What is AI for business?
AI is not a fad. It’s the news. It dominates the talk of every business conference. And it is the number 1 topic of countless leadership meetings. CEOs are challenging their teams to leverage AI in every facet of their businesses to increase their productivity. gain deeper insights into user behavior, automate mundane tasks, and drive deeper insights with which to make critical decisions. It’s here. The big question is: How can you use it in your business right now?
GPT has catapulted AI into the spotlight, but does it translate into measurable productivity and revenue gains? A recent study by the National Bureau of Economic Research suggests that even in these early innings of AI for business, it’s moving the needle for early adopters.The study’s authors, Erik Brynjolfsson, Danielle Li Lindsey, and R. Raymond, concluded that AI “disseminates the potentially tacit knowledge of more able workers and helps newer workers move down the experience curve.” Our takeaway from the study is that AI is already helping humans do things faster with often better results.
AI provides businesses with numerous automation opportunities, which can help save time and money while increasing accuracy at the same time. Automation technologies such as natural language processing allow machines to interpret human speech or text input without any manual intervention required from humans. This eliminates tedious data entry tasks from employees’ workflows so they can focus their energy on higher-value activities. According to a recent Zapier study, “76% of respondents said they spend 1 - 3 hours a day simply moving data from one place to another. Additionally, 73% of workers spend 1 - 3 hours trying to find information or a particular document.” Additionally, automated processes reduce errors due to human factors like fatigue or distraction, which could otherwise lead to costly mistakes. The same Zapier study confirms this, arguing that “83% of workers said they spend 1 - 3 hours a day fixing errors.” That’s a full workday moving data around, looking for information, and fixing human errors. Imagine what you or your team could do with their day if that were a thing of the past.
The fact is that your company will never generate less data than it does now. The issue is that traditionally, 90% of data generated and collected by businesses is dark—untapped, and often completely unknown. Thanks to applied AI, companies can now use advanced algorithms to easily analyze large unstructured data sets, allowing them to understand consumer and customer behaviors better. This application spans the organization from marketing to engineering, product to customer experience, business intelligence to RevOps, etc. Imagine if these teams had their very own AI teammate that organized every interaction with your prospects and customers and turned it into knowledge and insights to help make everyone’s job easier. Now that’s possible with AI.
Harness the power of AI today. Simple, smart, and safe.
On a more grounded note, AI will have growing pains along the way. It has problems of its own, especially for businesses that need to leverage it to stay competitive.
It’s clear that adopting AI is no longer a question of “if” but “when.” It’s an opportunity facing every leader in every industry. And as we confirmed in our research for this report, there are huge incentives to move quickly. But before you leap into AI-powered solutions for your business, you must evaluate them properly—not only from a technical standpoint but also from a functional perspective.
Let’s be honest, millions of dollars will be wasted trying to “roll your own” AI. Before attempting a bespoke AI project, it is important to understand the challenges.
Cross-modal data integration
Cross-modal data integration refers to the process of combining and analyzing data from different modalities or sources, such as email, voice, chat, CRM data, ticketing data, and more. This is one of the biggest blockers for teams trying to build their own AI solutions. Cross-modal data integration aims to extract meaningful insights and knowledge from diverse data sources that may provide complementary or redundant information.
Cross-modal data integration involves several steps, including data preprocessing, feature extraction, alignment, and fusion, or what we call data joining. During data preprocessing, the data from each modality is cleaned, standardized, and prepared for the AI to analyze. Feature extraction involves extracting relevant features or characteristics from each modality, such as text features. Alignment involves mapping the features from each modality onto a common feature space, which enables them to be compared and combined. Finally, fusion involves combining the features from each modality to generate a unified representation of the data.
Cross-modal data integration is integral for AI for Business. For example, when you want to know the most commonly requested feature for a specific cohort of your customers, you need to combine data from multiple inboxes, systems, your CRM, and other modalities to provide a comprehensive and accurate answer.
Data cleansing and normalization
Data cleansing and normalization are critical steps in preparing data for AI applications. In simple terms—garbage in, garbage out. They are also insanely laborious. AI algorithms rely on accurate and reliable data to make predictions or generate insights. Data cleansing and normalization are critical to ensure the data used to train AI models is accurate, consistent, and complete. Think of it this way, you can improve the performance of AI algorithms by reducing noise like duplicate data and other inconsistencies. This can help AI models to make more accurate and reliable predictions. And cleaning and re-structuring data helps to make AI models more interpretable by reducing the complexity and variability of the data. This can help identify the most important features or variables driving the predictions or insights generated by AI models.
One of the most common reasons AI projects fail is due to data cleaning and normalization or, rather, the lack thereof. Business data can come in different formats, such as email, tickets, call transcripts, and more, each with unique characteristics and challenges. Normalizing and cleansing data across these different modalities is a complex task. Now consider that the amount of business data available for analysis is growing rapidly, making it difficult to manage and process data efficiently. This can be particularly challenging when cleansing and normalizing data, which can be time-consuming and computationally intensive.
To compound matters even more, business data is always stored in different systems or databases, which are infrequently compatible with each other. Just think how many different people are emailing your customers. Daunting.
Data cleansing and normalization are important but challenging steps in preparing data for AI applications. It requires domain expertise, technical skills, and a deep understanding of the data and the business problem.
User interface
Let’s say you can get clean, structured data into an AI or large language model. Now what? Now you need a user interface (UI), so business people can inform decisions and workflows with AI-generated insights. This isn’t trivial. You’ll need another team of developers because the people cleaning and preparing the data aren’t the same people who will build an end-user UI. Now you need to build some more software.
A well-designed UI helps users understand an AI system’s capabilities and benefits and increase their willingness to adopt and use it. The UI communicates the outputs and insights generated by the AI system in a way that is easy to understand and interpret. The UI can provide a way for users to input data, provide feedback, or customize the AI system to their needs. This can help improve the AI system’s accuracy and relevance and increase user value.
Data permissions
What next? Sigh. Don’t forget about data permissions. You are going to need them.
Data permissions refer to the rights and permissions required to access and use the outputs from your AI project. They are a critical component of data governance and must ensure that data is accessed per your organization’s policies. Think of this as who can see and use what.
Depending on the nature of the business and the data being used in the AI project, there will likely be requirements around data access and use. Ensuring that the AI project has appropriate data permissions can help to ensure compliance with relevant laws and regulations, such as data protection laws like GDPR.
Simply put, you must build more software to manage who can see what. Data permissions are often reviewed and updated to ensure ongoing compliance with and changing business needs, so your permissioning software has to be commercial grade.
Automations and exhaust
Insights from operational AI systems involve integrating AI models and insights into existing business processes and systems. You must get AI-generated insights to the humans and systems needing the knowledge. This may involve building APIs that allow the model to be called from other applications or systems or integrating the model into your teams’ existing tools like your CRM, email, etc.
Once the AI model is integrated, you have to automate the generation of insights. This may involve setting up automated alerts or reports triggered based on specific events or data conditions or integrating the insights into a dashboard that provides real-time insights (data UI).
By automating insights from AI systems, businesses can gain real-time, actionable insights that can inform decision-making and drive business outcomes. Automating insights can also help improve businesses’ process efficiency and effectiveness by reducing the need for manual analysis and decision-making. Vernon Howard, Co-Founder & CEO at Hallo, exclaims that with AI,
I can automate 20 to 30% of my work now.”
The takeaway is that if you can generate insights, you must autonomously get them to the right place.
Integration requirements
One of the main failure points for AI-related projects is unsustainable methods of extracting and converting data. In the past, this was done manually by interns, business analysts, and data engineers. Technological advancements like APIs make modern data capture processes instantaneously and consistently. This frees data and BI professionals from arduous entry work, focusing their efforts on more rewarding, core business responsibilities.
When selecting any technology solution, it’s important to ensure that it has integration APIs (Application Programming Interfaces) to enable seamless integration with your other systems and applications. Integration APIs allow different software applications to communicate with each other, exchange data, and perform tasks without the need for manual intervention.
Ideally, look for solutions that build direct integrations to platforms instead of using third-party integration platforms. Third-party platforms offer a fast way to connect systems but often lack configurability and data classification controls. Also adding a third-party data integration solution can also introduce another data processor to consider.
Data privacy
The subject matter experts we interviewed for this report agree that privacy concerns are the number one reason AI initiatives fail to launch. Xiaoze Jin, the Lead AI/ML Solution Architect at Rackspace Technology, states,
We’re in a very early inning of cloud AI as a SaaS offering. Privacy, above all else, is most important. Beyond privacy lies responsible AI/ethics, federated learning, and zero-trust framework security.”
Due to privacy concerns and stringent compliance regulations, functional leaders are often stymied by infosec and privacy teams reluctant to allow access to collect or process user data, preventing them from taking full advantage of AI-driven insights. Yacov Salomon, Founder & Chief Innovation Officer at Ketch, states,
Be aware of privacy, the ethical use of data, and the governance of data.”
He continues,
...the world is evolving, and if ML and AI are involved, you need to scrutinize. Governing around data makes a big difference.”
Data privacy laws are designed to protect individuals’ privacy and personally identifiable information (PII). PII refers to any data or information that can be used to identify a specific individual, directly or indirectly. PII can include any information that can be used to identify an individual, such as their name, email address, social security number, phone number, home address, date of birth, driver’s license number, passport number, biometric data, or any other unique identifier.
Any commercial AI solution should have a detailed and thorough approach to privacy. Ask for a copy. Otherwise, look for solutions that automatically de-identify data. The de-identification process can involve removing or masking certain data elements, such as names or addresses. Of course, there are different levels of de-identification, ranging from removing obvious identifiers to more advanced techniques that involve complex algorithms and statistical analysis. The effectiveness of de-identification methods depends on the type of data being processed and the acceptable re-identification risk threshold. Patricia Thaine, Co-Founder & CEO at Private AI, argues,
The easiest thing to do is remove PII as early as possible in the pipeline... At ingest or as soon as you possibly can in your system in order to minimize risk.”
Pro Tip: Sending customer data that includes PII to large language models (LLMs) like ChatGPT is a bad idea and will likely compromise user and confidentiality agreements with your customers.
Security
AI solutions may require access to sensitive data. Ensure that any solution provider maintains a comprehensive Information Security Management program to manage Sturdy’s systems and products’ security, availability, confidentiality, integrity, and privacy risks. The vendor’s program must be independently audited and certified to meet the requirements of Trust Services Criteria SOC2 Type II. You’ll also want to ensure that all data communications into and out of a platform are encrypted-in-transit. That data is stored encrypted-at-rest using industry-standard encryption mechanisms.
Any solution vendor should have a current SOC 2 Type II for your team to review. This is the most comprehensive SOC protocol and attests not only to the suitability of a vendor’s processes and systems but their operational effectiveness of sticking to those controls over a period of time.
Considering these technical requirements when evaluating AI solutions, you can ensure that the solution is compatible with your existing infrastructure and meets your performance needs.
AI is the future wave, and those who do not embrace this ever-evolving technology in their business are already falling behind. Jin argues,
We’ve already begun to see the paradigm shift, creating a new way of living and working with AI as a co-pilot... And yet, we’re still living in the wild west of AI, with land-grabbing occurring left and right.”
Those who do understand the immense power that AI can begin to automate processes, secure insights and increase customer engagement across multiple channels reap immeasurable rewards. As a CEO and business owner, Howard exclaims,
This is a big deal... this is huge.”
Unsurprisingly, AI has become one of the most aggressive sectors in business today. Still, fortunately, there are now plenty of resources and expert advice on how to safely and successfully deploy AI into your company. It’s time to get ahead of the competition and take advantage of this revolutionary force before it zooms us into an entirely new world of business opportunities!
.png)
.png)
