Updated: Nov 26, 2021
Avoiding disillusionment with AI depends as much on the managers as on the data scientists. After all, while it is the job of data scientists and engineers to build a working AI model, it is the manager's job to see it used successfully. To put this in context, here are 7 lessons that helped us in adopting AI.
The big problem our team faced was on the sales side: how to win new customers. Back in 2018, our organization decided to reposition itself and we now wanted to target a new specific audience group. Despite a small sales team, we solved it using four tools, only two of which were AI enabled.
All our information on leads and sales cycles are stored in the CRM - our base platform. Let's call it Mickey. Being a central repository of all our sales data, the first thing for us to keep in mind was that any new software we use or that we bring in should be able to feed data into Mickey, to make sure that all our data stays in one place. And that is Lesson One: organizations usually end up having silos where different teams or functions have their data in different systems. Try to bring all your data into or at least connect them onto one system.
Lesson Two: The first step in approaching AI use is to always break down the core problem that you're trying to solve into specific use cases. That means breaking that big statement - 'winning more customers' - into more specific tasks that need to be accomplished to achieve that. To do this, you have to spend some time understanding and framing the problem more clearly. It’s important because AI models are designed to perform tasks, and there may be several involved in solving a single problem, not all of which may be addressed by a single solution. This is where the bulk of miscommunication between the business and technical sides takes place.
We wanted our small sales team to spend their limited time focusing on the leads that are most likely to turn into customers. So we needed to automate as much of the first half of the sales process as possible. This would require a software that could at least send out e-mails to our anti-spam compliant prospects, follow those up with emails that are being sent periodically, and alert the team when a phone call or another channel of contact could be helpful, which could be after every five attempts, for instance. Most importantly, it would also need to update all this information on Mickey - our central CRM. But that wasn’t the end of our requirements. With so many outgoing engagements we needed a software that could also understand positive, negative or out-of-office replies to alert the team accordingly, which meant a robust NLP capability at least.
As you can see, breaking a big problem into specific tasks and analyzing them helps us understand exactly what kind of solution to look for. This is important because in a real-life scenario, we do not always have time to look at every possible solution in the market.
Among our narrowed-down list of three vendors, one was quite popular and proven. However, there was a relatively new and lesser-known solution in the market, that could also learn progressively to determine when to send a follow up email to different leads based on the highest likelihood of getting response from each of them. Let's call the software Minnie. Minnie wasn't the most popular software out there but it was the right fit for our needs. That is Lesson Three.
With the list of interested leads identified regularly and updated on Mickey, our team could start chasing the resultant sales cycles. However, with many shortlisted leads with varying degrees of genuine interest and fit, our small team could still spend their time chasing a lot of cycles that looked good on paper but were actually not. Winning a deal depends as much on the sellers' behaviour as on customer interest. It depends on their compliance with proven sales best practices and operational requirements in place as well as the custom conditions faced in a given opportunity.
What we needed now was some intelligent guidance on not only the opportunity itself but every action being taken by the rep – on opportunities, on accounts and even in entering data into Mickey. We chose to go for Goofy here. While it could handle all our required tasks, what stole the show was its ability to explain the rationale behind its predictions. It clearly showed what factors it had looked at in order to judge and predict the opportunities. Given this transparency, salespeople were able to make contextual decisions on whether the AI recommendation was right or not, and when to listen to it and when not to. That is lesson Four: As far as feasible, opt for AI solutions that are not black box, and can explain the rationale behind their insights.
With its real-time look into data completeness, Goofy also delivered the Lesson Five. Because AI is only as good as the data it analyzes, ensuring that data is up to date, relevant, and complete is crucial for any company using AI. And that can also involve setting up a process for employees that are responsible for entering or updating data. Another great thing about Minnie and Goofy was that both were simple to use, thereby not requiring a steep learning curve for the employees to start using them. This helped with adoption. Effectiveness of any AI tool in an organization is determined by – apart from how well it is being used – not only its direct impact on business results but also its collateral impact on existing systems and resources. That is Lesson Six.
And finally, not everything has to be Machine Learning, which is Lesson Seven. The fourth software in our stock, which we haven't discussed, was not. Given the specific nature of our sale, which relies on an intense consultative qualification process among other criteria, we needed state-of-the-art solutions like Minnie and Goofy to manage the business end of the sale. The fourth software's job was focussed on managing the lead generation with our existing marketing stack. For us, this was an opportunity to reduce cost and training effort.
As you can see, there were several components involved in this chain of decision making around AI enablement. In the end, the 7 principles ensured that we could engage 10x more sales cycles while having reduced the operational cost by 15%.
To learn more about successful AI journeys and its management, visit www.classesAI.com.