This is the year of AI; where we see discussion about AI in every forum, from World Economic Forum, to every Technology company, to every Government, to Environmental bodies, to NGOs, to Industrial bodies and to individuals like you and me. AI is touching every one’s life and have a broad spectrum of feelings about it. Some are excited to the brim and some are frightened to the core; others fall in between. The discussions hover around topics like Benefits that AI can bring-in in terms of efficiency & effectiveness, challenges in terms of misuse, regulations that might be required, environmental impact of the energy usage, Data privacy and Copyright protection, Job loss and job creation, new strategies for business and war, realignment of human life, etc. AI has potential to disrupt everyone’s life; everyone is trying to realign and cope up with the change. This paranoia is justified to an extent as we have just seen the tip of iceberg. We have seen what technology is capable of in the laboratories and there is a lot of speculation on what is its full potential. So far three themes have emerged for industry to use i.e. Conversational AI, Generative AI and Agentic AI, where the technology has been productized for specific business scenarios. Significant experimentation is underway to test them and get the advantage as an early adopter. Multiple options are available for companies and they have to make a right choice to get the best ROI. One of the major challenges with this new technology is the investment required, even for prototyping or building a POC. Companies are in dilemma of where to invest and looking for help in making this decision. Since the technology is new, results at scale are unknown. ROI calculations do not give correct results as they are based on assumptions of unknown. Organizations are seeking additional validation, which looks at this situation differently and helps making the investment decision. I am going to talk about one such approach using Differentiate Needs Pyramid in this article which can help organizations in their decision-making process.
Let us quickly look at what is Conversational AI, Generative AI and Agentic AI.
Conversational AI is a type of artificial intelligence that allows computers to simulate human conversations. It uses natural language processing to understand and process human language. It can help machines interact with humans in a more natural way by simulating human conversations like Chatbots, Virtual assistants
Generative
AI is a type
of AI that can create new content like text, images, and videos. It uses
large AI models, called foundation models, to learn from existing data and
generate new data with similar characteristics. It can create
conversations, stories, images, videos, and music. It can learn human
language, programming languages, art, chemistry, biology, or any complex
subject matter. It can help streamline the workflow of creatives, engineers,
researchers, and scientists. e.g ChatGPT, DeepSeek
Agentic AI is a type of artificial
intelligence, that can act autonomously to achieve goals. It's designed to
interpret context, make decisions, and take actions with minimal human
intervention. It understands the context of a problem and the user's goals.
It uses machine learning, natural language processing, and automation
technologies to make decisions. It can adapt to new information and changing
circumstances. It can learn and improve through each interaction. It
is like creating a autonomous co-worker.
Let us take an
example of most popular playground for implementation of AI i.e. Service Desk.
There are multiple products and use case scenarios covering Conversational AI,
Agentic AI as well as Generative AI. Let us assume that a company wants to
introduce AI in its service deck operations for Claims processing. The
objective is to provide a better customer experience as well as reduce the cost
of operations in long run. Now we will assess the situation using Differentiated
Needs Pyramid. Let us first draw Differentiated Needs Pyramid for the
Service Desk Service with two customer groups in mind i.e. Customer who is raising
the claim and second, Management, which is internal customer. Leader of Service
Desk operations is trying to improve experience for both set of customers from
the operations by implementation of AI. The picture blow showcases what are the
needs at different levels for each customer group. The customer experience improves
as the needs at higher levels are satisfied.
Organization would have identified multiple use cases to be implemented so that collectively they give the required boost in customer experience as well as bottom line desired by management. We now need to look at all the use cases and test their impacts on each of the needs. If the impact is positive, customer is happier with implementation of that use case. This is a “thumbs-up” for implementation from that customer segment. If the need is not going to get satisfied, then customer is going to struggle and unhappy. This is “thumbs-down” for implementation, unless use case is tweaked to satisfy the specific customer need or additional investment done for training.
Let us assume that
for this organization, two needs “Immediate availability of agent- Level 1 need”
and “Agent helps through the entire process – Level 3 need” are not
satisfied for customer. To address this, organization wants to implement a
simple use case for “Initial conversation with customer and providing him
the status of the claim along with some information” as a part of conversational
AI implementation program. The ROI calculations look favorable and need
additional validation that ROI calculations will materialize. We now need to assess
the impact of implementing each step /action in the use case on the level of need satisfaction/ customer
experience improvement. If we get positive results, the customer will support
this initiative and it will be a success.
The high-level steps
could be as follows
- Customer Calls to understand status of a claim and what needs to be done to expedite the same
- Within 3 rings, Conversational AI bot picks up the call (This is a big positive with respect to needs satisfaction, if the conversation moves smoothly, customers will be happy to adopt)
- Bot welcomes customer and asks reason for call
- Customer tells the reason in his own style
- Bot deciphers customer answer and replays its understanding to customer
- Bot keeps on asking differently until its understanding matches with customer (There is a rule set up that if the understanding does not match within 3-4 tries, the call will move to human agent queue – This rule is applicable for every conversation) (if bot is able to understand the conversation, this is very positive. The % of time, bot needs to hand over the call to human agent queue will determine the success. Higher the % lower is the satisfaction. This also tells that model will need more training to be successful, i.e. more investment for the organization)
- Once the understanding is correct, bot asks specific questions (specific to a situation understood through earlier conversation) to get identity of customer and fetch the record.
- Customer provides the information
- Bot will try multiple options if the primary option is not available with customer during the conversation. This conversation to continue till bot gets the information to fetch the record from the system
- Bot shares another parameter fetched from the system with customer to reconfirm the identity.
- Customer confirms and conversation moves to the next step
- Bot provides the status of the claim to customer e.g. it is pending for xyz approval
- Customer asks how this process could be expedited, (or any similar question)
- Bot again performs
the routine about matching its understanding with customer and once match is
achieved, provides an answer e.g. they want some additional information from
customer (Success
here depends on training for open ended questions. Higher the training, better
results – overall customer satisfaction depends on completing the transaction
in one channel i.e. through bot or through agent. If they need to shift, the
waiting increases, resulting in dissatisfaction and reluctance for usage)
- Customer provides that information
- Bot records this information in to the system and confirms new date for approval
- Customer thanks bot
- Bot asks if there is anything else that customer needs
- Customer says that he is looking for information on an unrelated topic
- Bot again confirms the understanding and directs him to either another bot or a human agent for taking it forward.
Similar assessment could be done with needs of other customer segment i.e. Internal Customer- Management to see the impact
We need to look at
all the use cases under consideration using differentiated needs pyramid to
identify impact on the customer experience. This feedback is then combined with
the ROI calculations regarding investments, savings, productivity gains etc.
can provide correct picture of what will work and what will not. We consider
volume reduction as one of the parameters for considering savings, this reduction
will occur only if the “Complete” transaction is executed by bot. If bot has to
transfer the call to human agent, volume reduction calculations go wrong and
ROI is not materialized. Thus ROI calculations supported by Assessment through Differentiated
Needs Pyramid gives confidence to organization to go ahead with AI
initiative.
Detailed explanation
about Differentiated Needs Pyramid, its construction and usage are available in
my book “Customer Experience Decoded” available on Amazon https://www.amazon.com/dp/8195052657.
You can always reach out to me for any discussion you may want to have about
your specific situation.
