Support chatbots have the purpose of automating a task that, until now, could only be done by humans. In theory, any company could offer instant support around the clock, but the labor costs of doing so are simply too high to make it a realistic option. This is where chatbots come in and offer their savings potential.
However, any chatbot first incurs costs, which is why it’s important to calculate their ROI before implementation. Unfortunately, the field of chatbots is so dynamic that you need a good overview of which costs are involved.
In my first post in this series, I wrote about chatbot confusion and proposed a taxonomy to help differentiate chatbots. Now that it’s clearer what chatbots are used for and what they’re capable of, I want to address another important question I deal with every day: The ROI of a customer support chatbots specifically.
To frame it right: The following applies to text-based customer support chatbots only. Not for voice bots or social, branding, or sales-oriented chatbots (as defined in question #1 and #2 of the taxonomy).
There Aren’t any Good Calculations Yet
My research didn’t yield many insights. Some calculations were well written, but too generalistic, well explained but don’t offer a tool, a good starting point but don’t offer specifics or only available after downloading a White Paper.
There are a lot of variables that will impact the ROI of your customer support chatbot. Beyond the volume of requests you’re receiving and the resolution rate of the bot, you need to consider the volume of requests you can actually route through the bot, set-up costs, ongoing licensing fees, and much more. This is, what this post tackles.
We regularly do this calculation for our customers and know both which costs are involved and what returns are possible. I’ve turned our internal calculation into a generic template to share here, since it might be useful for you when making a decision about chatbots.
Sharing our ROI Calculator
We have deliberately kept it simple. Our goal was to highlight the key parameters of an ROI calculation in a human-readable form (for which you don’t need any instructions). You can download it on the link below
Hint: Use a split screen to open the ROI Calculator on the one side and this article on the other to get a great overview and easily follow the explanations:
- First we need to take into account the salaries of people involved in support. This allows us to calculate the direct cost and future savings.
- From the amount of tickets and number of agents in a support organization, we can derive the average price per contact.
- Routing requests to the bot is super important — if you train it but nobody uses it, there will never yield any return.
- Lastly: If you’ve trained a bot and your customers use it, it’s important that it truly helps customers.
Need-agent cases v. self-service cases
This may not be straightforward at first, but it’s important to differentiate:
- A self-service case means that the customer can “help themselves”. A password reset, for example, is usually a 100% self service case. The savings are 100% here, as the customer doesn’t contact your support team at all, but solves their problem with the help of your chatbot.
- A need-agent case means that an agent is needed to perform the desired task for the customer. Even after the bot has provided the correct solution, a human is needed to perform a task (e.g. “canceling an order”). In those cases, the request is correctly identified, still some work is to be done by a human. Not all costs are saved.
Important here is to take into account the actual occurrence of requests. In our experience, the need-agent cases are less popular as companies try to make the most common requests a self-service case as they continuously optimize customer experience. But still, a good portion of requests are need-agent.
With this in mind, let’s dive into the phases of your project and the associated costs.
Every bot needs to be set up. We call this training, some call it configuration. It should typically involve the following steps:
- A kick-off session to explain the scope of the bot, brainstorming the use case, setting goals, discussing responsibilities, and drafting up a project plan.
- Integration of the bot with the current IT infrastructure. Depending on the type of bot, this can either be “plug and play” or a complex tech project.
- Training of the bot. It can only be as good as the input, but since training is done in various ways, the amount of required work depends on the bot provider. Nearly always it involved human labor to either train a system directly or generate training data.
- Communication and feedback. Part of the support staff should try out the bot and give feedback. Also, account for proper communication with the support team to explain the scope and capabilities of the bot.
- Maybe: External support. Some bot providers’ business model is to charge for setup — which incurs costs you need to take into account.
Savings in the Live Phase
This is an easy calculation:
- Number of tickets routed through the bot x cases where the bot is right x cost per ticket
Costs in the Live Phase
- A support chatbot doesn’t function autonomously. Someone needs to maintain and improve it and the labor costs are an investment.
- The software provider charges a license fee.
- If you opt for paid support by an external agency, take those costs into into account.
Payback Period as Key Metric
Considering the initial phase of an investment and the ongoing costs ≪ ongoing gains, we can then calculate a payback time of the investment. This means the ROI of any chatbot project boils down to when the payback will take place. We measure that time in months.
Receiving payback within three months, for instance, is great — but if it only takes place after 36 months, you might wonder whether a project is worth it.
Find below a screenshot of the parameter section if you haven’t opened the ROI calculator yet.
Some Combinations Don’t Work
By downloading the ROI Calculator and playing around with the numbers, you’ll quickly find combinations that simply don’t work. If the ongoing costs are ≫ ongoing gains, you should kill the project.
If there are monthly gains, there’s another question: How big are they compared to the investment?
Obviously, the bigger the investment, the longer the payback period. It’s a linear relationship. If you plan on investing + €100k before even taking your chatbot live for example, you’re likely to have a very long payback period.
So let’s keep the investment fixed for now and play with the parameters for the gains.
The most critical factors are:
- # of tickets in general: If you have a small support team, the effort required to set up a chatbot might be too big. There is a minimum team size you’ll need in order to justify the costs. We advise our customers that support chatbots only start making sense if your team is > 10 agents.
- % support requests routed through the bot. If you only intend to route 5% of your requests through the bot, you will never see any significant decrease in contacts. This is simple math: The more requests you route through the bot, the higher the potential gains. We advise our customers to route more than 20% of their requests with the bot.
- % of correctly solved requests. If the bot doesn’t solve the customers’ requests, what is it good for? The higher its accuracy, the better. The majority of cases should be solved correctly. This is important, as the number implies all cases where the users quit the conversation, the algorithm stops and hands over to an agent or when users are proposed a solution but nevertheless want to talk to an agent.
- Human training needed: At Solvemate, we only require human training in a “software as a service” context. No programmers are needed. For a medium-sized project handling thousands of requests a month we calculate approx.10 days of a project manager for the same month.
- Software fees: Chatbot software isn’t free. But the fees should only constitute a fraction of the savings from deflected support requests.
- External labor/agency: In my previous post, I have mentioned that chatbot providers who act as “digital agencies” can ruin your ROI. If you need to pay significant amounts of maintenance costs, the project might never be worth it.
Why Chatbots Fail
Take a close look at the spreadsheet: If chatbots are done right, they are a no-brainer in terms of ROI. But very often, chatbots are considered a “strategic project” with the overall innovation being most important. I have talked to customers who spent a €100.000+ budget on a custom-build support solution they never used — just to prove they were “being innovative”.
Needless to say that those kinds of projects can’t even achieve a ROI on an operational level. But if you’re dealing with a real-world operational project, calculating the ROI is instrumental in setting yourself up for success. I hope that this article and our ROI Calculator will help you do that!
Give it a Try
My goal was to create an easy ROI assessment for a support chatbots. Even if the result looks a bit daunting (A > 70 row spreadsheet with > 20 assumptions), you’ll quickly note that calculating ROI isn’t rocket science.
As sample inputs, I’ve added a case of using Solvemate for a generic enterprise customer. Our customer see a payback within a few months time.
Call-to-action: Go and explore the ROI Calculator yourself (download it here) and play around with the numbers!