How to approach enterprise automation in 2024
How to approach enterprise automation in 2024
AI automation can be domain-specific or enterprise-specific. What degree of customization does your business need? Find out here.
Key Takeaways
AI is transforming enterprise automation, moving beyond simple RPA to offer end-to-end workflow solutions.
Businesses can choose between domain-specific LLMs for specialized expertise or enterprise-specific LLMs for adaptable automation.
AI should augment human capabilities, freeing employees for higher-value tasks while automating repetitive processes.
Transparency, fairness, and accountability are crucial considerations as AI becomes increasingly integrated into business operations.
Last year, I started Multimodal, a Generative AI company that helps organizations automate complex, knowledge-based workflows using AI Agents. Check it out here.
For the past year, as I’ve been building Multimodal and networking with startup founders in the AI/automation space, three key things stand out:
Enterprises are quickly moving towards adopting automation. Most already have some automation in the form of RPA (robotic process automation), and are trying to migrate to AI-based intelligent automation.
There are two main types of startups in the automation space: those that build domain-specific solutions and those with enterprise-specific automation solutions. In both cases, they use large language models as the foundational technologies.
Many startups are building point-automation solutions, but some offer a high degree of customization for end-to-end workflow automation.
If you’re thinking about adopting AI for business process automation, I know it already sounds too overwhelming. So I’ll break down this jibber-jabber for you, helping you make the right choice for your enterprise.
The rise of intelligent automation: RPA to AI
RPA was a game-changer, automating those repetitive tasks that used to tie up your employees' valuable time. But let's face it, RPA has its limits. It's great for rule-based processes, but when things get a little messy, a little unstructured, RPA starts to sweat. It struggles with tasks that require decision-making, handling highly variable data, judgment, or understanding of context. It's like a robot on an assembly line – efficient, but easily flustered when faced with unexpected variations.
That's where the next wave of enterprise automation comes in, and it's not just for the tech giants anymore. AI-powered enterprise automation is all about automating everything that can and should be automated. This isn't just about efficiency; it's about transforming the way your business operates, from top to bottom.
Unlike RPA, AI-powered automation can handle complex processes, understand unstructured data, and even learn and improve over time. It's like having a team of intelligent assistants who can analyze, interpret, and make decisions based on the information they receive. This opens up a whole new world of possibilities, allowing you to automate tasks that were previously thought to be impossible.
Also, AI-powered automation lets you stay compliant in a document-heavy industry. It brings much more flexibility for players in high-regulated industries as well.
Domain-specific automation vs. enterprise-specific automation
Within AI-powered automation, as I mentioned before, there are two models: solutions that cater to an industry or domain, and solutions that go a level deeper, catering uniquely to an enterprise within that domain/industry.
Domain-specific automation: deep expertise for specialized industries
What is domain-specific automation all about?
Domain-specific automation solutions harness the power of artificial intelligence, specifically large language models (LLMs), to automate tasks within a particular industry. These LLMs are trained on vast amounts of industry-specific data that is available publicly, enabling them to understand the unique language, regulations, and nuances of that field. This deep understanding allows them to provide insights, recommendations, and automation solutions that are tailor-made for ALL businesses within the industry. Think of it like having an AI expert who specializes in your industry, ready to tackle the tasks that bog down your team.
Typically, companies in this space fine-tune existing large language models by OpenAI, Google, and Meta to make them very good at handling information related to the particular industry.
How it works
Domain-specific solutions are built by fine-tuning existing LLMs, like those developed by OpenAI, Google, or Meta. The fine-tuning process involves training the model on a massive dataset of publicly available industry-specific information, allowing it to "learn" the ins and outs of your industry.
Examples of domain-specific automation
There’s a major player in the Legal AI space called Harvey. Backed by OpenAI, they have a fantastic legal LLM that can help with tasks that paralegals and lawyers spend hours tackling - filing applications, asking questions, pointing out missing links, etc. Theirs is a domain-specific, end-to-end solution that deploys on Azure.
Harvey’s solution will be similar across all law firms, so if you were a lawyer looking for a solution that speaks just your language and employs only your tactics, it might fall short. However, the vast knowledge base of Harvey’s models will come in handy for large law firms looking to make common tasks more efficient.
Deployment and integration - where domain-specific models don’t do well
Most players who take this approach to enterprise automation are not flexible with deployment and integration. They’re basically building solutions that can be used by every company in an industry, without much room for enterprise-level customization or uniqueness.
Businesses often need to adapt their processes to fit the solution, as it is not tailored to their specific workflows.
Who is domain-specific automation ideal for?
Domain-specific models are best for large enterprises with standardized processes looking to automate common tasks within their industry.
Companies that offer domain-specific automation likely won’t train their model on your business data, but general documents and data gathered from across the industry. Does this bring more accuracy than completely human workflows? Absolutely! But is it flexible enough to adapt to your existing business processes and give you a true competitive advantage? Likely not.
Enterprise-specific automation: going a level deeper than domain-specific
What is enterprise-specific automation all about?
Tired of cookie-cutter AI solutions that feel like trying to fit a square peg into a round hole? Enterprise-specific automation solutions offer a tailored alternative. Unlike domain-specific solutions, which are trained on general industry data and may not perfectly align with your unique processes, enterprise-specific automation is built with your specific needs in mind.
These solutions leverage your internal data, capturing your unique terminology, workflows, and best practices. They are designed to integrate seamlessly with your existing systems and adapt to your existing processes, eliminating the need for extensive retraining or disruptive changes. Think of it as having an AI system that speaks your company's language and understands your specific way of doing things.
While domain-specific solutions provide a broad understanding of your industry, enterprise-specific automation goes deeper, offering a level of customization that truly reflects your business's individuality. This not only improves efficiency and accuracy but also empowers your team to work smarter, not harder. In essence, these solutions offer a high level of customization and effortless usability.
How it works
Enterprise-specific automation is built upon a foundation of collaboration and customization.
The process begins with a thorough analysis of your internal structured and unstructured data. This data serves as the fuel for training the LLM, ensuring that it learns from your unique business context.
Leveraging state-of-the-art machine learning techniques, your data is used to train or fine-tune an LLM. This model is specifically designed to understand your industry's terminology, your company's workflows, and the nuances of your internal processes.
Once trained, the AI model is integrated into your existing systems and workflows. This minimizes disruptions and allows your team to start benefiting from the automation quickly. Also, the model doesn't stop learning after deployment. It continues to analyze new data, adapt to changing circumstances, and refine its performance over time.
Examples of enterprise-specific automation
Let’s take the insurance industry as an example. Underwriting is a common function across all carriers. All of them have some data extraction and analysis needs, which a domain-specific solution can address.
But the specific decision-making process and criteria in underwriting workflows vary widely between companies.
Some insurers may have a more conservative risk appetite, favoring lower-risk applicants, while others may be more willing to underwrite higher-risk individuals at a higher premium. Insurance companies also often have unique underwriting guidelines based on their target market, product offerings, and regulatory environment. These guidelines might include specific criteria for evaluating health conditions, occupation, lifestyle factors, or credit history. A generic AI model might not be able to interpret these specific guidelines accurately.
Using a domain-specific solution here can lead to several issues:
Inaccurate underwriting decisions: If the AI model is not aligned with the company's specific risk appetite and underwriting guidelines, it may make incorrect decisions, either rejecting potentially profitable applicants or accepting high-risk individuals who are likely to file claims. This can lead to financial losses for the company.
Inconsistent pricing: If the AI model is not trained on the company's specific pricing model, it may misprice policies, either overcharging customers and making the company uncompetitive, or undercharging and leading to inadequate premium revenue.
Regulatory non-compliance: In some cases, using a generic AI model that does not take into account specific regulatory requirements could result in non-compliant underwriting decisions.
The enterprise-specific advantage
An enterprise-specific automation platform, on the other hand, is trained on the company's historical underwriting data, policy guidelines, and risk models. This enables the AI to understand the company's unique approach to underwriting and make decisions that are consistent with its established practices.
By tailoring the automation solution to the specific needs and processes of the insurance carrier, enterprise-specific automation not only improves efficiency and accuracy but also ensures that the decisions made by the AI align with the company's unique goals and strategies.
At Multimodal, we worked with Direct Mortgage Corp., a residential lending company. Their workflow was mainly manual and extremely labor-intensive, for both employees and customers. With our enterprise-specific AI agents, they were able to decrease the time-to-approval by 20x, freeing up their loan officers to focus on complex cases that needed extra attention. The reason we were able to decrease the time-to-approval so much for them was that training the model on their internal data made it more efficient and suited to their use-case.
I’ve also noticed that enterprise-specific solutions often help maintain a business’s unique value proposition. If you have a unique loan origination workflow that makes you stand out from other lenders, the last thing you want is for automation to dilute the secret sauce that makes you so good. Unfortunately, a domain-specific solution would do just that, because it understands your industry, not your business.
The pros and cons
Both domain-specific and company-specific automation offer unique advantages, but they also come with their own set of challenges:
Domain-specific
Pros: Deep expertise, compliance-ready.
Cons: Limited flexibility, lower accuracy compared to enterprise-specific models, not easily integrated into existing workbenches.
Enterprise-specific
Pros: Flexible, adaptable, scalable, compliant. Also retains unique differentiators and processes and delivers better results than domain-specific solutions.
Cons: Might be more expensive than domain-specific solutions
The Bottom Line
The effectiveness of both domain-specific and enterprise-specific large language models (LLMs) hinges on the quality and relevance of their training data. However, the nature of this data and its impact on the model differ significantly between the two approaches.
Domain-specific automation
Domain-specific LLMs are trained on publicly available industry data, which may lack the granularity and specificity needed to address the unique nuances of your business processes. This can lead to several issues:
Limited accuracy: If your industry lacks high-quality, structured data, or if your business processes deviate significantly from industry norms, a domain-specific LLM may struggle to deliver accurate results. This is particularly true for tasks that require a deep understanding of your internal workflows, terminology, or decision-making criteria.
Adapting to the solution: Domain-specific LLMs are designed for broad applicability within an industry, which means your business may need to adapt its processes to fit the model rather than the other way around. This can be disruptive, time-consuming, and may not always be feasible or desirable.
Delayed results: The process of adapting your processes to a domain-specific LLM can delay the realization of benefits. Valuable time and resources may be spent on retraining employees, modifying workflows, and integrating the LLM with existing systems.
Enterprise-specific automation
Enterprise-specific LLMs, on the other hand, are trained on your company's internal data, encompassing the specific language, processes, and decision-making patterns that define your business. This tailored approach offers several advantages:
Enhanced accuracy: By learning from your company's data, an enterprise-specific LLM can achieve a high degree of accuracy in tasks that are specific to your business. This includes understanding your unique terminology, interpreting complex documents, and making decisions that align with your established practices.
Seamless integration: Enterprise-specific LLMs are designed to integrate seamlessly with your existing systems and workflows, minimizing disruptions and accelerating time-to-value. There's no need to overhaul your processes to fit the model; the model is built to fit your processes.
Faster results: With minimal adaptation required, enterprise-specific LLMs can deliver results faster, allowing you to quickly realize the benefits of automation. This can translate to improved efficiency, reduced costs, and enhanced decision-making across your organization.
So what businesses actually need an enterprise-specific solution?
While domain-specific automation can be a valuable tool for many businesses, certain scenarios demand a higher degree of customization. Enterprise-specific automation is particularly well-suited for companies with:
1. Unique workflows
Example: A pharmaceutical company with proprietary drug development processes that differ significantly from industry standards. A generic AI model trained on public data wouldn't be able to understand or optimize these unique workflows effectively.
2. Specialized terminology
Example: A legal firm specializing in a niche area of law, with internal terminology and case precedents not commonly found in broader legal datasets. An enterprise-specific solution can be trained on the firm's internal documents to accurately interpret and utilize this specialized language.
3. Highly regulated industries
Example: A financial institution dealing with complex regulatory compliance requirements that vary across jurisdictions. An enterprise-specific solution can be trained on the institution's specific compliance rules and procedures to ensure accurate and compliant decision-making.
4. Sensitive or confidential data
Example: A healthcare provider handling sensitive patient data that cannot be shared with third-party AI providers. An enterprise-specific solution can be developed and deployed on-premises, ensuring data privacy and security.
5. Decision-making processes that rely on internal expertise
Example: An investment firm with a unique investment strategy based on proprietary algorithms and market insights. An enterprise-specific solution can be trained on the firm's historical investment data and decision-making rationale to automate investment decisions in line with the firm's strategy.
By tailoring the AI model to the specific data, terminology, and workflows of these businesses, enterprise-specific automation ensures greater accuracy, efficiency, and compliance. It empowers companies to automate complex processes that were previously thought to be beyond the reach of AI, unlocking new levels of productivity and innovation.
Building an automation strategy for your business
Before you think about implementing AI across your business, you need a deep understanding of your business operations.
I know you’re thinking, “Ankur, it’s my business. I know how it operates”. But trust me when I say that you need to look deeper and dissect every aspect of how you function.
Process mining and task mining
Think of process mining and task mining as an X-ray for your business processes. These technologies give you a crystal-clear view of how work actually gets done, revealing bottlenecks, inefficiencies, and opportunities for improvement. It's like having a behind-the-scenes look at your operations, allowing you to identify where automation can make the biggest impact.
Process mining analyzes digital footprints left in your IT systems, such as timestamps, user IDs, and activity codes, to create a visual representation of your end-to-end processes. This helps you understand how work flows through your organization, identify bottlenecks, and pinpoint areas where automation can streamline operations.
Task mining, on the other hand, focuses on the individual actions that employees take to complete their work. By tracking clicks, keystrokes, and application usage, task mining reveals how employees actually spend their time and highlights repetitive, manual tasks that are ripe for automation.
Together, process mining and task mining provide a comprehensive view of your operations, allowing you to identify exactly where and how AI can be applied to maximize efficiency, reduce costs, and improve the employee experience. This data-driven approach ensures that automation initiatives are targeted, effective, and aligned with your business goals.
Once you’ve identified these areas, I want you to ask some questions:
Do I want to automate multiple workflows or multiple parts of my workflow?
Some enterprise-specific AI automation startups excel at automating parts of your workflow or just one workflow. For example, Sixfold AI is an expert at automating underwriting after an insurer’s specific underwriting guidelines. If you’re looking for a solution for just one or two parts of your business, you are better off going for a point-solution provider like Sixfold.
Similarly, if you’re just looking to automate data extraction for populating loan applications, there will be an intelligent document processing platform that specializes in this.
End-to-end automation is a different game altogether. It’s much more scalable and flexible. If you’re an insurance carrier looking to start with underwriting automation but also eventually venturing to claims processing or submissions intake, you need an AI partner that can offer multiple capabilities in the form of AI agents. These will learn from each unique workflow of your enterprise and integrate seamlessly into your systems.
(Btw, Multimodal specializes in automating end-to-end enterprise workflows. Check us out if you’re interested.)
Where do my humans shine, and what areas do I want them to continue to concentrate on?
This is a more evolved question that will help you augment your human employees’ abilities. When thinking about automation, you should also think about where you want your humans to focus.
For example, if you’re a lending company, think about which loan workflows your loan officers really need to devote their time to. The rest is where AI automation can come in and lessen their burden.
Obviously, any solution you choose should be compliant, secure, and easy to use. Compliance nowadays is so urgent that it’s no longer a differentiator for AI startups. If they’re not compliance-ready, they’re not even an option. The same goes for data privacy and handling.
I remember a conversation I had with the head of the neurology department at a hospital in NYC. They were experimenting with a new medical record automation software. While the hospital admin thought it would ease their physicians’ burden, it made them super frustrated because it was complicated and unfamiliar.
Remember, your people are going to be the ones using these autonomous systems, and you don’t want to spend weeks training them to use something super fancy. Technology should always adapt to humans, not the other way around.
Multimodal has been working on augmenting human intelligence and capabilities with AI. When I work with clients, I advise them to start thinking about their humans and their strengths, because technology will find its place in the gaps.
I also host an AI podcast and content series called “Pioneers.” This series takes you on an enthralling journey into the minds of AI visionaries, founders, and CEOs who are at the forefront of innovation through AI in their organizations.
To learn more, please visit Pioneers on Beehiiv.
Let’s talk about ethical AI
As AI becomes increasingly integrated into our lives, the importance of ethical AI development and usage cannot be overstated. This is especially crucial in industries like healthcare and finance, where sensitive data is at stake. My recent conversation with Sharifah Amirah also uncovered some insights for me. She particularly mentioned the use of black-box models and how they make anything AI-related super questionable.
Transparency, fairness, and accountability must be built into the very fabric of AI systems. We must ensure that AI algorithms are free from bias, that their decisions are explainable, and that they are used in ways that benefit society as a whole.
Make intelligent automation efforts
As AI models continue to evolve, we can expect them to become even more sophisticated, capable of handling increasingly complex tasks and generating even more accurate predictions. End-to-end automation will become the norm, streamlining workflows, reducing costs, and improving customer experiences across industries.
If you’re just starting to think about automation for your enterprise, you’ll need to reconsider everything from building and deployment to integration and evolution. I always recommend thinking years ahead with this, because you don’t want to invest in one thing today and overhaul it in a couple of years.
In the coming weeks, we’ll explore this dynamic more.
Best,
Ankur.