2020 letter from the CEO
The pandemic challenged us, but made us a better company
Sales dipped, then bounced back
The first half of the year was a struggle. The pandemic hit and from one day to the next, we moved to a work-from-home model, revised our business plans, and drew up new scenarios for the future. As a SaaS company with recurring revenues, our business model is robust - we even managed to grow slightly in the first half of 2020, although our biggest client churned due to the pandemic.
Q2 ‘20 was our worst quarter to date in terms of closing new business. This is understandable as customer service teams were struggling to bring all their agents to home-office and needed to deal with operative fires burning - it was not the time to tackle a cool automation project.
I am proud to say that the 2nd half of 2020 was the best half-year in terms of winning new business. We have been able to win awesome clients like UK-based eCommerce marketplace MusicMagPie, the famous Swiss sports brand On-Running, or, in our new geography, the well-known Swedish retailer Trademax.
Yes, the pandemic is still with us, but the general push for innovation and digitalisation has now also reached customer service teams and will give Solvemate strategic tailwind. That’s one of the reasons why I believe 2021 will be the year of the bots.
How it made us better
On May 26th, the company took a unanimous vote to change our culture deck, a public document which acts as our constitution. We changed our remote work rule to:
“As a company, we default to offering full flexibility for colleagues to choose [to work] in the office and/or [to] work remotely.”
Looks like a small change, but it had a cascading effect. In the following months, we have hired our CMO Sylvia Jensen, working from London; a developer working from Portugal; and an operations analyst working from Ireland. None of this would have been possible before! By becoming a remote-first company, we have a structural advantage of hiring the best international talent and are no longer bound to finding talent that is willing to relocate to our headquarters. I am proud to have 20+ different nationalities - and counting - in the team.
A second benefit is that existing colleagues gain flexibility and can live where they want (without the need to pay the extraordinary rents in Berlin). Our Danish colleagues work from Copenhagen, and a Spanish colleague has relocated to Madrid.
Once the worst of the pandemic is over, we plan to host quarterly multi-day “homecomings”, where we do goal setting workshops, strategic brainstorming, and socializing. Travel and meeting space would of course be paid by the company, but this is more than counterbalanced by rent savings from going remote.
We researched, experimented, and put a lot of thought into how to make remote-work “work”. We have written exhaustive internal documentation about transparent (over-)communication and decision documentation, using the right tools, having an excellent home-office and video-call setup, and staying connected with peers. Last week, we had our first annual virtual “offsite”, where we discussed our 2021 strategy. Before the event, I had mixed feelings, but thanks to clever use of technology and thorough preparation, the team was excited and felt as connected as they did during our previous physical offsites.
Ultimately, the forced remote work brought me to the conclusion that, as a high tech SaaS company, a distributed team is more effective and efficient in the long run for Solvemate due to a) access to better international talent, b) better work-life balance, and c) a more organised way of working.
2021 will be the year of the bots
Intercom states in their 2021 customer service trends report that 50% of support leaders plan to invest more in automation, including chatbots, to increase their team’s efficiency. Further, I have been following the Gartner Hype Cycle for Artificial Intelligence for the past four years and this is how chatbots have moved along the hype cycle from 2017 to 2020:
We see that chatbots had reached the peak of inflated expectations in 2019; and I believe that they will begin to ascend the Slope of Enlightenment in 2021. According to Gartner, the final plateau of productivity will be reached in 2-5 years.
Chatbots are crossing the chasm right now. 2021 will be the year of the bots.
Product vision 2021: The central service automation software
Our overarching product vision since our foundation is to empower customer service with automation and insights to provide the right help at the right time. Every year, we boil down our product vision to a yearly goal. For 2021, this will be
Our platform will be the central service automation software that owns the customer service flows.
We see the general trend of CS software, CRM, and CPaas merging together. We are official partners of Zendesk, Freshworks, RingCentral, Salesforce, and many more and will continue to deepen our relationships. I have already written an article on why I think bots ♥ CRM.
In order to achieve the above, we are launching two strategic features:
Voice as a channel
We are already offering our bots on all text-based channels (with our In-App Webview, and the Beacon for the Website and as well as Messaging channels). However, we see voice as being an important channel for customer service which will continue to be relevant in the coming years. Already in H1 2021 we will have built a beta version where our bot experience will be available on voice. Following that, our omnichannel offering will expand to speech-based channels and vastly increase the value proposition to our customers.
No-code automation sequences
When connecting to third party systems to trigger processes or fetch information, developers are needed to write small scripts which the bot is executing. We have learned that customer service departments ideally want to be less dependent on IT and in full control over their processes. We will abstract the coding layer, so that - after an initial setup - CS teams can build API calls and improve CS processes without coding.
Separate from the crowd
There are a lot of “bot vendors” out there talking trash. Full stop. However, there are a few true customer service automation platforms that have a track record of case studies that deal with significant improvements (see ours here) like contact volume reductions of up to 40% while increasing CSAT.
In 2021, it will become clear to customer service leaders that chatbots are only the tip of the iceberg; it’s not only about automation, but about creating meaningful conversations between brands and end consumers.
“Meaningful” conversations are those with significance, purpose, and value. I love the value-irritant-matrix for its simplicity, so let’s look on the right side of it:
We automate high-value tasks for end consumers that are of low value for companies. Our bots deliver instant answers and real time resolutions, are always available, and deliver a personalized experience.
We leverage conversations by enabling 1:1 conversations between humans, where human empathy is needed or desired.
My prediction is that CX leaders will in 2021 come to understand that a good service automation strategy consists of two parts:
Understand what the user wants
Automate the user’s request (or decide for human contact)
Good customer service is not about having a bot up and running: it's about creating meaningful conversations. At scale. And for that, leaders need a customer service automation platform that is integrated with their CRM and their tech stack; that authenticates users; that delivers a personalized experience to every single customer; that gives them incredible insights; and that lets them define and optimize their CS processes.
Why we win
An analyst from a big market research firm asked me the question: “There are hundreds of vendors out there - why do you win?”
Our bots deliver the best return on investment, because our Contextual Conversation Engine finds the right intent with little maintenance effort and our Automation Builder is highly flexible to define any automation or escalation strategy.
In order to explain this, we need to take a step back and deep dive into both topics mentioned above: (1) Understand what the user wants and (2) Automate the user’s request.
Understand what the user wants (with our Contextual Conversation Engine)
In the overview graphic below you see that Static Logic Trees and NLP-first both each have their disadvantages:
Static logic trees - there are plenty of cool browser-based static decision tree builders which get up and running quickly. However, there is no AI involved that supports the bot manager. They do work for easy, small bots, but no human can efficiently maintain a monstrous decision tree with 50-100 intents. That’s the reason why expert systems already failed in the 80’s and 90’s, and that’s the main limitation of the static logic tree builders of today.
NLP-first bots - there are also plenty of NLP first tools, which typically also have a service agent interface and start by providing text snippet suggestions for the service agent to improve their response time in live chats. However, there is two issues with NLP first bots:
Training an NLP-first bot takes time and human effort. I have seen companies investing hundreds of thousands of Euros into a team of bot managers to review and refine the NLP models. But keep in mind: each minute spent on bot training must return a magnitude of minutes saved from the actual use of the bot! Initial and ongoing human bot training efforts are a key input for the ROI calculation.
Humans are still needed due to a lack of understanding. I am sure that every reader of this article has experienced the “chatbot frustration” themselves. Understanding natural language in a multi-step conversation is just too hard. That’s why agents are always only “one click” away with those tools. Their main value is augmenting agents, since they are not able to fully automate requests. But augmenting agents only creates a fraction of the value for the company as agents are still involved, while having a mediocre experience for the end user.
Sidenote: Don’t get me wrong - I love NLP, I use it every day to talk to Siri or dictate my instant messages and it’s an incredibly powerful technology that creates a tremendous amount of value in the world. It’s just that for CS automation, we need s.th. different in addition.
Both Static Logic Trees or NLP-first bots are not the ideal solution. If you follow the product strategies of both types of vendors, you’ll see that Static Logic Tree vendors are adding NLU capabilities and NLU-first bots add rules and buttons for more accuracy. The holy grail to identify intents using bots is to combine both technologies.
Let me repeat that one more time: The holy grail in finding a customer’s intent with a bot is to combine decision trees with NLU capabilities.
Solvemate does that. And I am proud to say that our approach is unique.
Conversation Builder: We have invented a system of “dynamic decision trees infused with NLP”. In our Conversation Builder, an unlimited number of intents / solutions (typically 50-100) are defined. Then, diagnostic multiple choice questions are added to distinguish every intent from every other intent. This process takes hours, not days or weeks. There is no painful NLP training needed - it can be done in one minute per solution / intent.
Conversation Engine: Out of this knowledge, the system does information gain and contingent probability calculations to dynamically decide which diagnostic questions to ask the user. This process narrows down the number of possible intents / solutions. On average, after 3.7 questions our bot has found the correct intent. The bot shows a solution in > 95% percent of the conversations, meaning that there is minimal “sorry I don’t understand, I need to loop in an agent”-frustration. On top of that, the end user can also type their request in natural language and we use NLP to get to a solution even faster.
Conversation Trainer: We use NLP to analyze all existing interactions with the bot and cluster conversations by topic. Together with our granular dashboard, this gives the bot manager Insights on what needs to be changed in order to increase the bot’s performance.
Thus, we have closed the loop; Our Conversation Builder, the Conversation Engine and the Conversation Trainer go hand-in-hand and lead an ongoing improvements cycle.
So let me summarize: Our Contextual Conversation Engine is fast to get started, is easy to maintain, and provides a highly efficient way to find the end-users’ intent.
Automate the user’s request (with our Automation Builder)
Once the bot has understood what the user wants, the next step is to define a process to automate the request. So ultimately, Customer Service Automation is about process automation. We have built a highly flexible solution to map out those processes.
Automation sequences per intent: For each intent / solution, an Automation Sequence can be set up. Automation Sequences consist of pre-configured modules, which can for example be Forms (with File Uploads), Functions (to make API calls), Live-chat handovers, Phone handovers, or Ticket handovers.
Examples:
When the intent is to “know the shipping costs”, then a simple text answer is displayed; no Automation Sequence is set up as it’s purely informational.
When the intent is to “get the order status”, then a Form and a Function are connected to fetch the latest order status.
When the intent is to “book a consultation”, then the user can be prompted to either begin a live-chat with an agent or to directly book an appointment within the bot.
When the intent is to “give honest product feedback”, then it could be useful to add a Phone hand-over and/or a call-me-back option.
When the intent is to “make a claim for a damaged delivery”, then a Form could be prompted, asking the person to take a photograph of the package (all other data would be pre-filled due to the user being logged in). Once the form is submitted, a customer service ticket is generated and processed alongside other CS tickets within the CRM system.
Automation Sequences per channel: If Automation Sequences per intent aren’t enough, each Automation Sequence can be configured individually per Channel. So, for example, a certain intent shown on a FB Messenger Integration could have a live-chat handover, but on the Website you could offer a call handover.
Automation Sequences personalized per user: If all of the above is not enough, different user groups can be shown different Automation Sequences. A VIP client shall be shown a telephone number, while normal clients can only send an email? No problem, we’ve got you covered.
Our clients embrace the unprecedented granularity of being able to first define those processes, then to change them with a few clicks. This is true process ownership. Through that, we give customer service teams the tools they need to drive excellent customer experiences and create meaningful conversations.
Outro
In 2020, we experienced a pandemic and one of the deepest recessions of recent history. It's been a rollercoaster of a year, but I’m optimistic about the future. I hope that vaccinations will be effective and that next year will herald a return to normalcy.
As far as the future of Solvemate goes, I strongly believe 2021 will be the year of the bots, and we plan to more than double our revenue in the next twelve months.
Stay safe, stay healthy, and have a great holiday season! 🎄
Erik Pfannmöller + all the Solvemates
(A screenshot from our virtual offsite last week - yes, we have matching Solvemate sweaters…)