makes themealunique: It is a data workflow, composed of reusable data operators to
- extract customer attributes often from equally formatted data,
- prepare for an algorithm and
- compute groups of similar customers.
We use segmentation as an example. But the message holds for a wide variety of other use cases as well: Turning low value data into high value insights is always achieved by set of recurring and reusable data operations.
So, build data operators once, make them configurable and reuse many times.
We define reusable AI building blocks or ingredients at the level of data connectors and operators and consider them as a fine-grained foundation to respond to recurring AI business cases. This is the prerequisite to reduce costs and time-to-value and guarantees a constant level of quality.
Following renowned analysts, the next-generation AI working horse ships with a completely different user experience, inspired by todays’ music streaming services:
Individual pieces can be organized in music playlists creating concerts of a certain sentiment or style. Without being a song writer and composer. These playlists can be interpreted as AI data workflows, and orchestration is an agile method to organize them and respond to ever-changing market needs fast.
It offers a similar user experience to AI than streaming services to music:
Reusable data connectors and operators can be arranged and re-arranged to form data workflows without writing a single line of code.
We love cooking, and from this perspective “orchestration” is a means to customize entire AI recipes or hand-picked parts of them.
Stop AI projects from remaining alchemy and unify AI training & operating phase.
Todays’ AI landscape offers great platforms for data science work and current trends aim to bring more efficiency to scientists. But business impact happens in the operationalization of AI models. This includes model deployment, maintenance and fast integration into business processes.
IT solution providers know that workflows to train and use AI models do not differ much and have many components in common. Different technologies, however, hinder immediate access to trained and retrained models in production.
Key to fast operationalization is the implementation of a new paradigm: Predict as you train. It is a demand to use the same platform, the same technology with the same user experience for both training & operating phase.
Business First with Integrated Case Management
Stop the risk that AI solutions work fine, but do not solve the business problem.
No AI project really starts with the question which workflows to orchestrate and which data connectors and operators to select and configure. The beginning is marked by a business problem or question.
Generating data-driven answers or solutions must be part of a structured business process. But how do we seamlessly connect business and technical perspective?
We believe that (dynamic) case management, a flexible method for structured and also unstructured problem solving processes, is an appropriate candidate to start with.
Adopting the concepts of business cases, desired outcomes and intermediate tasks, and connecting them to AI workflows is an innovative way to provide a 360° view onto AI projects keeping problems and solutions closely connected:
Similar problems can be described by similar cases. Cases break down into reusable tasks and each task is executed by one or more AI workflows.
Integrating (dynamic) case and reusable AI workflow management is a natural next step to reach out a next-generation AI business platform.
Reach more customers with similar problems in shorter periods of time.
Despite the common perception that every use case needs its own AI solution, we observe an increasing number of cloud platforms that share data science results as technical recipes.
This approach, however, gives no attention to business aspects. On the contrary, reusable AI ingredients and workflows in combination with reusable business cases and tasks from dynamic case management define a solid foundation to build code-free, end-to-end AI recipes from business problem to AI-driven solution.
These AI business recipes or templates describe what companies do with their data and why they are solving a certain problem or answering a question exactly the way they did.
It is time to shift focus to holistic AI business templates as valuable business assets.
For IT solution providers with customers who have similar business problems and seek for similar solutions, end-to-end AI templates offer a valuable instrument to significantly reach more smaller companies in shorter periods of time.
White Boxing with Templates
Build trust with comprehensible AI business templates and turn customers into repeat customers.
Back to the common observation that most SMEs are not prepared to embrace AI by their own but need IT solution providers for corporate adoption. But how can a business user trust an AI solution he does not understand and that is implemented by an external solution provider?
References that thousands of projects have been carried out already, even for Fortune 500’s, offers no direct answer to this question. What if the provided AI solution almost explains itself? Every building block describes its purpose, every workflow shows which blocks come first and which follow next. For every task and for the whole business case.
Business users still do not know how a certain prediction algorithm exactly works. But with every deployed solution, users have a chance to understand how things are interrelated to move their data to insights: Based on a knowledge management approach where every AI solution reveals itself as a visual network of its elementary building blocks and aggregated components.
Do not reduce speed when deploying AI solutions.
What is the fastest way to deploy orchestrated and templated AI solutions? Let’s take a practitioner’s perspective and ask, “do we really have to change horses?”
The future AI business platform introduces the “predict as you train” and integrates both training and operating phases of AI models.
Why not continue and integrate construction and operating phases of the entire AI solution as well? With the same platform, the same technology and the same user experience?
This dual approach is the best way for a fast and seamless deployment.
On the one hand, solution providers use the platform to customize AI templates or recipes and on the other hand, business users leverage it to execute deployed customized recipes.
In the upcoming final article about how to implement reusable AI, we introduce the world’s first AI business platform ready to become the next-generation AI working horse for IT solution providers.
Even more: We illustrate how PredictiveWorks. seamlessly integrates with an AI (business) canvas as an end-to-end guide from problem definition down to its reusable AI solutions. All this finally turns todays’ data science work into business-ready project work with accurate estimation of costs, quality and time-to-value.
Originally published at https://www.linkedin.com.