We are developing a tool based on Artificial Intelligence to qualify all types of sales leads. There will be numerous sales leads and incoming enquiries. Attending all of them is a huge waste of time. You can’t increase the number of sales members based on the number of sales leads. You ideally want to categorize all of them so that you can focus. This will help the sales team to focus on sales leads which have a higher chance of conversion.
This article will explain our tool under final stages of development for improving sales leads. This is based on a machine learning algorithm which learns on its own after an initial period of supervised learning. This would help the customer to identify quality leads using their own company descriptions.
If you look at your current manual process, you may notice that this is already happening. This is based on the experience of the sales person. This is mostly an intuitive process and not scalable. Instead we can automate the whole process using Artificial Intelligence and Data analytics.
A typical result of the AI tool will be as below:
The AI tool will analyze a combination of parameters such as the product/service sought, country/city of origin, type of the customer etc.
The machine learning algorithm combined with data analytics works in the following manner.
The first step in the process is called data collection. This dataset will consist of qualified or disqualified companies in terms of quality leads. The qualified list will consist of your regular clients. You may already have done a business with them at least once. And the disqualified list will consist of companies which do not really relate to the profile of your potential customer. This part is easy. Now there will be lot more which are not in both of these categories. Our AI software will analyze these sales leads based on lot more parameters.
There are several ways to collect this kind of data. Some opt for a manual technique while others prefer automated machine aided analysis. Manual collection of data will result in more accurate and complete data in contrary to automatic analysis. In an automatic process there is a chance that data will not be accurate and incomplete. However, automatic analysis does offer the ability of time conservation. During the initial period of learning, the data collection will be mostly manual. This helps the AI tool to learn the process.
Now that the dataset is taken care of, we will have to move towards cleaning the data. This process is generally called data processing or data cleaning. Cleaning the data essentially means getting rid of irrelevant information. This can be achieved by writing a natural language process script. The idea is to make a proper dataset in a useful form. Our machine learning algorithm now uses this data to qualify the sales leads.
In our case, one of the cleaning techniques we applied was using regular expressions to get rid of non-alphabetical characters so that our model would only read words from the description.
Another problem was reducing the repetition of words in the data. To tackle this, we used a stemmer which would group the repeating words and simplify it to a single word.
The final problem was to get rid of the stop words such as for, I, it, to etc. We used Natural Language Toolkit to solve this problem.
Finally, after processing the data we must transform our dataset into a machine-readable format. We used Bag of Words (BoW) approach to turn the description into vectors. Machines can only read data as 1s or 0s. Therefore, to train a machine learning model, all the data is converted to binary format. This essentially aids the machine in reading data and learning from it.
All the afore mentioned techniques fall under the topic of natural language processing. All of this was carried out using a python library called NLTK.
Now that our data is ready to be used in our machine learning algorithm, we split the data into two different sets. We slicked it into 70% training data and 30% testing data and ran in through a famous machine learning algorithm Random Forest. And after some fine tuning we reached a high accuracy on the test dataset which means the output would be helpful to our sales team.
Random Forests are a supervised machine learning algorithm. We use Random Forest algorithm for classification. However, the algorithm can be used for regression problems too. The algorithm stems from decision trees. However, decision trees tend to overfit. To overcome this, Random Forests train multiple trees and take the majority decision to give an output.
Various other algorithms can be used to train the same model. To investigate the best model, multiple algorithms should be used, and best performing models should be used. Decision Trees, Support Vector Machines (SVM), Liner classification and K-Nearest Neighbor are some of the examples of good and efficient algorithms.
Of course there are pros & cons for every tool. As AI is becoming an integral part of life, you can take a calculated decision based on your applications.
Pros:
Cons:
Vacker Global provides various tools based on Artificial Intelligence, Machine Learning, Data analytics etc. mainly for B2B segment.