Hyper-personalisation, dynamic pricing and beyond — the business value of AI agents

Gartner defines AI agents as “autonomous or semiautonomous software entities that use AI techniques to perceive, make decisions, take actions and achieve goals in their digital or physical environments.”* At their core, an AI agent is a software program that can interact with its environment, collect data, and use the data to perform self-determined tasks and meet predetermined goals. Humans set up the agent and the goals, but once that’s done, the agent can work independently. 

Agents can vary widely in complexity, from simple rule-based systems that respond to specific conditions, to advanced systems that utilise machine learning and natural language processing for more sophisticated tasks.

Their ability to operate independently, learn from experience and adapt to new information makes agents invaluable for tackling complex tasks, optimising processes and driving innovation. As technology advances, the importance of AI agents continues to grow, promising transformative impacts across various fields. 

There are several different categories into which agents can be classified, depending on their specific abilities and functions. Autonomy (fully or semi-autonomous), task orientation (task completion or social) and learning methodology (reinforcement learning via trial and error, or learning via pre-planning.) 

The classes into which agents can be placed refer to their degree of intelligence and capability. These are:

1. Simple reflex agents: Operate based on a set of predefined rules responding to specific conditions. They do not have memory of past actions or states.

2. Model-based agents: Maintain an internal model of the world, which allows them to handle more complex situations by considering the effects of their actions.

3. Goal-based agents: Make decisions based on achieving certain goals. They use planning to determine the best actions to take to reach their goals.

4. Utility-based agents: Consider not only goals but also the associated costs and benefits of actions, aiming to maximise their overall utility.

5. Learning agents: Improve their performance over time by learning from their environment and experiences. They typically include components for learning from success and failure.

Agents follow a process that starts with using sensory input to gather information, and moves through analysis, and output based on programming. In some cases, agents may complete this cycle multiple times to perform an iterative process, and if capable of doing so, they may share their output with other agents, and enlist them to complete a new task.

When multiple independent agents interact with one another, this is called a multiagent system (MAS)*. Multiple agents can tackle complex tasks that individual agents can’t, such as inpatient care coordination in healthcare, and work toward a common goal. A MAS may be created by human engineers, or in some cases, by another MAS seeking to solve a problem. 

It’s important to note that today’s AI agents are most useful for augmenting, not replacing human staff. Though some can learn from data, make decisions and carry out actions autonomously without further human intervention once prompted, it is necessary to maintain human supervision to ensure that they are functioning properly and delivering the desired results.

Hyper-personalisation can drive sales with enhanced customer identification and acquisition and power exceptional customer experience with data-driven communication. Dynamic pricing enables companies to maximise revenue and profitability by evaluating and adjusting prices in real-time to take advantage of fluctuations in the market.

AI agents can analyse vast amounts of customer data including past interactions, purchase history, browsing behaviour and demographic information. Using advanced data analytics and machine learning, the agent can automatically segment users into groups for marketing purposes, and provide insights to tailor products and experiences to individual user preferences. 

Effective and responsive hyper-personalisation can reduce customer acquisition costs by as much as 50%, lift revenues by 5% to 15%, and increase marketing ROI by 10% to 30%. It also makes customers feel seen, heard and appreciated, which can drive enhanced customer satisfaction and customer loyalty.

Another way AI agents are helping to deliver business value is through enabling real-time dynamic pricing. Using many of the same capabilities that enable hyper-personalisation, the agent can analyse supply and demand fluctuations, competitor pricing, customer purchasing patterns, historical sales data and market trends to determine an optimal price for any given user at any specific time or place. 

This strategic process of targeted price adjustment creates the ideal circumstances to maximise revenue and profit. Common examples include airline tickets that cost more during peak travel times, and Amazon’s practice of adjusting prices based on customers’ buying behaviour, competitor prices, profit margins and inventory. Future developments could include businesses using cameras to change prices at the individual shopper level. This is something which is being proposed by retailers and that raises serious discrimination and privacy concerns. 

Since agents are adaptable by design, they can be applied in a variety of use cases including:

• Selling or buying products and creating solutions to support customers in consumer environments.

• Optimising, automating and/or executing processes in an industrial environment.

• Analysing, augmenting, collating, assessing and summarising information. 

• Generating or optimising creative content in a variety of formats.

• Communicating, or facilitating communications in social environments.

• Creating and optimising logistics for transportation and supply chains.

Currently, agents are being applied in finance, healthcare, education, gaming and robotics. McKinsey & Company notes that agent “capabilities have raced ahead of leaders’ capacity to imagine how these agents could be used to transform work”. Undoubtedly, additional uses will continue to be discovered. Due to their innate flexibility and scalability, especially when considering multiagent systems, their promise is almost limitless.

The vast and diverse possibilities opened by AI agents can cause either decision paralysis, or a leap to embrace them without a plan. While it is advisable to explore them as soon as possible, organisations are best poised for success when adding AI agents to their system based on a clear business case and plan for implementation. Engaging an experienced digital partner can provide advice and guidance, as well as speed the development and deployment of an AI agent system that brings value. 

As with all applications that require access to vast amounts of data and use sophisticated algorithms to generate output, there are potential risks to be managed. Insecure data systems are vulnerable to breach, and without strong guardrails, agents can produce incorrect or biased content. Their ability to function with minimal human oversight once they are set up, means that people may have less visibility into their behaviour. In those cases, an agent could produce faulty output, or malfunction, and if it’s not monitored regularly, errors could go unnoticed.

Improper or incomplete design, data training with unverified or biased data, and lack of oversight can cause agents and multiagent systems to produce improper, unintended or unpredictable results. 

Some practices that can minimise the occurrence of undesired results are when users:

• Establish clear and transparent policies for what data can be collected, how it’s used and who/what can access it.

• Ensure that anyone whose personal data is used has explicitly given permission for their data to be a part of an AI application.

• Verify compliance with all organisation policies covering data and AI use, and any relevant government regulations in the areas where the data is collected and used. 

• Confirm that developers include instructions and guardrails for how the agent system should respond under different circumstances, including if it encounters missing or unverifiable information, or receives an improper prompt or response.

Human collaboration is of paramount importance during planning and development, such as in engaging stakeholders from different business areas to clearly define an agent’s objectives. Before launching an agent, designing plans for ongoing monitoring can ensure that relevant employees will regularly be checking in to make sure the agent(s) are working as designed.

Organisations that bring significant experience with AI development and data management can help ambitious enterprises ensure that their agents have seamless UX design, handle data properly, and that they’re compatibility with existing systems. For these reasons, most organisations exploring integrating artificial intelligence applications collaborate with a digital partner that has domain expertise as well as a track record of success in similar engagements.

Nearform’s AI and data experts are always looking for innovative ways to solve challenges. Sometimes solutions suggested to address partner issues originate from internal experiments. 

Nearform Field CTO Shaun Anderson embarked on an experiment with an agent framework to help streamline his own workflow. He described the process, how it helped him in his work, and how this process can be applied with clients.

“I built an agent framework to explore what an agent can do. I asked it to tell me which software component makes sense to modernise, and then create a plan to do that. The agent knows which tools I’ve got available, and it asks ‘What’s the nature of your data?’ In this case, the data is the software schema graph. With that information, it knows what tool to use to execute queries and analyse the graph. Now it can narrow down the algorithms that might be good candidates for modernisation, then engage other agents to suggest a plan. 

By the time it gets to that output, it’s used a few different tools, and along the way it picked what tools to use on its own, executed queries and reported the response. Now, if at any point it doesn’t have enough information, it can prompt me for it. It leads me down the path toward the solution like a trusted advisor. 

When you have the right data, the agent can select the appropriate tools to use, iterate to gather solutions, and, according to its code, either deliver a recommendation to a human user or engage more tools to continue the workflow. 

We can apply this kind of workflow to many different situations, each one with specific instructions based on the type of data and the desired result. The agent system follows its process, and depending on the type, it can learn and suggest adding additional agents to streamline the rest of the process and get to the desired result faster.” 

From the very start of determining what the potential business cases for an AI agent might be and how it might provide value, an experienced partner like Nearform can provide guidance. Oftentimes AI projects don’t live up to expectations, because of incomplete or incorrect assumptions about the possibilities made at the outset, perhaps by not engaging a cross-functional set of stakeholders.

Engaging Nearform’s experts with an exploratory “accelerator” can not only maximise the chances of developing a valuable AI solution, it can also speed up the process from concept to modelling, to development and deployment. Contact Nearform today to discuss how we can help your enterprise.

*Innovation Insight: AI Agents, 3 April 2024 - ID G00806843, GARTNER is a registered trademark and service mark of Gartner, Inc. and/or its affiliates in the U.S. and internationally and is used herein with permission. All rights reserved.