Deep Learning for Lead Nurturing
Lead Nurturing is the process of keeping your leads engaged, so that they eventually buy. It's important to send your leads something that they will open and read. Despite the clear benefits of lead nurturing, a study by MarketingSherpa indicates that only 36% of marketers actively nurture their sales leads (Source: 2013 Email Marketing Benchmark Report).
To build an AI system that uses deep learning one must understand what leads to an effective lead nurturing email.
One such thing is targeted content. So given a lead and their info the task the AI system has to figure out which content is most targeted. First of all, you need to understand each of your unique buyer personas. Prospects receive an average of 10 touches from the time they enter the top of the funnel until they’re a closed-won customer
However this is different for each product and service and that's where deep learning can come in. It can help understand the ideal touch points for your product and customer.
Most successful content is the one that answers common questions that help the buyer in the lead conversion process. So what are those common questions and concerns? Deep learning can help mine key questions and concerns from text you give related to your product or service. For instance you can train a system to identify which sentences are questions and which ones are not. Using amazon mechanical Turk you can train a system to extract key questions and concerns from any text. All you have to do is feed content relevant to your product and you will get back a list of common questions and concerns. Using text similarity or auto encoders you can train a system to group questions that are similar into one group. You do this because you want to group together different variations of the same question into one.
Based on the recent browsing behavior of your lead you can then figure out what questions and concerns are most common. For instance if someone just visited your page on on premise deployments and then downloaded your product brochure it suggests that your lead is most interested in knowing about how your product will work on premise.
Deep learning can help here as well. It can take as input browsing behavior and rank the top questions and concerns. The interesting thing to note here is that you must use a RNN or a LSTM neural network because you want to take the sequence of actions performed by the customer and then predict the top questions and concerns they are going to have next.
So how does one train a system like this?
First a deep learning system should continue to scan the content your marketer creates. This could be whitepapers, blog posts, pages, etc. to make it even simpler you can do this by doing a crawl of your own website. You can also feed public information to this system in the form of urls. What you are doing is effectively giving the AI system a repository of content that can be sent to the user.
Next this system should take as input the users activity.
If we follow that analogy, the contact properties would be the collective knowledge and memory that is stored in your brain. Contact properties store information about people like:
• website activity
• email engagement
• social media activity
• form submissions
• conversion information
• data from other integrated software.
One simple way to build a AI system is use training data of past behaviors. The deep learning system can now predict what the next best action that customer will take. It will learn from the behavior of your own customers and then suggest the next best action. For e.g. If a lead visited your website and clicked on case studies but the leads that convert tend to read a popular case study then the next best thing to do is to send that case study as an email to your lead.
Unlike other behavior targeting systems which are rule based, what SublimeAIs engine can do is that it can use all signals in your database. It's highly unlikely that a behavioral targeting system will use data like
• Annual Revenue - annual company revenue.
• Associated Deals - the number of deals associated with this contact.
• Became a Customer Date - the date that a contact's lifecycle stage changed to Customer.
• Became a Lead Date - the date that a contact's lifecycle stage changed to Lead.
• Became a Marketing Qualified Lead Date - the date that a contact's lifecycle stage changed to Marketing Qualified Lead.
• Became a Sales Qualified Lead Date - the date that a contact's lifecycle stage changed to Sales Qualified Lead.
• Became a Subscriber Date - the date that a contact's lifecycle stage changed to Subscriber.
• Became an Evangelist Date - the date that a contact's lifecycle stage changed to Evangelist.
• Became an Opportunity Date -the date that a contact's lifecycle stage changed to Opportunity. Became an Other Lifecycle Date - the date that a contact's lifecycle stage changed to Other. City - a contact's city of residence.
• Close Date - the date that a contact became a Customer.
• Company Name - the name of the contact's company.
• Country - the contact's country of residence. This might be set via import, form, or integration.
• Create Date - the date that a contact entered the system.
• Days To Close - the days that elapsed between when a contact was created and when they closed as a customer.
• Email - a contact's email address.
• Fax Number - the contact's primary fax number.
• First Name - the contact's first name.
• Industry - the company's industry.
• Job Title - the contact's job title.
• Last Modified Date - the most recent date that any property on this contact was modified.
• Last Name - the contact's last name.
• Lifecycle Stage - a property used to indicate at what point the contact is within the marketing/sales process. It can be set through imports, forms, workflows, or manually on a per contact basis. For more information, check out this article.
• Message - a default property to be used for any message or comments a contact may want to leave on a form.
• Mobile Phone Number - the contact's mobile phone number.
• Number of Employees - the number of company employees.
• Owner Assigned Date - the most recent date that a Owner was assigned to a contact.
• Persona - the contact's persona.
• Phone Number - the contact's primary phone number.
• Postal Code - the contact's zip code.
• Salutation - the title used to address the contact.
• State/Region - the contact's state of residence.
• Street Address - the contact's street address, including apartment or unit #.
• Website URL - the contact's company website.
• The beauty is that the system can practically use all data you have about your contact and then use that to predict their next action.
So to summarize a deep learning system can be used in two ways for lead nurturing.
• Given relevant content about product or service it can predict the top questions and concerns users have. This can help your marketing team create content to address those concerns.
• Given the behavior of your users and their contact info a system can predict the next action successful customers take which can then be used to send automated emails, text messages or calls.
The power of deep learning here is that truly can use all the data you capture about yours users unlike traditional marketing automation solutions that rely on standard set of inputs.