Artificial Intelligence

Setting a British telecom up for AI adoption with 91.36% success rate

  • Industry Telecommunications
  • Year 2020
  • Services Artificial Intelligence
  • Hypothesis AI can improve sales rate
  • AI implementation success probability 91.36%

Client

4Com offers leading business phone products, mostly in the United Kingdom. The company’s core product HiHi2 is an innovative desk phone with a tablet and business phone system. They approached us with the task to validate whether AI can improve the overall sales rate by identifying high quality leads more likely to convert during the sales process. They wanted to find out if the available data can be analyzed deeper to draw conclusions and used more efficiently to help telemarketers close more deals.

Introduction

About project

The Client presumed AI could help create a more predictable and scalable business model and wanted to prove this before going all in with a company-wide adoption.

Our tasks

task icon
Validating the hypothesis for a business case

The Client presumed there are dependencies between specific leads and their conversion rate. Validating this hypothesis would later help telemarketers focus only on the leads that have high chances of closing.

task icon
Building a Proof of Concept

During the AI Sprint, our data scientists build a set of models that either proves that a business case can be solved with AI or that it’s not possible with the available data.

Challenges

01

1. Low telemarketing efficiency

The Client did not have a lead scoring system helping identify the leads with high chances to convert. Thus, telemarketers had to talk to leads that were unlikely to become clients.

2. No single data funnel in place

There was no single data funnel in place and leads had to be processed from three different funnels thus making it difficult to track relations during the process.

3. A need for more efficient use of leftover data

Telemarketers kept a lot of data about their leads in their databases. This could be used more efficiently during the sales process and help increase lead conversion rate. There was also a lot of useless information that had to be filtered.

4. High maintenance cost and overhead

The leads were managed manually, which created a dependency on one key person in charge of this process. With manual work dominating the process of managing leads and data collection, the Client faced increasing costs of keeping the system in place.

5. Checking if outsourcing works remotely

During the duration of the AI Sprint, slightly more than 3 weeks, the Client wanted to test out if the team of our developers can work efficiently with their in-house team remotely before engaging in a long-term cooperation.

Solutions

02

1. Performing exploratory data analysis

The Client provided us with ready to test data about call history. After analyzing the records, our data scientists decided to drop certain rows as their quality was low. This process helped our team train the model in a more efficient way and, as a result, create an efficient and robust AI model. 

 

2. Creating the first version of the predictive model

During the AI Sprint, our data scientists created the first model with the data the Client supplied us with. The model would help validate the hypothesis that AI can be used to overcome the Client’s business challenge.

3. Conducting the AI Sprint with the Client’s team

We met the Client’s team for a 2-day workshop to set the goals and the hypotheses for further validation. We discussed the available data and set up success criteria for the project. During the next 3 consecutive weeks, we built the first model that validated the hypothesis and confirmed over 91% probability of success for AI adoption.

 

Technology we used

Google Cloud Function
NumPy
Python Pandas
Python
scikit-learn

Project results

03
Positive AI validation

During the AI Sprint, we have validated the hypothesis saying that telemarketers’ efficiency can be increased by assigning the right leads and leftover data in a more efficient way.

Identifying high success rate of AI adoption

The certainty for success of AI adoption was rated at over 91% and the Client decided to proceed with further model optimization and AI adoption.

The Client decided to continue cooperation

After the AI Sprint and presentation of our findings, the Client decided to continue working with our data science team on the company-wide adoption of AI. Apart from continuing work in the AI area, we also started a web development project.

Conclusions

We validated the hypothesis that finding potential leads interested in booking a meeting with a telemarketer faster and more efficiently, was possible with AI. With the estimated probability for AI to succeed at over 91%, 4com decided to continue with analyzing and cleaning data as well as scaling the project finding. Our team paired with the developers from 4com again to continue working on the company-wide AI adoption.

Do you like our work?

Let’s talk

about your project!

Download now!

Enter your personal and working information.

AI in e-commerce - Greg Gwozdz

Fill out the form below to watch the recording.

Fill out the form below to watch the recording and download the infopack.

Fill out the form below to watch the recording.

Fill out the form below to watch the recording.

Download now!

Enter your personal and working information.

Download now!

Enter your personal and working information.