With over 80% of all enterprises looking to implement a chatbot by 2020, as a CMO you’ve undoubtedly seen a handful of proposals floated to you. You’ve probably seen a few competitors use chatbots or have had a few peers rave about the tech behind it.
But should your brand even use a chatbot? And if you’re going to, what do you consider before tying yourself to one?
Let’s get started with the big question first.
Do you need a Chatbot?
The first question is the simplest. And the most pertinent.
Does your brand even need a chatbot?
A lot of brands lose themselves in the noise of new tech, so let’s get one thing clear – Chatbots are not for everyone.
While it is unlikely that having a chatbot on your website or app will severely hamper your brand if you choose to get one, you don’t want to invest in something that isn’t giving you appropriate returns.
As a CMO, consider a chatbot if your brand is struggling with one of these three umbrella problem-statements.
- You’re struggling with quality inbound lead generation and qualification.
- Your brand suffers from sub-par customer engagement metrics
- Laterally, if retention is a problem, a la, a lack of at-scale, quality customer support.
A chatbot can definitely help you generate and qualify more leads. It can also help you automate a large chunk of your customer support. Additionally, chatbots (WhatsApp especially) are shown to be great at engaging customers
So if your brand struggles with these problems, look into investing in a chatbot. Otherwise, maybe hold off for a while or do more research.
In-house vs. TPP
Now that you’ve made a choice about whether you need a chatbot or not, it’s time to tackle the other big question.
Do you build your chatbot inhouse?
To keep this short, 9 times out of 10 – the answer is no.
Several companies build their websites and apps built-in-house. There are pretty good reasons for this – it gives brands creative control, often reduces costs and allows quick turn-around time for fixes.
However, chatbots development differs from website and apps in one crucial regard – ML data.
A chatbots’ ability to understand and process text is powered by machine learning. In short, the more data you can train your machine learning algorithm with (aka training data), the better your chatbot will be.
To fully utilize this data, you’ll need a great team of dedicated ML enthusiasts. You could choose to repurpose your current team to a bot project, but you’ll lose out both on opportunity cost and more importantly, quality.
A bot that you build in-house is not going to have as much training data to work with. Your algorithm won’t be robust enough, the people manning them won’t
And an untrained bot simply won’t be able to process, understand and reply to messages from your customers.
A TPP that works with a large number of companies across industries will have a lot of training data for their bot – therefore making it quicker, smarter and more accurate.
What’s better than a provider that has a lot of training data?
A provider that has a lot of training data for your specific industry.
Not all data is created equal.
A dataset for real estate queries will vary humongously from a similar dataset for e-commerce queries. “Returns and refunds” will dominate intent classification in ecommerce queries, but the same term will be non-existent in real estate.
A chatbot provider that has several clients but in just one industry – may not be able to provide the kind of conversational coverage your brand desires.
To get a better understanding of where the chatbot provider stands, consider asking the following questions –
- What are the gains their existing clients in your industry see?
- How much training data do they have for your vertical?
- What is their prediction accuracy for other customers?
Think of chatbots as a pair of golf gloves.
You can have the best, prettiest and most functional gloves on the course. But if they don’t fit your hand, they’re not going to help your game.
It’s the same principle when you’re choosing a chatbot. Even the cheapest, most powerful, and aesthetically pleasing chatbots are useless if they don’t integrate with your existing tools.
A chatbot that's generating real estate leads from your website is useless if it can't push that lead data into your CRM. A chatbot that's handling ecommerce customer queries is useless if it can't pull tracking data from your OMS.
As a CMO, your brand uses hundreds of software providers to track, engage and service your customers. If a chatbot doesn’t operate in sync with these services, it’s going to cost your brand time, money and resources.
Ensure your chatbot provider integrates with a wide suite of tools. Ask questions about future integrations, timelines, and how receptive they are of custom requests.
When extrapolated properly, a chatbot is one of the world’s most powerful behavioral analysis tools.
Simple decision-tree heatmaps tied to complex keyword mapping allow brands to have unprecedented insights into the way their customers think, act and behave.
So what should you look for when comparing analytics amongst prospective providers?
A simple rule of thumb is that your analytics should appear as a funnel. At the very top – a massive data dump. At the bottom – easy to read visual representations of said data. In between, reports based on your specific requirements.
A dump usually presents itself as a .csv file filled with those days/weeks/months chat logs. This massive trove of raw data gives you a lot of free reign. You can sit with your team and paint broad strokes to identify general trends or dig deep to observe even the finest details.
This data dump should be paired with mobile-friendly graphs and bars – ideally on a dashboard. This one-glance interface keeps you posted on key metrics that YOU define as important.
When choosing a chatbot, make sure to ask your prospective providers to send you templates of these reports, dumps, and dashboards.
If you’re the CMO of an enterprise company and you’re dabbling in new(ish) technologies like chatbots, you’re probably worried about the all-important word – scale.
There’s no easy way to determine whether a provider can handle your scale. Or, if in fact, your scale is too much for them at all.
Talk to them about what their message load (in an hour) is. Quiz them about who their largest existing customers are and how many conversations they handle in a specific time frame. Get more specific and ask about concurrent message loads, latency metrics, number of concurrent active agents and how many outbounds they can process.
Most importantly, ask about their future plans to scale up to understand if the vendor can keep up with your growth.
Freetext is defined as an “out-of-the-flow” interaction that customers have with bots.
So off-script questions like, “what’s the weather in New York?” or “who made you” qualify as freetext.
But most questions aren't that ludicrous. Often, customers are just trying to learn more about your product or solve a query that they deem important.
For example: During a lead generation flow, a customer may have questions about the pricing of your product. To ignore the questions and prompt an error doesn’t make for great customer experience. Instead, you want your bot to answer that question and then move on.
There’s also the matter of spelling mistakes and portmanteau words (eg: Spanglish, Hinglish, etc) that complicate a bots’ ability to handle free text.
No chatbot vendor is going to be 100% accurate with free text. Your brand may not even need it. But if you do, it’s important to understand how well your vendor can manage free text.
Realistically, pricing is going to be the most important aspect of your decision-making process.
Different vendors use different pricing strategies based on a variety of factors. And these strategies often make a HUGE difference in how you’re billed.
Keep in mind that there are two aspects to chatbot pricing – a flat fare + surcharges for add-ons. With that in check, there are three basic methodologies to calculate pricing for a bot.
- Cost per message: ideal for easy-to-close interactions.
Cost per message is the easy one here. A message is defined as a string of alphanumeric variables exchanged between your brand and the end-user. You're charged a certain amount for each message, and that figure is usually in cents or a similar nominal denomination.
- Cost per chat/conversation: ideal for interactions with lots of back-and-forths.
A conversation is a collection of messages. It doesn't matter how many messages are in a conversation, cost per conversation charges you for the number of conversations you have in a specific timeframe. This pricing is usually tiered in nature.
- Cost per customer/contact/user: ideal for a small, dedicated user base.
Cost per customer is a bit tricky. Often used by enterprise-style chatbot providers, cost per customer means that you pay for each unique customer that you interact with. Even a single message or conversation adds to the sum total of customers, assuming they're talking to you for the first time.
While vendors may use other synonyms, they’re likely talking about one of these three umbrella terms.
Identify what kind of pricing works for your brand and negotiate pricing based on that.
‘Cost per conversation’ is the ideal plan for a medium-to-large ISP, for example. An ISP would have hundreds of thousands of users asking millions of queries. The other two plans would cost the ISP at least 4-5x more.
The best answer to the question “in-house vs TPP” is actually quite boring – it’s a combination of the two processes.
Combining the expertise of a chatbot provider with the creative control of an in-house product is key to your brands’ success. This is what the DIY (do it yourself) function of most modern software aims to enable.
Being able to edit your brands’ bot by yourself allows for quicker turnaround time, easier experimentation, and better personalization.
Even if you’re the CMO for an enterprise firm, you want a UI that’s easy enough for anyone in the company to use. Get your hands dirty during the sales process. If that’s a bit out of reach, ask if you can see a demo the bot building process to gauge the platforms’ DIY-ability.
If you’re the CMO of an enterprise company, you want a dedicated customer success manager to handle your account.
No ifs, and or buts.
During your decision-making process, remember that you're choosing a good success manager just as much as you're choosing a good chatbot.
A chatbot provider that doesn’t give you a success manager should be taken off your list immediately.
A good customer success manager is your walking, talking key performance indicator. They’re charged with understanding your requirements, creating an action-plan, onboarding your brand on to the platform, tailoring bespoke solutions for you, monitors your progress and helps you optimize your flows to see maximum ROIs.