As chatbot technology is becoming more sophisticated and serving the customer better, it’s becoming easier and bringing a better return of investment to build a chatbot for your business.
However, this can be a daunting task with a lot of unknowns if it’s your first time.
In this guide, we’re going to cover the high-level requirements to consider when bringing a chatbot into your business in the same way we do with our clients.
We’ll also cover some examples of rough timelines, prices and provide examples of what it takes to complete each part.
So, what does it take to implement a chatbot?
The complicated process can be boiled down to these key steps:
It will typically take an experienced and specialist chatbot agency (one like Filament) ~8 weeks and ~£50,000 to get running.
From there, it’s approximately £80k for the following year to run.
Year on year after that, the cost should be expected to grow along with chatbot capability, improved customer experience and returns of investment.
Take these numbers with a pinch of salt. This is the average we found building hundreds of chatbots. Prices can go as low as £10,000 all the way to £600,000+
Chatbots can have a low barrier to entry and fortunately, there are tools being built making it easier and more accessible month on month.
However, the difficulty quickly increases as you try and get past level 1. It’s here where you want your chatbot to start integrating with other systems. That way, it can increase its ability to handle more complex use cases.
Since bot building is such a complex topic, we’re going to make things a bit easier by using what we think every company should be aiming to have by now: an advanced level 2 chatbot.
A level 2 chatbot is typically built to automate FAQs and autonomously complete authenticated/transaction-based enquiries without human intervention enquiries. To achieve this, it’s likely to be:
Without further ado, Let’s dive down into how this works.
The first and most important step is creating a business case for the chatbot program and developing a long term strategy.
You need to ask yourself some fundamental questions such as:
You’ll need to carry out the difficult balancing act of weighing up what chatbot capability can you afford, what does that get you and what return can you expect to get from that investment?:
We go through this in detail in our guide here.
Typically most businesses will already have a level 1 – basic chatbot or proof of concept (PoC) by now and are now looking to grow their use cases and scale across multiple departments or sectors into level 2.
For enterprises, the race is on to reach level 3 and beyond!
As mentioned, many companies have a proof of concept (PoC) chatbot (also called “MVP” or prototype- these terms vary) already. If you’re the exception and are looking to create your first one, then it’s still worthwhile reading this article.
Take note that your proof of concept is definitely not going to be this complex. PoC’s are usually the very minimum of every function and feature you need to demonstrate that the chatbot can function in your organisation.
Check out our guide here on how to start your own PoC on a budget.
Just because you’re going to build it, doesn’t mean your employees will use it.
One of our favourite methods to help with employee adoption and awareness is a bot naming contest.
Some key questions you need to ask yourself:
The number of times we see chatbot projects done as one-trick ponies where they think they can just do one and leave it alone.
Chatbots are not a “build and forget” endeavour. For a couple of reasons:
If you’re going to invest in a chatbot, which you should, because you simply won’t keep up with competition otherwise, then you also need to include in your plan how you aim to expand and continue increasing the effectiveness of your bot.
Front end developers, back end developers, dev ops, project managers, chatbot developers and conversation designers. In many trades, like in software development: “full-stack” developers exist.
Unfortunately, Chatbots are still in their infancy and it’s not possible yet to go beyond a level 2 chatbot without the full team we mentioned above. (This is the minimum needed for an effective customer support/service chatbot that brings you an ROI).
We write about in details what the ideal team looks like here.
Here’s a summary of the average team a chatbot agency will use:
Using the example of our taking a chatbot from level 1 to level 2: to get up and running you typically need to be looking at having at least a team of 3.
Planning the following year is hard to predict as it depends on your personal business goals and how aggressively you’re adopting new AI technology and how many use cases you want the chatbot to start automating.
Due to the current AI tech hype, many people think chatbots and AI are the best solutions to their problems.
Often, by using some design thinking techniques, we find there any many cheaper, simpler or more effective solutions to solve the problem.
This is why it’s so important to enter with an open mind.
Making sure you have an implicit understanding of your customers demographics, their language and a whole bunch of other variables is a vital part of chatbot creation.
The conversation is a speciality branch that combines a mix of user experience (UX), copywriting, psychology and technology.
There are 5 key components to conversation design, the first two you’re likely to have already:
What will likely need to be invested in from scratch is:
For a proof of concept (PoC) or level 1 bot, this may only need a couple of days of work. For level 2 and beyond, this turns into a continuous cycle as you continue to expand the bot capability and tackle each new use case, thus requiring a team of people.
We cover this in more detail in our comprehensive conversation design guide.
In an overly brief summary, there are two types of chatbot you can develop:
Depending on which of these depends on the type of historical data you need to collect.
If you use platforms like Salesforce, Zendesk or Intercom, then you’ll be able to gather all your previous conversations. The reason this is hugely valuable is you can use this data to gain insight into what the common questions your users are asking.
Depending on your budget, agencies can use techniques like word clustering to really dig deep into the data and provide a comprehensive list of intents and training phrases, each with an estimate of the value of automating that phrase and the expected ROI.
You need to go through a much more elaborate process in order to get your data to a stage where it can be used by machine learning. We go into this in an extensive depth here.
Like any project, you need key metrics to help you measure the success of the chatbot program.
The key metrics we usually recommend are:
We cover these in more detail in our chatbot analytics guide.
As we run through these next stages such as picking your NLP platform and front end interface, we will be putting together all these pieces and reviewing our architecture program throughout the process. Our level 2 bot example will look something like this:
If you’d like to learn more about the pieces of the chatbot puzzle, such as what the message broker is and how it works, check out our architecture article here!
To say there are quite a few is an understatement.
A chatbot can require an array of tools. From natural language understanding (NLU) like Dialogflow, sentiment analysis using Watson, bot management platforms & analytics platforms like EBM.
Some key questions you need to consider are:
This topic is a beast in itself and also one of the bigger decisions of this whole process. It affects your costs and future capability.
To cover the basics of what is NLU and why your chatbot needs one, read this:
When done, to find out which NLU is best for you, check out this:
For our example level 2 bot, you’re going to want to use something like Dialogflow or Watson for your NLU engine.
This question heavily affects which NLU you use as some are more capable than others.
Also, ensure you plan and budget for translation between the bots if you need multiple languages.
This should be highlighted during the design and bot persona phase and as I’m sure you already know, a critical part to a chatbot’s success is you mirror the customer demographic well.
Language is full of nuances that can easily be lost in translation. Ensure this doesn’t happen!
Most companies will have their preference. Typically we push back to deploy on to Google Cloud Platform because of how easy it is to deploy to and, from our
We typically develop and deploy an application on the Kubernetes platform. This is based on a technology called Docker and can be run in many different cloud environments. We recommend Google Cloud due to Google having developed Kubernetes in the first place and we find it is the most supported and stable environment for these types of applications.
We can also support deployment to on-prem environments, usually with the stipulation that those environments also run either Kubernetes or Docker.
This is a whole beast of a topic to debate by itself. Your company is likely to already have its preference on which server architecture it’s using.
“Front end” usually means anything to do that interacts with the user.
Hands down, the most popular messaging channel is Facebook Messenger. That being said, it depends where your users hang out and what pains you’re solving.
Secondly, the general consensus is your company should be looking to become omnichannel, which most bot platforms are now supporting.
Some of the most common channels are:
The only other thing front end developers need to worry about is if you have requirements for a custom widget.
We touch on integrations and APIs in a moment. Custom logic would be for things such as
Your business may have a lot of logic taken care of with robotic process automation (RPA) tools such as with ServiceNow or Blue Prism in which case you may also need to integrate with these platforms.
On the other side, the back end, this is where the user doesn’t see happening, such as the transfer of data, understanding the language and so on.
Tools like Dialogflow & Watson are technically part of this list, but since we covered them earlier, we’ll look at every other tool & platform you may want to integrate into.
One common problem we come across with current chatbot builders is deployment management.
Or lack of a defined deployment process specifically.
How do you ensure whatever changes your team makes get checked and approved before going live?
In particular, the likes of Dialogflow and IBM Watson have limited functionality in terms of:
One current solution is to use chatbot design tools such as botmock, or mind mapping tools like Lucid charts to design your conversations and version control. However, there is a ton of time wasted here with duplicating and many areas for errors to still occur.
So you need to ask yourself:
What processes & systems do you need to put in place to make sure that you have all the above capabilities?
This is how EBM was born, to solve all the challenges above and more.
When thinking about data security, there are two categories we consider when implementing chatbots:
For off-the-shelf software, you will want to look at the security capabilities/characteristics to make sure they meet your requirements (eg supporting end-to-end encryption) and that where there are interaction points with your own software that there is a way to ensure that security is maintained through those interactions.
For your own software, you need to go a step further and ensure that you are writing it in a way that addresses the security concerns directly.
From our experience, data privacy can be broken down into:
The reason it’s important to differentiate between the two is it affects what GDPR and data masking considerations you need to make:
From our experience, the data that could travel through your chatbot that requires security measures are:
Depending on whether your user is logged in or talking with your chatbot publicly, you’ll need a tweak your message broker accordingly. By default, message brokers in platforms like EBM help manage GDPR.
As a general tool, it assumes that any user message might contain personal information that needs to be protected and relies on the chatbot creator to use the EBM tools appropriately to ensure that their chatbot complies to GDPR.
EBM provides the following features:
The message deletion schedule allows the chatbot creator to configure a duration after a user submits a chat message when the text content of the message is deleted from the system.
Data masking allows the chatbot creator to define patterns of text that should be replaced when they appear in the user message. For example, a pattern could be made to replace text that looks like a credit card number with stars.
Building a business case and creating your first chatbot for the company is a monolithic task as it is. Since data security is such a complex and serious issue, we recommend keeping things as simple as possible for your first for by not storing confidential information
The cost, complexity and risks are high and should only be considered if it’s mandatory to proceed (which means you’ll need to accommodate extra budget for it) .
Tying in closely to business transformation management is ensuring users can discover and engage with your bot (driving traffic etc)
Yea, me too, as soon as I hear the word “legal” my brain switches off and I want to do basically anything else.
Sadly, legal is a critically important part alongside your data management.
The key considerations in legal are:
Disclaimer: This is an example of the upper limits of complexity. It’s likely your project will be much shorter.
Project Weeks 1-4:
Project Week 4-8:
Project Week 8-12:
As you have now learnt: the number of variables that affect cost are numerous.
The short answer is: the typical range to use someone like us, is:
We have an extensive guide that covers every variable that influences the cost of your build here
Creating a chatbot as you can see is a complex beast. Estimating costs is challenging and calculating ROI is an added complexity.
We have a full breakdown of how to do this effectively here.
Once the bot is up and running measuring results with chatbots is fairly easy. However, we’re more concerned with anticipating ROI.
Sadly, because it’s still such a young and ever-advancing industry, there aren’t any agreed industry benchmarks for the average saved amount of money saved in customer support centres or increased in revenue. Most examples we give are from our own results.
So we can answer this question briefly: we’ll give you the low down and the general minimum requirements we found are needed to invest in a chatbot program.
By getting our hypothetical business to a level 2 chatbot, that enables the new chatbot team to start creating and automating a conversation flow every 4 weeks at the cost of £X each use case saves the customer team increases their productivity by 6.5x, based on a traffic of 10,000 enquiries for that particular use case, they’ll get their return back in 5 months.