IT ecosystems: how to create platforms from services and why it is needed

Bots: test for durability

Speech synthesis and recognition technologies have become more

democratic:no programming skills are required to create a primitive voice bot, and convenient no-code platforms make it easy to develop skills for voice assistants. There are even services that can "clone" any voice, and voice deepfakes are becoming more accessible and realistic. On the one hand, the industry became democratic, but at the same time, the market was filled with "raw" products. Unsurprisingly, Gartner analysts have recognized chatbots (both text and voice) as an overrated technology.

Opportunities of bots in some areasreally overestimate: for example, it is difficult for a virtual interlocutor to conduct abstract dialogues, joke and show empathy. But the lack of developed EQ and sense of humor does not prevent robots from successfully handling millions of calls in call centers around the world. According to Invesp, over the past year, 67% of consumers had at least one conversation with a bot, and in 2020 the number of conversations with bots increased by 426%. The number of successful cases is also growing: for example, the bot of the American railway company Amtrak processed 5 million support requests in a year and increased its revenue by a third.

But not all companies are able to benefit fromrobots. The problem is that businesses often launch a virtual assistant “for show” and do not integrate it with key services. As a result, bots do not work efficiently enough, and managers and marketers do not interact with them in any way, doing their tasks. The solution to the problem is platform omnichannel ecosystems - a new trend at the intersection of AI technologies, management and marketing, which is changing the approach to business communications.

Ecosystem elements

If a company develops custom-made voice and text assistants, it offers product... The client decides how to use the bot and sets up the integration himself. For example, Botsify and ManyChat work according to this model.

Platform ecosystem operators act differently:they create a foundation to which various instruments can be connected. A bot itself is a “cog” of a mechanism, and an ecosystem is a ready-made mechanism with a thousand of such “cogs”. The ecosystem model is widespread in many markets, from fintech to e-commerce, but it is a recent emergence in the virtual customer service space. There are three key features that distinguish it from a stand-alone product.

  • The ecosystem is more stable.

A company with an average of 200-500 peopleuses more than 120 SaaS solutions, and it is not always possible to build synergy between them. Connecting virtual operators often leads to more chaos. Startups often offer standalone point solutions: for example, some create a virtual operator to order - write scripts and synthesize replicas. Others offer only bot platforms, and still others - "screw" billing systems. It is not always possible to integrate these tools into a CRM system and "make friends" with analytics services. As a result, technologies from different providers conflict with each other and are not efficient enough.

Platform providers usually offer a combinationturnkey services: for example, synthesis and recognition service, transcription and notifications, as well as analytics. They usually provide the customer with access to a personal account with various modules - one employee can manage them. In our practice, there were cases when such monitoring made it possible to coordinate 900 people at the same time.

Some voice platforms work withclosed model and include only their own developments in the ecosystem, others embed third-party solutions into the infrastructure - for example, in TWIN we combine our own TWIN ASR / TTS technology with speech recognition and synthesis systems from Yandex and Google. At the same time, our task as an ecosystem provider is to make sure that everything is working stably and smoothly, and the services do not conflict with each other.

  • Ecosystems are built on omnichannel. Customers do not like impersonal calls andprefer a personalized approach, so a bot that is not included in a unified communication system irritates them. Such a virtual operator always calls at the wrong time, uses an unsuitable communication channel and, in general, does not take into account the wishes of the client.

Many people, in principle, do not like to talk on the phone and prefer instant messengers: according to statistics, 55% of consumers are more likely to use the services of a company if it is possible to contact it in the messenger.

But how do you know what exactly users like?The most effective way is to measure conversions and collect analytics. To do this, you need to monitor the work of each operator, and this is a long and laborious process. Alas, in Russia up to 80% of companies do not collect statistics on each employee and communication channel, so they do not know what works and what does not. It is really difficult to measure customer service conversion because it is not measured by sales and revenue, but by a complex combination of metrics. And to collect them, you need a smart analytics system, including BI tools.

At TWIN, we collect and account for hundreds of parameters.For example, we call the client and determine what device he is using - if it is a smartphone, then you can send a short SMS with a link to the company's website, and if it is a push-button phone, then we send him the details in the message. If we know that a client uses Telegram, then we send him messages in the messenger and generally stop calls, and instead of a voice bot we use a text bot.

The same principle should be applied for mailingnotifications. A client who uses a mobile application receives a push notification, and to the rest we send a link to Viber, Telegram or WhatsApp - much depends on which communication channel the user prefers and what data he has provided to us. This approach is only possible if the company has connected analytics tools, and the voice technology provider has access to them. These are the classic principles of omnichannel, but now not only real operators, but also virtual employees rely on them.

  • Fault-tolerant architecture. The platform ecosystem consists ofnumerous modules that are hidden "under the hood" of the service. This microservice architecture helps the provider to provide services around the clock without compromising quality. Since the operator does not rent a single server, but uses a whole network of distributed servers and data centers, his system is less vulnerable - it cannot be overloaded or collapsed. If one module fails, the rest are activated. And if the customer needs additional capacity, the provider connects auxiliary servers and data centers.

For technologies based on machine learning, sucha solid foundation is a must. Imagine what it would be like if a robotic car suddenly stopped recognizing objects on the road, because "the server is not responding." In the case of bots, the main thing is that the virtual assistant remains in touch and can stably maintain a dialogue with any interlocutor. To do this, during recognition, our bots sometimes ask for an answer option from several neural networks at once and automatically select the most relevant one. This also has a positive effect on fault tolerance. The use of redundant systems, including third-party ones, is a normal practice when developing solutions based on artificial intelligence.

How do platform ecosystems help businesses?

Platform ecosystems make it easy to communicate withcustomers, optimize processes and generally contribute to marketing "warm-up". The company spends less time on ineffective calls, and most importantly, it faster evaluates which tricks work and which don't. Virtual operators are also cutting costs - on average, according to our calculations, one minute of voice bot operation costs 5-7 rubles, including all additional costs. The operator's services will cost 10-15 rubles if you conclude an agreement with a third-party call center. When creating your own CC, one minute of an employee's work will cost 35–45 rubles. Many companies do not even suspect how much the operator's services actually cost: they usually divide the employee's salary by the number of formally worked minutes. But not a single person works non-stop without pauses and downtime, and many costs are simply not taken into account: for example, the maintenance of a CC, bonus and social payments.

Because of this, the benefits of bots may not be appreciatedthat's it: if the business isn't doing analytics and tracking key metrics, the voice ecosystem won't do it any good. Therefore, for now, the main customers of omnichannel platforms are digital-first companies that have digitized most of the processes. Banks, transport and freight services, and online retailers benefit most from voice services. At the same time, 58% of clients work in the B2B segment. Most use ecosystems to make communication with them more efficient: with the help of virtual operators, brands generate leads, increase conversions and reduce support costs by an average of 20%.

What's in the future for platform ecosystems?

Gradually new ways appear on the marketapplication of voice services. For example, some people use bots to work with newbies and trainees, as well as to establish internal communications. The voice assistant calls employees, sends them invitations and meeting reminders - both real and virtual.

Ecosystems will also connect more and moremicroservices - TWIN offers 12 different add-ons, including emotion and gender recognition by voice. Some are experimenting with age definitions as well as biometrics. Add-ons that improve bot performance are becoming the new standard. For example, autoresponder recognition services - with their help bots automatically perform this function and promptly end the dialogue.

Another challenge for voice developers isit is a continuous improvement in speech recognition and synthesis. For example, we manage to accurately determine up to 95% of the spoken text - this is the standard on the market and it is still difficult to overcome it. Many companies are trying to raise the bar, but each percentage is hard to come by. Algorithms have already caught up with humans - now the task is to go beyond human capabilities, and this is not easy.

See also:

Abortion and science: what will happen to the children who will give birth

Check out the most beautiful pictures of Hubble. What has the telescope seen in 30 years?

An artificial object was found in soil samples from the asteroid Ryugu. Like this?

Facebook Notice for EU! You need to login to view and post FB Comments!