Common challenges in AI adoption
Enterprises across industries face a few key challenges in adoption of AI:
- Lack of adequate guidance or frameworks in terms of what criteria should be used to check whether a business case can be best-resolved using AI techniques.
- At times, due to over-enthusiastic approach to cognitive, we often pick up business problems that are either too vague or are not amenable to solution using AI techniques.
- Our selection criteria usually weigh very heavily in favour of direct manual efforts-reduction related cost-savings considerations. In that process, we often deprioritize business cases that have huge indirect cost saving potentials. For example: by improving accuracy and reducing human errors, we can reduce rework which is usually at least double the cost of the work done the first time.
Significance of selecting the right business case
Cost take-out is definitely a big win for any automation project, be it simple industrial automation or AI-based, intelligent or smart automation involving machine learning and reasoning. Also, people-costs typically being the most visible parameter, it often over-shadows the benefits of AI in other aspects and therefore business problems where AI could make a far bigger impact in terms of top-line, e.g. QoS improvements, gains in productivity, agility, TTM etc. get deprioritized and put on the back-burner.
That’s why- there is a need to put a simple lens in place that will help both clients and their AI solution partners to drive the due diligence in choosing the right business cases.
A framework for guidance
AI techniques work best if the business cases depict the following filtering criteria, briefly presented in figure 1 below and explained further:
- Need for speed: One great strength of AI is that it doesn’t suffer from fatigue, like human, can work 24X7, can massively parallelize depending on compute availability [which is anyways made possible thanks to GPU, TPU- AI-oriented hardware designs and AWS, Google, Azure types of cloud services]. Therefore, process latency, response times, waiting time, queueing etc. are often no longer the most significant concerns, when an AI-enabled solution is available.
- Need for scale: Business/ IT problems and use-cases where rapid scaling in terms of data processing is required, to generate quick, near-real time insights are amenable to AI techniques. AI modules, again thanks to their massive parallelize-ability, can scale much faster than any other manual options.
- Uncertainties: Solution state-space is defined but has uncertainties and several options e.g. similar incidents have loads of different resolutions available as proposed and executed by many stakeholders at different points in time depending on their comfort levels with specific tech stacks. In Such scenarios, AI modules esp. NN-based classifiers and recommenders can learn and adjust the applicability scores as weights of the nodes, dynamically, based on usage data and success criteria.
- Lack of accuracy: Human error-prone business processes and systems where the process accuracy is questionable can be good candidates for AI & automation. For example, monotonous tasks such as data extraction from multiple disparate sources, checking various data entries and summation formula’s etc., general data entry functions. Repetitive tasks often induce human errors due to lack of concentration and focus on details, boredom and lack of interest.
- Lack of consistency in output and response of business systems, due to human behavioural variations, are also good candidates for AI usage, because the bots behave in a predictable and consistent manner i.e. the same set of inputs, events or triggers will result in evoking the same process and logical event/action chains.
This is not an exhaustive list; there can be various other specific criteria of the problem state-spaces that can be best addressed by AI techniques. These represent a basic set of criteria and attributes that are most commonly seen as the principal factors that determine suitability of AI and can leverage the strengths of AI techniques to overcome certain human limitations, beyond just cost considerations.