Jul 23, 2025
The Problem With Traditional Automation Logic
Traditional automations have been based on deterministic flows according to which predetermined actions are executed if certain conditions are met. This type of automation is currently referred to as deterministic, as it does not allow variations beyond those that have already been configured.
It is known precisely how it will behave, and it is predictable, since the automation flow has concrete actions for each case.
However, this automation system, although efficient and very powerful, has a significant limitation: rigidity.
On the one hand, as the actions are very well defined, the moment an unforeseen factor enters, the automation issues an error and, in most cases, stops. This can lead to production stoppages and efficiency errors. For error correction, manual intervention by automation professionals is necessary. All this can increase complexity in cases where many scenarios and complex automations are contemplated.
Below is an infographic on how these types of automations work:
Currently, there are various automation tools, with Zapier, Make, and n8n being the most widely used.
Despite being the most popular ones, they also have their limitations today.
Regarding the use of LLMs, Zapier and Make have minimal use, focusing on straightforward tasks and semi-deterministic automations. Since LLMs are not natively integrated, they cannot use AI conversationally or build complex AI agents.
In the case of n8n, one of the most popular options, it is the most advanced in terms of AI usage. However, it requires manual configuration or even code entry to set up AI agents. Additionally, it is visually complex to observe the behaviour of AI agents and LLMs.
Another critical point is regulatory compliance. The main regulations SOC2, HIPAA, GDPR, Zapier and Make only ensure basic security.
The On-Premise option, the only tool that allows it is n8n, but it does not include, by default, compliance with the regulations required at enterprise level.
As for the visualisation of automations and API endpoints, only Zapier, in a minimal way, makes it difficult to see how the information flows throughout the automation.
Introducing AI Routing: Smart Decision-Making at Runtime
The AI Routing feature enables automations and workflows to reason and contextually make decisions in real-time. This allows you to maximise the benefits of LLMs and automate workflows without requiring programming knowledge.
How AI Routing works in Stack AI
Dynamic conditional flows: With Stack AI, you can create automations that automatically apply advanced logic and even natural language processing based on the data passing through the workflows.
Example: In the case of a finance company that may receive many loan applications every day and whose documents require credit verification.
It could be set up to automatically verify and generate each action for each type of credit assessment of the personal profile:
Classified documents: When a loan application is received, an AI agent first verifies whether sufficient documentation has been submitted. It then assesses what type of documentation is available and classifies it based on this information. All this is done thanks to logic based on LLMs.
Smart routing: Once all documents are uploaded, automation takes effect, leveraging credit assessment to execute the most appropriate action.
Doubtful assessments: If credit documents with doubtful scores are identified, LLMs, thanks to their real-time capabilities, can execute an action to trigger a manual review.
This is how it would look in Traditional Automation vs Stack AI:
Why AI Routing Changes the Game
With Stack AI, in automations, the “If This Then That” conditionals. This makes it very easy for companies to scale both their internal processes and even customer service.
Without the need for “If/Then/Else,” you can take advantage of and optimize the use of LLMs. Using “If/Then/Else” in traditional automations makes sense in particular cases and is very predictable. For companies, the more processes, factors, and variables there are, the less viable this option becomes and the more prone it is to errors.
The potential of AI Routing
Using Stack AI, the actual context is taken into account, and the intention behind each input can be accurately interpreted. With routing, the most appropriate action is taken without having to determine, as in traditional automation, which process to activate with each variable.
Regarding the context of Artificial Intelligence, LLMs, and the use of Stack AI, the platform is capable of:
Understand the meaning of the input, rather than focusing solely on keywords.
Direct, prioritise, and even classify according to criteria, including the user's language.
Without human intervention, adapt responses and take action in new situations.
For example, to assess its potential, you can create an agent in Stack AI that can understand the tone of messages or emails, such as complaints, common queries, and urgent cases.
Modifying workflows in Stack AI is really easy. The primary benefit is the ease of modifying any workflow without having to start over or disrupt the automation. This is thanks to the modular nature of the flows and their ease of updating. These modifications can be made by adding, adjusting, and reordering at all points in the workflows.
It is also possible to maintain excellent and easy traceability of how the automations are working. Additionally, changes can be audited and scaled (a crucial feature for companies) with the “add,” “swipe,” and “recipe” options. All this is presented in a very visual interface that allows any manager to insert any adjustment easily.
Use Cases That Only AI Routing Can Handle
With Stack AI and its use of LLMs, they can be helpful when analysing the history, description, and even nuances of tickets. This allows you to:
Depending on the type of request, identify whether the user is sending an ambiguous message or using unexpected phrases.
Recognise intentions or specific topics in a single message. For example, there is an error on my invoice, and I am unable to access the portal.
On the other hand, it can also determine the urgency depending on the nature of the customer. For example, it can recognise a complaint from a VIP customer who needs an immediate response.
How does AI Routing with Stack AI solve Lead Qualification?
LLMs in Stack AI are capable of analysing any message from Leads (thanks to natural language), with automatic identification of valuable data. For example: position, sector, company size, objections, or level of interest. All this despite the information arriving in a scattered or ambiguous manner.
The flow of information may proceed to an interactive process, as it may be necessary to ask follow-up questions or extract information through the interaction of different communications.
If an ICP (Ideal Client Profile) is defined, AI agents have the potential to evaluate and qualify clients in real-time according to the specified criteria.
Based on the qualified lead, corresponding actions are automatically performed in accordance with the commercial process requirements.
Let’s start with an example Workflow of Escalation IT:
The input is a message that we receive: CRITICAL: Web server cluster offline - 503 errors, revenue impact $50K/hour
Then, the LLMs of OpenAI have the Instruction to analyse the incident with the context, including description and severity. The Prompt requires the classification of the incident by:
Category
Business Impact
Priority (which helps to escalate to the right team)
In order to understand the best team depending on the category, business impact and the different teams, we explain which are the teams available for each situation.
Once we get the answer, the AI Routing is able to decide which team is the best with every incident.
In that Example Case, the incident is Critical. That means the AI Routing goes to the option for Critical incidents. The desired output is: consider the business impact, mention, if necessary, additional specialists, assign an incident commander, create a bridge call, compile all the context, and set a communication cadence.
After gathering all the information, it is now necessary to have effective executive communication in case of a critical incident. Now it’s needed to have three communications:
Executive brief for the CEO or CTO.
Customer status page update.
Partner/vendor notification.
Finally, this is the output:
Executive brief for the CEO or CTO.
Customer status page update.
Partner/vendor notification.
In summary, this is an example of how, from an input, the Stack AI can escalate in an Enterprise IT.
This is the complete workflow with all the options:
When to Use AI Routing in Your Workflows
Using AI Routing is highly recommended in situations where free text and data require natural language processing.
The main uses of AI Routing can be:
Free text from variables from forms, messages, chats, or emails.
When a better understanding of both the meaning and intent of the message is needed.
Interpreting ambiguity to extract data, understand feelings, and even the history of previous communication with the same person.
When scaling your automations, it is not feasible to use conditionals such as “If/Then/Else.”
Always consider Stack AI's AI Routing when you need intelligent and dynamic responses, without static limitations and operating optimally and efficiently.
Use “If/Else” in cases such as structured inputs, for very clear and expected situations.
AI Routing is ideal for scaling without needing to determine multiple actions with numerous variables that are impossible to manage.
Analyses are performed within the context based on an input, allowing you to understand nuances, complexity, urgency, intention, and the type of task.
For companies that receive a high number of inputs, AI Routing is ideal for scaling. The logic is simple to implement.
When new needs arise, simply modify the AI Agent to continue the automation.
Paul Omenaca
Customer Success at Stack AI
