10 Things You Should Do Differently To Introduce AI Into Enterprise Companies

The adoption of AI in most large companies is slower than expected. We believe the following list can help AI companies and large enterprises work better together.

Guy Ernest


Artificial intelligence (AI) is one of the most transformative technologies of our generation. AI is applicable to every industry and in endless use cases where people are doing manual work. From driving trucks to answering customers' calls, inspecting plane engines, monitoring food safety, staring through security cameras, reading support tickets, and many others.

Nevertheless, the adoption of AI in most large companies is slower than expected compared to the maturity of the technology and the need of these companies to use it. aiOla is focused on unblocking the delay, and we believe the following list can help other AI companies and large enterprises work better together.

10 Ways to Introduce AI into Your Enterprise

  1. Use natural language — Brilliant and talented business people built large enterprise companies. If you want to meet the business people where they are, you need to speak with them in their natural business language. aiOla interface is based on the many advancements in natural language understanding (NLU) and automatic speech recognition (ASR). We believe that we don’t need to teach business people how to use new systems and interfaces, and we should learn (using machine learning) to use the interfaces that people already use, such as text messages and free speech (in the sense of speaking freely).
  2. Learn the (private) business language — Every large enterprise has a unique private language, and every industry domain has a unique set of standard terms and concepts that are not part of the public knowledge. If you try to use general language models, you will miss many of these terms and internal vocabularies, and the level of AI accuracy will be too low to be helpful to these companies. If the AI models are not very accurate in their understanding of the above natural language, the trust of the business people is lost, and the efforts they have to spend on fixing the errors of the AI are more significant than the benefits of the AI. aiOla developed and uses an advanced set of AI models that are fine-tuned to the specific vocabulary of every use case of every company in every domain.
  3. Use very few examples to train your AI — Most AI models are based on vast amounts of samples and enormous data sets to train these general language models. Companies such as Google, Amazon, or Apple can have enough sources to get as many examples as they want and can spend vast sums of money to pay people to annotate these examples. On the other hand, large traditional enterprises are much more limited in their ability to generate similar large data sets. aiOla is focused on utilizing small data sets created by specific domain experts who are rare and limited in time. The better quality of the data and the unique AI algorithms we use improve the accuracy of the AI outputs to meet the business needs of these specific use cases the AI is trained to integrate into.
  4. Augment human intelligence and don’t try to replace it — Some AI companies ultimately promise to automate some human tasks. They promise that AutoML can replace data scientists, Self Driving Trucks can replace truck drivers, chatbots can replace call center agents, etc. We believe that AI technology should be integrated with human intelligence and not replace it. It is not only due to the AI models’ maturity but also because artificial intelligence and human intelligence differ in their qualities. If salespeople are very good at selling and bad at data entry, aiOla AI models focus on facilitating data query and data entry for talented salespeople. Thus increasing the sales by allowing the salespeople more time to sell, with better data from these data systems to augment their already existing experience and knowledge.
  5. Deploy into the enterprise cloud accounts — Many data and AI products are built as a pure SaaS Multi-Tenant model. It can make sense for many smaller companies to adopt an external process quickly. However, larger enterprises have many security and privacy policies and regulations that prevent them from using such pure SaaS services with sensitive data for core production use cases. Suppose you want AI to impact the business significantly. In that case, it should have an “on-premise” version, which is installed in the cloud environment of the enterprise and not only serve as SaaS, where the customer loses control over its data.
  6. Support multi-cloud deployments — Almost every enterprise we meet has some cloud usage and ambitious cloud plans. AI systems are an excellent example of cloud-native implementations, and it is not hard to convince even the most traditional IT people to use the cloud for it. However, each enterprise IT team feels more comfortable using one cloud or another. Since the system should be deployed in their cloud accounts and be under their control and operation, the system should support the major cloud providers (AWS, Azure, and GCP, at least).
  7. Coordinate with multiple IT administrators — The size and complexity of large enterprises make it impossible to have a single administrator that knows everything and can do everything to deploy a new AI service into the organization. aiOla’s product deployment addresses the different administrators, from the cyber security experts to the network administrators, Active Directory team, cloud excellence team, and application administrators. Other companies might have a different split of responsibilities, for example, a dedicated team for Salesforce. However, there are always separated teams, and it requires streamlined coordination.
  8. Allow both manual and automated deployment scripts and operation run-books — Many technical companies adopted DevOps methodologies and tools and try to automate all deployments and configurations to avoid manual and slow processes that can also lead to mistakes. However, enterprise companies must review and test these automated scripts and often prefer to execute some of the steps differently based on their policies and procedures. Some of the IT teams (the cloud team, for example) might choose to run the automated scripts (Terraform, for example). In contrast, others (the Active Directory team, for example) might want to make the changes through their interfaces and not use the automated scripts.
  9. Invest in training of IT personnel — Big enterprises usually have relatively small IT teams to support their massive operations. Sometimes it is easy to mistake the cautious and slow adoption of new technologies as a sign of unwillingness of the IT people to learn these new technologies. However, it is usually not the case. Most experienced IT people in enterprises know the need to learn new tools and technologies. If you get them to discover new knowledge that will serve them in their career growth, you will win champions for your system, who will support you along the way.
  10. Offer modular tiers of support — Plan the support of the AI system in a modular way, with the first level of support by super users and early adopters among the business users. The second tier should be provided by the organization’s help desk, similarly to other digital systems used by business users. It is best to have the third level of support provided by the internal IT people (that you trained as described above), and the vendor of the AI system should handle only the fourth level of support.

If you follow the above 10 guidelines, it will be easier for your AI product to be adopted by both the business users and the IT personnel of large enterprises. The system will pass cyber security reviews, be deployed quickly, and operate smoothly.