Overcome Your AI Barriers

Businesses see the tangible value of AI and are racing towards creating their first AI project. Business owners, Chief Information/Technical/Digital Officers, Data Specialists, and Designers need to understand the barriers they must overcome to realize a successful AI project.

You have likely felt the buzz in the business community about artificial intelligence (AI) – how it is transforming every business process, from sales and marketing to customer service, and throughout the entire supply chain.

Despite all the talk, only one in five executives have deployed an AI solution to support core aspects of their business. Most of this has to do with how alien AI technology is to the existing Chief Information, Digital and Technical Officer of a company. Most businesses do not know where to start, how to manage for success and then how to drive adoption so the business is onboard with AI as a game changer for their industry.  


 - The 7 Common Barriers to AI Adoption - 

1. I cannot find the people across all the required roles of AI

While it is possible to build an AI Innovation Lab organically, in today’s world hiring the right AI team is next to difficult, time consuming and expensive, and keeping specialists is near impossible.

You first have to know that AI is not just hiring Data Scientists. As fields mature, you see specialization of roles.  For example to deliver a successful AI project you can identify up to 7 roles needed.

  • AI Architect is somebody who knows the domain and is able to use the art of feature engineering to extra the hyper-parameters necessary to optimize a deep learning solution.

  • Data Engineer who knows how to extract, transform and load data across different systems, formats and networks.

  • AI Software Engineer is somebody who knows how to setup, configure, manage a TensorFlow implementation

  • UX Software Engineer is somebody who knows how to visualize data as well as present the right user experience for users to experience and/or interact with an AI solution. See AI is the UX to gain an understanding of how AI can enhance the user experience.

  • AI Scientist or Data Scientist is somebody who knows the details of mathematical algorithms and along with the AI Architect, are able to select, customize and optimize specific algorithms to specific problems.

  • AI Domain Consultant or Project Manager who has to negotiate with all of the stakeholders, keep all parties informed and ensure all parties have the tools and information to do their job. A good AI PM understands the entire AI lifecycle.

  • AI Ops specialist is somebody who knows how to setup, manage, optimize and support AI solutions in the cloud.

You should give yourself a budget of $250,000 to staff a minimum team:

  • Staffing costs for a team should include possible attrition as you build the team,

  • Equipment costs,

  • Cloud service costs,

  • Training costs, 

  • Initial project cost to validate learnings and get team working together.

Most skilled people will want to work for a successful project and might get bored if there is not enough work. The other problem is we see teams going off and doing make work projects because the leadership team did not correctly understand AI and did not understand how to manage an AI team.


2. I don’t know “my” steps required for a successful AI project

Yes you can go to the internet and find the 7 steps for an AI project. However you need to know your steps for your organization so that AI sticks. Once you consider the specific dynamics of your company, you can then create an AI solution that is tailored for your organization and industry.

The general phases from a Design Thinking perspective are: Empathize, Ideate, Define, Prototype, Test. You will want to follow this with an active Adopt/Adapt phase.

The steps when you bring Design Thinking and AI together:

  • Empathize: Master the domain

  • Define the Problem

  • Ideate

  • Prototype

  • Test by building an AI MVP:

a. Define the Test Plan

b. Data Acquisition

c. Data Cleansing

d. Data Visualization

e. Feature Engineering

f. Dataset Preparation

g. Model Selection and Development

h. Training and Optimization

i. Evaluation

j. Prediction

  • Adopt and Adapt
  • Across the phases you have to consider:

    • Privacy

    • Data Retention

    • Legal Compliance

    • Security

    • False Positives and Negatives as your results are stated in terms of probabilities and outcomes and not as absolute facts

    • Capital and Operating Costs

    • Data Integration

    • Cooperation and voting with-respect-to other AI and human systems


3. I just learned how to manage software code; now you are asking me to think in-side-out and to think about data being the new code?

The biggest learning is for the Executive and Information teams to understand that AI is not deterministic code but rather a statistical process that is susceptible to imperfections and false positives based on ever evolving data and data models

AI learns from patterns of data. You will spend more time defining, training and tuning your model than writing code to optimize a specific algorithm.

Garbage in leads to garbage out. The predictions of your deep learning model depend on the quality of data you use in the training process.

  • Training Data

  • Evaluation data to validate your training process. Your intellectual property is in the training process that creates a usable data model.


4. I don’t have the infrastructure to manage an AI project

  • Tensorflow from Google is the largest and most popular AI framework in the cloud.

  • AI, like Blockchain, is processor and energy heavy. If you go into the cloud you will spend a significant amount of money training your model/engine.

  • Surprisingly the infrastructure to create and train your model is bigger than your runtime environment where you run your model, depending on your model selection.

  • In some cases it might be better for both cost, time, moving of data and security to have a local high performance system that does not have to maintain 99.999% uptime for the development and training environment and then having your TensorFlow runtime in the cloud.   


5. I don’t know the processes and tools to define, create, refine and drive adoption of an AI project.

You don’t know what you don’t know. If you start organically, your first set of projects are for you to learn how to get your team to work together, select your tools, define your unique process and drive adoption with your users.


6. Once I deploy the AI algorithms, my organization does not understand that I have just delivered a learning system that will improve over time; it is not perfect upon inception. Just like a new baby coming into the world, the model has to learn.

AI is a statistical predictor where false positives are expected.

The organization needs to see existing AI solutions in action and realize what they are getting is a system that might automate their best person to drive efficiency but, that the recommendations coming from the AI may be susceptible to the flaws of learning, imprecise data and imperfections in the model. All these flaws can be improved over time with continued use and tweaking.


7. My organization wants AI but does not want to, or does not know how to, train the AI continuously to keep the model current and correct. Just like a child you can train incorrectly and end up with “AI gone bad”.

One thing organizations have seen over time is a trained working AI, that is not maintained, will start to degrade where accuracy of the predictions decrease. AI models must be maintained. Training has to be reinforced. New learnings might require updating weights and biases, or even updating the model with new layers.


There are other barriers, that will flush themselves out once you get past these first 7 steps.

The best place to start is find the right partner who has the experience, team and confidence to overcome the barriers.   


Choosing the AI process that is right for you:

If you would like to learn more on how AI and machine learning can transform your organization, reach out today to set up a consultation. Liquid Analytics with our unique Design Thinking for AI offerings can jump start your AI project and team.


Liquid Analytics can:

  • Create AI Architecture

  • Setup your AI Innovation Lab

  • Provide Executive Training

  • Assist with staffing and team building with our unique learn-by-building offerings

  • Deliver an MVP using DTAI, or Design Thinking for AI. DTAI is a guided project to help you and your team through your first AI project

  • Help you select a team, or partners or consultants to deliver a successful AI project

  • Audit, guide, train your team

  • Provide AI Operations Services (AIOps) for an existing AI project

Screen Shot 2018-07-13 at 3.33.28 PM.png

To make AI work for you, your company must be agile and flexible enough to adapt to new business models, new team models, and new workflows across all departments. The best AI will compliment, enhance and augment your best people. AI will significantly change the way you do just about everything. Be the disruptor in your industry and change the world with AI.