We were recently asked a probing question by a CIO - “What can AI automate for me that I couldn’t achieve with other simpler automation techniques?”
We think this is a good question to keep ourselves honest with. A lot of the example use cases shared by AI evangelists would be better served with a simple piece of static code or a little data science. Anyone who uses these use cases to sell AI or LLMs is being at best disingenuous. That said, we believe there is a delta, things which a business can automate using AI and its cousin Machine Learning which they can't do today. This delta is encapsulated by a phrase we created - "Intelligent Automation".
Naive Automation
Businesses have been able to implement basic or “naive” automation using software for a long time. Whenever a customer or an employee enters data into a system, some process can be triggered which involves running code, calling APIs, or moving data around.
For example, when a product is returned to a warehouse, a refund could automatically be triggered by the finance system 48 hours later. Or, when a new employee joins a business and is entered into the HR system, they could automatically be issued a Salesforce account. This meets the definition of automation because there is no human manually logging into online banking to issue payments or manually provisioning accounts.
However, this type of automation lacks intelligence. In reality, a software developer has simply taken a business rule and implemented it in a deterministic way, perhaps calling APIs across system boundaries. At best, there might be some hard-coded if/else logic buried within the automation like this:
if(employee.department().equals(“sales”) {
create_salesforce_account();
}
This is automation that incorporates business rules, but it's not in any way intelligent.
Moving To Intelligent Automation
Intelligence is defined as "the ability to solve complex problems or make decisions with outcomes benefiting the actor."
Intelligent Automation is about giving our software systems the ability to do this, elevating our automation beyond simple deterministic business rules to being able to handle complex situations and consistently making the optimal decision with less human input.
Today, computers are very good at following deterministic business rules based automation, but humans are much better at making complex, subtle and nuanced decisions. Experienced employees will know how to handle the tricky cases which don't quite fit into standard practice. They will know when to bend the process, which cases to prioritise and where they need to step in and take manual action. This skill and awareness is why human employees still run the show despite our efforts to digitise.
Intelligent Automation is an attempt to replicate these very human skills into our automated processing, making our systems behave more in line with an intelligent and motivated employee. The opportunity is that by introducing this intelligence into our processing, we can automate significantly more work. This means that more tasks can flow straight through the business without human intervention, making everything faster, more efficient and cheaper, whilst also delivering a better customer experience. Even better, a system which has been properly Intelligently Automated will take preemptive action based on the framework you have given it.
An Example
A business has a process in place to review refund requests and decide whether or not to issue refunds based on the circumstances of the request.
Today, this process is entirely manual, because they feel they need a human to evaluate the request. The business policy involves looking at the order value, the customer lifetime value and the number of refund requests made by the customer in the past. If the employee agrees a refund should be made, they will log into a finance system to trigger the payment manually.
My suspicion based on waiting a long time for refunds is that most businesses rely on this manual process today!
The company grows and finds themselves needing to issue more refunds. They decide to progress to simple business rule based automation to process low value refunds without human involvement. Perhaps they come up with an algorithm like this that is implemented in their system:
if( order.value < 100
&& customer.spend > 1000
&& customer.number_previous_refunds < 3
) {
issue_refund();
}
This type of deterministic business rule led automation works well where we have clear well defined logic, few exceptions and relatively simple inputs. It also assumes that things are static over time.
But what about the higher value orders? Because they are riskier, the business retains a team of employees to manually review each request.
For each one, an employee will pull up the order details and the customers complete transaction history, and make a decision whether or not to refund based on legal obligations (e.g. Consumer Rights Act), internal policies, and commercial decisions such as the customers likelihood of churn or predicted lifetime value and the ability to resell the product. It is a much more complex and nuanced decision which relies on experience and intuition as much as following the rules. Furthermore, each decision could take approximately 10 minutes of analysis meaning that it isn't a particularly scalable activity.
Intelligent Automation is about bringing this complex decision making process into our systems so that we can more closely replicate the experienced and intelligent employee and ultimately automate more of our business operations.
Fortunately, our fictional business have multiple tools in their toolbox for introducing this level of automation into their software systems:
They begin by incorporating some relatively simple predictive analytics to understand the probability of the customer churning. If there is more than a 50% chance of the customer churning after declining the refund request then they choose to issue the refund without any human involvement;
They recognise that there are so many dimensions to the decision that it makes more sense to move to a machine learning model, where the optimal decision is predicted by looking at the outcomes of previous decisions. This machine learning model is inserted into systems and used to drive the decision inference around the refund instead of business rules;
They decide to use a Large Language Model to bring more unstructured data into the decision making process. We process their communications from the CRM system and find that the customer has recently emailed in a complaint about the order, a fact which we bring into the overall decision making process;
They use a different Large Language Model to run an automated check of the order against company policy documentation to ensure that nothing about the refund is being overlooked. As an example, vulnerable customers should be given extra leeway in any refund decisions. This is the type of edge case which an employee could easily overlook;
If appropriate, they make use of a Large Language Model to draft an apology note to the customer and offer a 10% discount code in an attempt to reduce churn;
The AI ultimately completes the reasoning whether or not to issue the payment, and sends it for a final review by a human employee prior to the payment being made. The human workflow task includes a concise summary of the situation of the recommendation drafted using the LLM.
This is a perfectly acheivable example of Intelligent Automation which closely replicates the human employee, looking at all of the context before making the optimal decision in a complex and dynamic situation.
How AI Supports Intelligent Automation
Recent developments in AI including Large Language Models and Autonomous Agents are key tools in our toolbox for achieving this dream of Intelligent Automation. They bring the following new capabilities to the table that go even beyond machine learning:
Automated Knowledge Acquisition - By using an LLM to ingest unstructured text based data, we can use inputs such as forms, emails and chat transcripts as part of our decision making. The LLM can give us the ability to bring much more knowledge and facts into our decisioning;
Automated Reasoning - The LLM can be used to actually carry out the complex reasoning process outlined above. With careful prompt engineering, they can weigh up and reason through all of the characteristics of the problem in a way that is very closely aligned to the employee making a decision;
Automated Actions - An Autonomous Agent can be developed which can essentially automate the same process as the knowledgeable employee would take. The agent could pull in the order details and the customer history, reason about the correct response and then actually carry out the action for us in order to automate the end to end task. These agents can then be set up to collaborate as part of an end to end business workflow.
Using these tools, we can automate entire workflows from knowledge acquisition to reasoning to eventual action, giving us the maximum possible chance of achieving straight through processing - all with Automated Intelligence at the heart.
This Is Transformative For Business
If we can bring more AI powered Intelligent Automation into our business and increase the level of straight through processing, it is possible for businesses to save huge sums of money spent on manual back office activities.
Most businesses are filled with business processes which are carried out by humans. Some of these processes could move from no automation to being entirely automated, whilst others might move from say 70% automated to 80% automated. The business case for doing this across a single large enterprise could be overwhelming even with small incremental uplifts.
This doesn't necessarily mean that these employees would be displaced, the same budget could instead be spent on doing valuable differentiated work or scale the business without the need for adding oodles of additional headcount. This drives real, tangible value creation whilst completely changing the nature of their cost base by eradicating the toil.
Equally, this is absolutely transformative for the customer experience. Sticking with the same example, instead of waiting for 2 weeks for your refund, it could be evaluated and approved in 30 seconds by the AI.
Intelligent Automation has enormous potential, and we believe it is where the value of AI will eventually find a home within business.