Bitter but Useful Truth About Bots for Business. Part 2
How the value of a bot is comprised? Let’s do the math
Victor Seleykov
The first part of this article defines the main fallacies most businessmen face when implementing a voice solution, and describes ways to get rid of them. This part will show how the value of a bot is comprised — let’s work this out.
Traffic and the number of topics a bot is covering
It is much harder to trend data when you got too few queries. Let’s say you got 7000 queries per month a bot is covering 20% of it. Meanwhile, this solution costs $15000 — because your bot covers a variety of topics. This project will never be viable. Traffic and the number of topics your bot is covering are closely linked to the development cost and its pay-off period because the automation level is tied with it as well.
Even a small event agency would have a branchy scenario — children’s parties/weddings/anniversaries (which means quite a different audience a bot is supposed to talk to), bonuses and discounts, catering and venue partners, décor, entertainment, and other services. And the more branchy scenario is — the more resources you will be required.
🔥 We should not forget that every industry has topics a bot is not supposed to cover — instances when empathy is required (insured accident, for example). Or a client authentication, or negative reviews and stuff elaboration
Logs
You need to analyze your clients’ conversations in order to build and mentor a bot — a dataset formed on the basis of the chats, mail, and phone conversations are here to serve. Without this data, a bot learns really bad or way too long. You got scripts and patterns prepared? That’s fine. But if you don’t, get ready to shell out some resources, both yours and vendor’s.
Moreover, logs got to be linked with the project goals — sometimes our clients bring tech support logs to automate sales, but that’s an injudicious decision. For those who do not have their own dataset, we recommend to build a chat for their clients — to gather logs — and then use these logs to build a bot.
Channels
A bot can be implemented anywhere — a chat widget for a website, social media, WhatsApp, mail, or voice channel. Some platforms allow connecting one bot to different channels simultaneously. You can create one scenario for a WhatsApp, kick it around, and then easily connect it to the Facebook chat. Although you have to keep in mind that each platform has its own technical constraints and conceptual possibilities. For instance, a WhatsApp bot adaptation to Alexa would probably entail the redefinition of a UX scenario, additional stuff setting, and new hypothesis testing.
There are even more pitfalls in a voice channel. A phone bot’s quality depends not only on the NLU engine — but ASR (automatic speech recognition) too. To understand a client your bot has to have a nice speech transcription. In case the ASR system has poor recognition, your bot will get all the trash info and eventually it simply won’t handle it.
🔥 Another point to underline — People not only talk differently when using different channels. These channels most probably have different audiences
Besides, businesses hope to hit a paydirt with a bot in a channel that has never been promoted before (or ever existed), but it’s not gonna work. Think of how you going to push the traffic into this channel, who your audience is. And remember, a new channel would increase the number of queries, enlarge the target audience, but it will never decrease the burden on staff.
SLA requirement (service level agreement)
This actually means the depth and scope of vendor tech support. Expanded SLA implies that a client pays an extra for an ability to call to their vendor at all times of day and night to solve a problem. There are companies that have their own SLA standards — high-level tech support is supposed to solve the problem within 3 hours after it was detected.
Content providing
A bot is needed to be nurtured and renewed, and the content is needed to be continuously evolved, even during the pilot. So, it should be decided in advance — who is going to take care of that — you or your vendor.
Quantity and the number of difficulties of integrations with IT systems
A bot with no logs may show a low automation level. The presence of logs levels it up to 60–70%. But if bot integrated with internal systems (like CRM and billing) and databases, it will level the automation up to 80% and higher. It requires the forethought to define which systems you are planning to use in your work with clients, whether there’s access and whether it is easy to provide it.
Most products allow internal systems collaboration through HTTP-calls, but some complex cases (CRM system integration, for instance) may require platform support.
IT solution for a bot development
The costs of a chatbot builder vary pretty much and depend on its capabilities — ranging from customization and analytic tools to some serious business benefits like NLU (natural language understanding) techs, dialog context saving, speech recognition and speech synthesis, machine learning, and neural algorithms. Small businesses can use a lower-cost solution that do not require coding skills (any bot builder — DialogStudio, for instance). There are a lot of those nowadays and the price for an NLU product starts at $100 per month. NLU algorithms help a bot understand live speech and the context of questions, make it possible to manage traffic, and support the integration with CRM, billing, and other systems and services; but keep in mind that not every constructor offers NLU algorithms.
🔥 When picking up a solution always discuss the possibility of a scheduled update. You will need it once you start making money — you don’t want once to find out that the platform you are using is outdated, the technology has leaped forward and now you have to expend money again in order not to go back to the very beginning
Products constantly get new functionality, analytic tools develop, the quality of an NLU engine is improving. Plus, metrics, new classifiers, new APIs for third-party platforms integrations, and new channels are adding.
Here’s an example. Not so long ago a business wanted to build bot into WhatsApp, but that was not possible because messenger didn’t offer an open API. Now imagine you bought a version with no updates. And now, that WhatsApp became available, your system doesn’t connect to it, because a year ago there was no such channel. In the end, you got two options — to pay the bill for a renewal, which payback would probably take a year, or reach out for a new platform with WhatsApp onboard (then WeChat, then Skype for Business, etc.). It stands to reason that there’s more bounce to the ounce when you use one system, which keeps up to date and the market’s trends, instead of multiplying different products in your IT architecture.
On-prem or Cloud
Needless to say, a decision to build a bot inside your IT environment would impact on the cost of the project, but it also affects bot’s efficiency and it’s scaling feature.
Part 1 of this article defines the main fallacies most businessmen face when implementing a voice solution, and describes ways to get rid of them
Part 3 will tell you how to steel yourself for a bot launching
