The 7 levels of AI-powered CX automation
How to move beyond basic AI and develop the maturity needed to transform your contact centre operation.
AI has been transforming customer experience (CX) for years, yet many businesses struggle to scale beyond basic automation. They often apply AI at the interface level, like chatbots answering FAQs, without integrating it into core processes. This results in limited benefits, inefficiencies and stalled progress.
So how can you leverage AI in the best way, to add undeniable value to your customers and your bottom line? How can you prevent your AI initiatives from getting stuck in pilot? And evolve from simple AI implementations to fully automated workflows and business transformation?
To help you navigate these challenges and scale your AI automation efforts, I’ve put together the 7 Levels of AI-powered CX Automation. This will provide you with a way to assess your AI maturity and strategically advance toward proper value. By following this model, you can move beyond surface-level AI applications and unlock real efficiency gains, cost savings, and superior customer experiences.
TL:DR: The 7 Levels of CX AI Automation in a nutshell
Interaction optimisation: AI enhances the interface layer, but doesn’t solve root inefficiencies (the ‘lipstick on a pig problem’).
Automated qualification: AI helps filter requests to prevent unnecessary processing.
AI-mediated triage: AI understands customer needs and routes to existing self-service solutions.
Verification, information collection & agent handover: AI verifies users, collects information, routes to the right agent, streamlining agent interactions.
End-to-end automation: AI fully resolves customer requests autonomously.
Problem elimination: AI predicts and prevents customer issues before they arise by either eradicating the problem before it arises or self-healing the issue before it affects the customer.
Process transformation: AI reshapes entire business functions and workflows.
The problem with your AI program to date
Many businesses deploy AI for interaction optimisation (Level 1), layering AI on top of outdated systems (the ‘lipstick on a pig’ problem) and stop there. They implement chatbots to deflect calls or automate simple FAQs, but fail to integrate AI into backend processes. This results in:
Minimal impact on key business metrics (e.g., CSAT, first call resolution, average handle time)
Increased agent workloads due to ticket creation without resolution
Frustrated customers who still need to wait due to manual processing
Stagnated AI programs that struggle to gain maturity and hit a ceiling
A sense across the organisation that AI is limited and doesn’t live up to its promise
It’s not like it’s easy. Silos between AI, IT, and operations teams prevent holistic automation in most cases. Lack of integration, misaligned priorities or outdated technology infrastructure all contribute.
So here’s a step-by-step process for progressing through the 7 levels, what each stage entails and how to bring your organisation with you on your journey.
Level 1: Interaction Optimisation (IO)
At Level 1, AI is applied to improve customer interactions without addressing underlying inefficiencies at the journey or process level. This is what I call the ‘lipstick on a pig’ problem, where AI is applied at the interface layer, over the top of outdated systems. This creates the illusion of progress, but doesn’t achieve anywhere near what it could.
An example of Interaction Optimisation (IO)
An example of this would be a chatbot that recognises a user’s intent, then logs a tickets into a CRM for human agents to process later. Or worse, sends the request as an email for it to be processed according to the existing manual workflow.
While this might feel like a step forwards, it doesn’t solve the root problem: customers still have to wait for resolution and everything surrounding the process has stayed the same, barring the interface used to collect information.
That means you’re not helping the customer because value is being slowed down, introducing the potential for repeat contact in the meantime. And you’re not helping your agents because they still have the same work to do. It can, in fact, lead to more work for agents because the chat channel can often uncover previously untapped demand, which can result in more tickets to process.
Another example would be an AI chatbot that answers surface-level questions for users using Retrieval Augmented Generation (RAG).
Benefits of Interaction Optimisation (IO)
The benefits of Interaction Optimisation include:
Adds a new customer interaction channel: AI chatbots can serve as an alternative to calls, emails, or forms.
Reduces initial friction: Customers can quickly log requests without needing to call support.
Provides a low-cost automation option: No need for complex integrations or backend modifications.
Can enhance the experience: If designed well, this level of chatbot can reduce effort for customers by handling simple queries.
Acts as a stepping stone to deeper automation: While limited, Level 1 automation can serve as an entry point for AI adoption.
There’s nothing wrong with Interaction Optimisation. Sometimes, opening up another channel or providing another interface can be a good idea, provided that interface can manage the customer interaction more efficiently, with less expense and provide a better experience.
This example is the AI equivalent of putting an online form on your website that logs tickets or sends emails behind the scenes. It’s a small step forward for the user, and a smaller step forward for the business, but it’s not going to improve CSAT, reduce average handle time or effect any of your metrics that matter. ‘Deflection’, maybe, but what use is that metric if the customer issue isn’t resolved?
Challenges of Interaction Optimisation (IO)
That being said, there are some challenges with this approach that will urge you to progress to Level 2 sooner rather than later. For example, Interaction Optimisation:
Doesn’t solve the root problem: Customers still have to wait for resolution as backend processes remain manual.
Potential increase in agent workload: AI chatbots may generate more tickets rather than reducing work.
Limited impact on key business metrics: No significant improvements in CSAT, Average Handle Time (AHT), or First Call Resolution (FCR).
The problem many enterprises have is that they implement this Level 1 automation and stop there. Sometimes, stopping here is due to technical limitations and not having the systems in place to go any further. It can be hard to make full end-to-end automation a priority if you’re sitting in a conversational AI team or an operations team, and those responsible for the process and underlying systems sit in another department, managed by another person and dealing with a different set of their own priorities.
However, that’s not to say that progress can’t be made. You must strive to get yourself into a place where you can move to Level 2.
Level 2: Automated Qualification (AQ)
This stage is all about filtering users out of a journey to prevent unnecessary processing and wasting the user’s time. Sometimes, you need to verify that a user meets certain conditions before allowing them to access a specific service or process. Automated Qualification is where you use AI with business logic and often some backend data retrieval to determine whether a user meets the right criteria to continue.
An example of Automated Qualification (AQ)
An example of this would be a retail chatbot that pre-qualifies return eligibility. For example, let’s say a customer made a purchase 6 weeks ago, and your policy is to allow returns for up to 30 days after purchase. The AI chatbot in this instance would take the user’s order number, return the order details, check whether it’s eligible for a return, and communicate this to the user.
Benefits of Automated Qualification (AQ)
The benefit of Automated Qualification is that businesses can filter incoming contact for relevance and quality, to make sure that only those that should be processed are. This means that your people can stop wasting their time and customers are told up-front whether they qualify or not, so they don’t have to wait.
Benefits include:
Reduces unnecessary contact: Filters out ineligible cases before they reach human agents.
Saves time for both customers and staff: Customers get instant answers, and agents handle fewer, more relevant inquiries.
Enhances customer experience: Users get a fast go or no-go response without waiting in a queue.
Lowers support costs: Fewer escalations mean reduced contact centre workload and lower operational expenses.
Challenges of Automated Qualification (AQ)
This doesn’t always mean that you’re going to avoid having to bring your people into the conversation. Often, after hearing bad news from a chatbot, customers think that speaking with a person will enable them to circumnavigate the process. And sometimes they’re right! What’s important is to reinforce the message when your people pick up the phone so that there’s consistency between the AI and your people. They need to work together.
Customers may seek human intervention: Some users will insist on speaking to an agent, even when AI denies their request.
Agent inconsistency can undermine AI decisions: If human agents override AI rulings, customers may learn to bypass automation.
Not all systems integrate easily: Some of these use cases need real-time data from backend systems, which can be challenging for some.
Poor communication can frustrate customers: If the AI system doesn’t clearly explain why a request is denied, customers may lose trust in the system.
Level 3: AI-Mediated Triage (AMT)
At Level 3, AI is not just qualifying requests but actively routing customers to the most appropriate self-service automation channel. Instead of escalating every request to an agent or relying solely on basic automation, AI triages customer inquiries and redirects them to existing self-service solutions that are already effective.
This approach bridges the gap between AI-driven intent recognition and existing digital investments, helping businesses scale AI usage without deep system integrations.
An example of AI-Mediated Triage (AMT)
Imagine a customer calls their bank’s contact center to close their account. Instead of placing them in a queue to speak with an agent, an AI-Mediated Triage Assistant would:
Identify the intent (“close my account”) using speech recognition and NLP.
Check for existing digital solutions that can handle the request.
Send a text message or a push notification with a deep link to the bank’s mobile app, where the customer can complete the closure process via a pre-existing, fully automated workflow.
By redirecting the customer to a highly efficient, purpose-built self-service journey, the bank reduces contact center volume, improves efficiency, and enhances customer experience.
Benefits of AI-Mediated Triage (AMT)
The benefits of AI-Mediated Triage include:
Faster implementation without deep AI integration: this doesn’t require AI to execute the process itself; it simply guides users to existing automation solutions.
Maximises existing digital investments: Triaging to existing high-performant self-service options means you increase adoption and leverage of existing or prior investments.
Reduces contact centre load and operational costs: Fewer customers need human agents, lowering call volumes.
Improves customer experience: Customers are immediately directed to the fastest, easiest resolution instead of waiting in a call queue.
You might think that customers will never go for this channel-shifting approach. Whereas, in fact, according to research in The Effortless Experience, customers will typically use the channel that has the best chance of resolving their issue, over and above personal channel preference.
Challenges of AI-Mediated Triage (AMT)
As with every solution, this isn’t without its challenges, which include:
Not all customers will accept triage: Some customers won’t trust self-service or may insist on speaking to an agent regardless.
Seamlessness of transition to self-service: If the digital journey is difficult to navigate or incomplete, customers may abandon it and call back, negating the benefits.
However, when done right, you’ll see this have a significant impact on your demand channels and will then allow you to chip away at Level 4.
Level 4: Verification, Information-collection and Agent-handover (VIcAh)
Next, you'r goal becomes to start reducing your average handle time (AHT). To progress towards this, you’ll need to pay a visit to the vicar (VIcAh), which involves:
Verification (V): Identifying the customer and verifying that they are who they say they are.
Information-collection (Ic): Collecting the relevant information about the customer’s situation (intent and relevant details) to prepare the interaction for human handover.
Agent-handover (Ah): Passing the interaction to the right agent, with the right skills, and passing across the customer’s intent, verification status and all required information to enable the agent to hit the ground running.
An example of Verification, Information-collection and Agent-handover (VIcAh)
Imagine a customer calls their bank wanting to query an unrecognised charge. They’re greeted by a voicebot that asks them why they’re calling.
Once they’ve explained what they need, behind the scenes, the agent routing is prepared. The customer is then asked to verify their identity via a push notification and pin confirmation sent to their mobile app (there are many ways to varify, including using AI with voice biometrics).
Once verified, the voicebot asks for the user to explain the amount, date and description of the charge. They are then handed over to the right agent who can begin their conversation with something along the lines of “Hi Mr. Simms, I understand you’re calling about an unrecognised charge on 12th December for £45.89 to Rode, is that correct?”
Benefits of Verification, Information-collection and Agent-handover (VIcAh)
There are many benefits of this approach, including:
Reduced average handle time (AHT): The agent skips verification and data collection, focusing directly on resolving the issue, shortening the call duration.
Lower call transfer rate: Routing the call to the right team the first time minimises the need for transfers.
Improved first call resolution (FCR): The agent has all the necessary information upfront, reducing back-and-forth with the customer and any grounds for misunderstanding.
Enhanced customer experience: The interaction feels seamless and efficient, as customers don’t have to repeat themselves. You’re also bringing some value up-front to the customer immediately so that they feel like they’re getting some progress right from the off.
Reduced wait times: having the user verify, and answer a few questions, can take place instead of waiting, so the user won’t have to wait as long in a queue. If you wanted to get fancy, you could also prioritise callers based on the nature of their intent (high value intents prioritised) or their status (loyal customers prioritised) or similar.
Challenges of Verification, Information-collection and Agent-handover (VIcAh)
The challenges in implementing this approach include the verification piece itself, such as:
Verification success rate: Being able to verify every user consistently is difficult, and different company policies require different levels of verification for different processes. So it’s not like you can have one method and be done with it.
There are also challenges with the agent handover piece that I’d divide into two buckets:
Handover logic.
Determining the right agent routing logic is complex: AI must match users with the best-suited agent, considering skills, queue availability, and priority levels.
Lack of sophisticated AI-powered escalations: Most AI platforms do not natively handle intent-based routing or queue prioritisation.
Middleware tools may be required: Solutions like SentioCX can act as middleware to manage routing and escalations, but without something like that, businesses must custom-build these capabilities.
Contact centre integration.
Voice channel complexities: The AI system must pass verification status and collected information to the agent before connecting the call.
AI platform limitations: Many contact centre AI solutions lack robust, scalable agent-handover features.
Data transfer issues: If using a third-party AI or custom-built solution, contact centres must support SIP headers or similar protocols to pass information effectively.
Chat platform inconsistencies – Different live chat platforms have varying integration requirements, making it harder to ensure a seamless handover.
Once you’ve got this kind of set up in place, you have everything you need architecturally and culturally to progress to Level 5.
Level 5: End-to-End Automation (E2E)
At Level 5, the focus shifts to automating full processes and workflows. Instead of logging tickets, or qualifying eligibility, the AI solution resolves the issue directly, from end-to-end, eliminating the need for human intervention. This is where most businesses will see more of the benefit of AI because it starts to impact the inner workings of the business, rather than the outer layers of interaction.
Most will start with tier 1 queries and escalate tier 2 and 3 to human agents initially. However, as you mature, and as the technology improves and AI agents mature, it’s likely that tier 2 and perhaps one day tier 3 use cases also fall in scope of AI automation.
An example of End-to-End Automation (E2E)
An example of this would be a retail chatbot that, once user’s have been qualified to make a return, generates the return label for the user to send back with their parcel.
Another example might be requesting a payment delay to a credit and collections company. Rather than this request being granted by a person, an AI solution could do this based on the request falling inside a certain degree of tolerance. The AI solution would then update the line of business system to make the required changes on the customer account and add notes or populate the relevant fields in the CRM to log that the interaction was successfully facilitated.
The list of use cases are endless because they’re tied to the needs your customers have and the things your company actually does to facilitate the purchasing, distribution and support of all customer journey activities.
Benefits of End-to-End Automation
End-to-End Automation is where you start to see some real value. The list is endless, but includes:
Faster resolution times: Having AI assist in workflow automation means reduced delays and improved response times.
Lower operational costs: Automating full processes can reduce reliance on human agents, cutting labor and processing costs if you choose to extract them.
Improved customer experience: Conversational interfaces (the front-end of most AI applications, are typically easier to use, reduce effort and improve experience.
24/7 availability: AI doesn’t take a day off sick, and can work 24 hours a day, meaning you can transact round the clock.
Scalability: Handle higher volumes without adding more staff, making operations more efficient.
Eliminates manual errors: AI-driven processes reduce human mistakes, improving accuracy and compliance.
Consistent and standardised service: It works the same, every time (as long as you’re not using LLMs for business logic(!))..
Seamless omni-channel experience: You can integrate AI across multiple touch points (chat, email, voice, apps), allowing customers to switch channels without disruption.
Free-up human agents for complex issues: Employees can focus on high-value tasks, like handling edge cases and providing personalised support, rather than dealing with repetitive tasks.
Data-driven decision making: AI can gather and analyse real-time insights, helping businesses refine processes and optimise service delivery.
Reduces backlogs and bottlenecks: Automated workflows can prevent service request pile-ups, improving efficiency across teams.
Future-proof operations: AI-driven automation enables businesses to adapt to digital transformation trends and remain competitive.
Challenges of End-to-End Automation
Achieving this level often requires integrating siloed data and building robust APIs. Basically, you need to be able to do, programmatically, everything that your people do to fulfil customer needs. Many businesses that haven’t already transitioned their infrastructure to the cloud and got themselves into a digital or mobile first environment, get stuck here due to competing IT priorities or system limitations.
Another key challenges is related to the lipstick on a pig problem, which is that, the processes in place in many organisations today are outdated and clunky. This means that, technology aside, the steps required for customers or staff to go through in order to get an outcome are often excessive.
Typically, there are numerous inefficiencies baked into these processes, which are then carried over into automated solutions. This can lead to automating waste rather than eliminating it, which negates the potential benefits of end-to-end automation or leads to poor experiences. To truly reap the rewards, organisations must first re-engineer their processes with Lean principles in mind.
Level 6: Problem Elimination (PE)
At Level 6, you go beyond automation to eliminate the need for those processes altogether. The goal is to predict and resolve issues before they arise, creating a seamless experience for customers. This requires more than having the right technology available, it requires a cultural shift organisationally to share data across the organisation, and having teams empowered to act on this data to make process, product and service changes, based on what’s being learned.
This problem elimination concept comes in two forms:
1. Pre-empt
Pre-empting is when a potential issue is about to arise that you stop in its tracks. A basic and common example is when you arrive in a foreign country, turn your phone off from Airplane mode and you receive a text message from your carrier telling you that they’re aware that you’re in another country, as well as information on how much you’ll pay for texts, calls and how to enable internet access.
This is a real basic example from today, but AI will give you the capability to constantly analyse and pre-empt issues within your business before they arrive, not just pre-empting contact based on environmental changes.
2. Self-heal
Self-heal is where the issue has occurred, but you’re able to resolve it before the customer realises and reaches out. For example, let’s say you book a cab and, after booking, the app detects that a driver has unexpectedly cancelled. Instead of notifying the passenger and making them rebook, the system automatically assigns the nearest available driver and updates the ETA, so the passenger experiences minimal disruption and doesn’t needing to take any action.
Benefits of Problem Elimination (PE)
By using insights and data to be proactive and resolve issues before they arrive, or before the customer recognises, is to offer almost the ultimate experience. The impact on loyalty, satisfaction and churn could clearly be positive, given that these are the metrics affected from poor relationships and experiences. There’s of course a cost reduction element here because this will lead to a decline in overall contact in the support function, too.
Challenges of Problem Elimination (PE)
This requires combining data insights with proactive actions, which often demands investment in advanced analytics and real-time monitoring. This is often a big challenge for businesses to do, even in the most simplest of ways because it requires a strong data science discipline and technical expertise to implement. You then need to have the ability to take actions to resolve these issues as and when they’re detected in an automated fashion, so you have the additional requirements of process redesign, orchestration and all of the additional analytics and monitoring that comes with that.
This is where I think we’ll see some innovation from startups over the coming years, however it’s easier said than done because your analytics and automation requirements are completely unique to you and depend on the systems and processes you have. No two businesses are the same and so there isn’t a scalable way for a vendor to provide these capabilities. An orchestration platform, yes. But the solution itself? I suspect that’ll have to be done in house or through partners.
Level 7: Transformation (T)
At Level 7, AI enables you to rethink how work is done in the first place. Manual, complex tasks like drafting contracts, negotiating deals, analysing data, serving customers, distributing products and everything required to build, sustain and grow your business become transformed with AI and technology.
Every business function will leverage AI at its core and have sequenced technologies that deliver value that you can’t imagine today. This means redesigning your workforce, technologies and, dare I say it, leadership.
An example of Transformation (T)
The reality is that you now have the technology available to begin automating things that you wouldn’t have previously imagined could be automated, or getting more leverage and performance out of the things you do automate. If you’d have told me 5 years ago that our content repurposing plan could be done in 5 minutes, I’d have thought you were mad. But we’re heading in that direction with our automated podcast repurposing pipeline.
Just imagine all of these areas within your business where you have people sitting around typing, copying and pasting… All of that is open to transformation.
All of your existing automation solutions that are built on legacy technology and that still require people to push things along… All gone.
Every process within every business, every system and every function has the potential to transform because of the combination of technologies we have access to today, none mores than the contact centre.
In the very near future, Tier 1 requests will be a thing of the past. That means that, rather than spending 80% of their time on operational management and 20% of time on strategic initiatives, contact centres will be able to flip that on its head. Staff won’t be taking requests, they’ll be managing and optimising the AI workers that are taking the requests. Operationally, this will require significant changes in job roles, skills and even fundamental KPIs.
If you were to design your business from a blank canvas today, would you design it the way it is? Probably not. So how would you design it? That’s the place to start when thinking about Transformation.
Benefits of Transformation (T)
The benefits of process transformation are fundamentally based on relevance, longevity, scalability and profitability. Businesses need to constantly reinvent themselves to survive and thrive. How you do things today simply can’t be the same as you did them 50 years ago. And they won’t be the same in 50 years time.
Think about the businesses that have been built in our generation. Netflix, Uber, AirBnB, Monzo. All of these companies, and many others, are built from the ground-up using technology. And they’ve all caused other companies to have to adapt to this new, streamlined and efficient competition.
Every bank on the planet is striving for a mobile-first banking experience. Every media company is aiming for a Netflix experience. In order to compete at the highest level, you have to fundamentally change the way you do things, and, sometimes, even what you do.
Challenges of Transformation (T)
The biggest challenge by far in business transformation is related to people. Firstly, your leadership have to have the imagination to put together vision of what the world will be like if you were to become the best possible version of yourself. That has to come alongside the ambition to make it happen.
Then, those leaders have to get your people on board. And not just on board theoretically. Nodding and agreeing in a meeting doesn’t count. You’ve got to change the culture. The underlying beliefs and attitudes that underpin the way you’ve always done things. This is often why the Netflix’s of the world can adapt to change quicker; the culture is about just that. Your culture probably isn’t, and so you have a lot of work to do with carrots and sticks to encourage and mandate the behaviour you need to enact true change.
We could write a whole series of books on change and people management, which I’m sure have already been done numerous times over. Needless to say that when transforming your business, although technically challenging, it’ll be the people and culture that’ll be hardest to change.
How to develop AI maturity
Where does your business sit on the AI automation spectrum? Are you still applying lipstick on a pig, or are you ready to leap to Level 7 and reinvent how you work? To find out where you stand and how to develop your maturity, consider taking our AI Readiness and Maturity assessment.