A conversational AI can streamline your customer service and internal operations but implementing it may not be as straightforward as traditional software deployment.
Conversational AI agents are attractive options for organisations seeking to automate customer support and streamline internal operations. But without a sound implementation strategy, the solution can end up being less than satisfactory and with considerable ambiguity about the success and future of the project. Here are 6 critical aspects to consider before embarking on a chatbot implementation project.
In most conversational AI use cases, the aim is to automate conversations in order to free up resources, save time and costs and improve productivity. This could be the number of enquiries directed at a customer support team, the number of leads to qualify through a marketing and sales funnel or queries among internal stakeholders. Before proceeding with a chatbot implementation, get an estimate of the volume of such queries that can be automated. Specifically, think about:
If the volume of repetitive queries or tasks is not big enough, you may not see a clear cost savings benefit in deploying a conversational AI.
If the volume of repetitive queries or tasks is not big enough to necessitate the need of automating them, you may not benefit from the cost savings of deploying a conversational AI.
|PRO TIP 1: As a guideline, if you receive more than 1,000 queries per month on average, then you should consider automation using a conversational AI.|
Think also about the value that the conversational AI will deliver to the broader organisation. It is useful to look at a breakdown into four quadrants as follows.
There are the “Irritants’; tasks that repeat a lot and at high volume and which are at the lower end of value-add to both the customer and the organisation. These should be the first to get automated – the low-hanging fruit.
Then comes tasks like rescheduling of appointments, checking of finances and incident reporting, which can be categorised under “Efficiency”. These add significant value to the customer if automated properly.
The conversational AI can also be a major Value-add to the organisation by scaling consumer education, first touchpoint appointment booking and automated follow-ups.
Finally, if implemented correctly, conversational AI solutions can also go on to automate tasks of a more Strategic nature – those involving highly complex interactions and requiring human discretion.
For beginners, it might be ideal to start off with the low value add quadrant automating the Irritants and then proceed to higher value quadrants such as Efficiency, Value-add and eventually Strategic.
We covered the importance of data preparation as the foundation of every conversational AI solution. The readiness of the data also indicates the readiness of the organisation regarding implementation and the timeline for deployment.
Take time to consider if there is already a substantial knowledge base of the questions and answers the bot is expected to learn. This could take many forms – product, marketing and sales collateral, pdf documents. Past support tickets, chat logs and emails can also be used, provided the bot is able to parse the information and learn.
Take time to consider if there is already a substantial knowledge base of the questions and answers the bot is expected to learn.
Often such a knowledge base may not be ready in documented form. In such cases, a customer-facing team within the organisation plays a crucial role. Through interviews and collection of data from current support channels, they will be able to supply real world queries that are representative examples to train the bot.
Other hubs of constantly updated data, such as website pages, employee training materials , internal wikis and collaboration documents could also come in handy for the bot to answer questions which are not pre-trained.
The age old question on whether to build in-house or buy from an external vendor remains relevant for conversational AI. The answer depends on several things like the in-house R&D team and operational resources available, the domain knowledge and multi language support that a vendor platform might bring in and so on.
Large technology enterprises can often afford to divert resources to create an in-house team to build chatbots. These companies often have data scientists, engineers, developers, user experience researchers and others already within their workforce. This makes creation of dedicated conversational AI teams within the enterprise quite plausible. A general guideline is that it takes at least a team of 20 to hundreds of highly skilled researchers in an AI lab, such as that of Lenovo, to achieve a certain acceptable level of performance.
For smaller enterprises that do not have a solid technology expertise or organisations like healthcare institutions, which have a large and complex ecosystem of mission critical systems to maintain, the story is different. Here, it makes sense to outsource the implementation to chatbot platform vendors and service providers.
|PRO TIP 2: For organisations like healthcare institutions, with a large and complex ecosystem of mission critical systems to maintain, it makes sense to outsource the conversational AI platform implemention to vendors and service providers rather than building in-house.|
This allows organisations to focus their resources on upgrading and maintaining their core systems and not take further technology risks in AI R&D and software development. Besides huge investments in labor, these in-house teams may lack the expertise in implementation, resulting in suboptimal outcomes. Specialised startups or vendors can fill this need. In fact, enterprises may also find it more beneficial to leverage the skills and platform features that these vendors have developed over time to achieve and ensure continuous state-of-the-art capabilities.
Other factors to consider include domain knowledge and language support. Some providers of conversational AI can be expected to have built up a specialised knowledge database in your industry.
Some providers of conversational AI can be expected to have built up a specialised knowledge database in certain domains.
The language of the users you intend to interact with also makes for a key deciding factor. Some languages are high-resource, meaning that there are numerous data resources and data sets that already exist for these. This makes it easier to develop ML and NLP models for conversations in this language.
English is easily the most well-resourced language, followed by others like Spanish, French, German, Chinese and Japanese. Others like Thai, Vietnamese and many languages from Africa are low-resource which do not have a lot of data sets already available.
KeyReply has worked on chatbot implementation projects across multiple industries, especially in healthcare and insurance. We have also built up our competencies in working with Southeast Asian languages and sentences with mixed languages, which tend to trip up many pre-trained language models.
Most of the time, a conversational AI system is not a standalone solution within an organisation. There could be many channels that you already use as customer touch points, like call centres, email, CRM, social media and messaging platforms.
For example, it could be the case that you have a business account on WhatsApp, Messenger or LINE that is active and engaged with customers. In such a scenario, it makes sense to add a conversational AI agent to this existing channel rather than create a separate app or a feature on the website.
The conversational AI needs to be able to integrate with tools, channels and products already being used actively within an organisation
To provide customers seamless end-to-end experiences, the bot will need access to backend systems. In healthcare institutions, these could include the data warehouse, electronic medical record (EMR) systems, appointment booking systems and insurance policy administration systems. The bot agent needs to be able to integrate with these systems via making calls about application programming interfaces (API) .
|PRO TIP 3: Explore options where a conversational AI can parse such information but then ‘de-identify' in order to protect user privacy.|
A common scenario in conversational AI projects goes something like this. An enterprise works with a vendor, prepares the necessary data, trains the bot and during the testing phase, different functions expect different things out of the chatbot. The call-centre teams want more useful responses, customer service heads want the bot to be more accurate, marketing teams want more leads while other departments may just want the bot to be more “smart” or “human”.
These are all different KPIs so it is imperative for the entire organisation to align at the start of the implementation. Not only that, it is important to measure and define what the “current state” is, and extrapolate that to what Years 1 to 3 would look like, with the assumptions stated clearly.
It is important to measure and define what the “current state” is, and extrapolate that to what Years 1 to 3 would look like, with the assumptions stated clearly.
Such a future-looking view is critical to ensure that the right measurement systems are put in place. This will also help to avoid conflicting goals, and allow for the refinement and adjustment of goals should the business environment experience substantial changes. What’s more, it is possible to meet all these requirements but only with the right training, testing and improvement in an iterative process over a period of time.
A well-planned conversational AI implementation can deliver the results that attracted you to the technology in the first place. On the other hand, skip any of the above key considerations and you end up with a less than ideal result.
For more insights on how conversational AI can help your organisation achieve digital transformation, contact us today.