Is Your Business Ready for AI?
A walkthrough of AI readiness
Artificial intelligence has become one of the biggest technological developments of the decade, transforming nearly every industry it touches. A recent McKinsey survey found 55% of companies use artificial intelligence in at least one function. With costs coming down, businesses of all sizes are looking for opportunities to incorporate AI into processes. But companies will need to do a fair amount of research and planning in preparation to deploy AI appropriately.
A recent research study of organizational readiness factors highlights five areas in determining whether a project is suitable for AI. These include:
1. Strategic Alignment
Businesses should evaluate use cases for either solving an organizational problem or addressing new opportunities through AI. As part of this evaluation, it is important to consider the relative advantage of AI over other solutions.
Because AI applications can be complex, systems often lack transparency. Each critical member of the project team—from C-suite to customer—must be brought along throughout the processes so that there is a clear vision and expectation of what the AI system can and cannot do. If the AI solution can deliver what it promised, there will be higher acceptance. This customer readiness and team awareness will make future AI projects more plausible.
Strong top management support is also crucial for successful AI adoption. Experts stress companies should not proceed with AI solutions without top management first providing an organization-wide signal.
Changes to incorporate AI may disrupt business processes, especially if there is not already good standardization in place. Mapping out an AI process fit provides the necessary linkage between the proposed AI strategy and organizational processes.
Lastly, strategic alignment requires data driven decision making (DDDM), not gut instincts. DDDM increases AI readiness and is the best opportunity for performance improvement outcomes.
AI adoption budgets can be cost intensive. Organizations are investing in time to build know-how, learning how to incorporate AI, and tailoring AI to a company’s unique context and data.
Regarding AI personnel, business analysts are a relevant part of the team, in addition to AI specialists. These are the professionals that have an abstract understanding of AI and can serve as the communications bridge between technical and business function teams.
Underlying IT infrastructure should be evaluated and developed to include: a large amount of data storage capacity; a strong network with fast processing and ability to transfer data quickly; and the computing power to handle heavy workloads without bogging down.
Perceiving, predicting, or generating are some of the cognitive functions of AI that employees working on the project should understand—at least in abstract terms. With this base level of understanding, teams can appreciate the versatility of AI as a tool and envision its application for their particular industry or project context. Building AI awareness also helps employees hold realistic expectations of AI.
Because of a shortage of skilled AI specialists, companies would be wise to invest in upskilling. This encompasses not only providing employees with knowledge related to AI, such as data management, analytics or engineering, but also domain knowledge.
As part of AI readiness, companies should review and update protocols to prevent discrimination. In AI, this can come through either bad input data or biased learning. The book, “Artificial Intelligence in Human Resource Management,” outlines a good example. If, for example, there is gender bias in the data sets fed to AI for creation of hiring tools, the result could be discriminatory hiring practices. By preventing discrimination at all information levels, organizations reduce liability risks.
Ability to adapt and initiate change at a rapid pace goes to an organization’s innovativeness. Such behavior in an AI project includes experimentation, risk-taking and diverse problem solving. Breaking out of “business as usual” and embracing innovative behavior should be encouraged.
Collaborative work embodies the ability to successfully work together in cross functional teams that include AI specialists and IT departments. Siloed work is a downfall here, but companies can overcome this by deploying various methods of collaboration, bringing together employees with unique skill sets.
Change is hard for many people. With effective change management, staff is better able to understand and cope with organizational changes brought on by AI. One area of concern to address is fear of job loss. AI rarely replaces whole job profiles, but does create efficiencies, taking over repetitive tasks or processes.
Information is at the core of AI, so having the relevant amount and correct types of data are critical.
Data quality relates to its suitability to be used by data consumers. Data quality can be especially challenging when dealing with historical data that may be incomplete. If you feed garbage data in, you will get garbage data out. Focusing on data preparation, processing and quality assurance can increase data readiness for an AI application.
Data accessibility, which ensures quick and easy access, can be enhanced by an access management system. Data can be contained in a centralized scheme, such as a data lake or warehouse. Personnel may be granted authorized access to the data they need.
Keeping things flowing in the data pipeline, or data flow, takes information from source to use. This continuously facilitates the implementation of AI systems and maintains ongoing processing after initial training.
Keep it simple to start with
Starting with the simplest AI solution first is an effective strategy. Simple solutions are efficient, run faster, with high performance and processes that are easy to explain to stakeholders.
In the Domino’s model below, they started with the simplest-first business model. The first development round was a simple regression model that came close to meeting project goals. A decision tree model was deployed next to look at more angles. This was rounded out by a neural net to explore more variables.
As an Amazon Associate, Arvig earns from qualifying purchases.
Good and bad AI implementation examples
Taking an AI-first approach can be incredibly successful, but not for every application.
The Good: Pizza chain, Dominos, has quietly been an innovator in an otherwise traditional line of business. Dominos built up 13 digital methods to order pizza, achieving 70% of sales through online channels in 2020.
Domino’s used AI and machine learning to enhance business operations, provide a better customer experience and route orders more efficiently. In one application, they used AI to drill down on more accurate delivery and pickup times. The time to bake a pizza and the time to drive from the store to the delivery point can be calculated. Even traffic hold-ups can be tracked in real time. This is good data that can provide accurate outcomes.
Domino’s customer satisfaction increased, as did tips for its drivers.
The Bad: Many following business news know Zillow lost nearly half a billion dollars with their foray into “Zillow Offers,” their online home flipping strategy that backfired. What is lesser known is the AI debacle behind it.
For many applications, one can apply a set of data to a curve and determine with some level of accuracy what will happen next. This is using history to inform. But with the unexpected events of real life, these predictions become less reliable. Covid 19, lockdowns and a temporary freezing of the housing market were followed by supply chain problems and a housing supply and demand imbalance that led to a rise in home prices at a rate that was without precedent. Zillow’s AI could not predict or compute these events.
You might think these are once in a lifetime circumstances, but the practice is actually a common AI misuse. For example, when businesses try to create an AI model to predict future revenues, even vastly smaller variables can have a tremendous impact on the outcomes. Without knowing what will happen in the future, humans have to make assumptions when feeding data into an algorithm. The data could then inherit biases that would skew the results.
While AI can still have some insights into certain portions of these types of projects, if there are factors that can’t be predicted or controlled, or factors that the company doesn’t have any data for, AI may not be the right solution. Just like the Great Zorba carnival fortune telling machine, AI cannot really accurately predict the future…yet.