I’ve been exploring the impact and importance of leveraging unstructured data in my previous articles for MarTech, and for this piece, I’m digging into perhaps the biggest source of insights and risk: our email.
Much of the genAI email discussion is centered on the use of GenAI to produce better targeted, personalized mass email campaigns, with the aim of improving the effectiveness of outbound campaigns.
In this article, I’m going to dig into genAI use cases for inbound emails captured in the CRM. These messages include one-to-one messages and one-to-many reply-all emails. I am assuming here that the organization integrated its team’s individual inboxes with the CRM, as is common among today’s marketing, sales and customer teams.
Every email that goes out or comes in is tracked and stored as part of the activity record. As a result, what were unstructured, scattered email conversations are now part of that business’ central repository of data, and can be analyzed using a CRM’s embedded AI and natural language processing (NLP).
This allows us to move the discussion from tracking outbound engagement metrics, like opens and clicks, to instead analyzing the content and sentiment of the emails in the CRM, particularly as their capabilities becomes more common in leading CRMs like Hubspot.
From email campaign metrics to conversational insights
| Use Case | Example KPIs | Key questions answered |
| Analyzing traditional ‘outbound’ email metrics. | Open rate, click-through rate, conversion rate | “Did my campaign perform well?” |
| Generating Insights from genAI analysis of ‘inbound’ emails. | Sentiment scoring, intent classifications, response to content | “How do our customers feel and what do they need?” |
Dig deeper: The future of the martech stack and marketing operations is ‘unstructured’
Insights and concerns
A more complete email analysis takes you beyond the high-level “who” (email senders) and “what” (subject lines) to the underlying sentiments of the email. Let’s break this down further into the opportunities for marketing, sales and customer-facing teams, and cover new considerations around privacy and compliance.
The pros: Opportunities for customer-facing teams
Enhanced customer insights
Using NLP to analyze email interactions reveals deeper customer sentiment and pain points, leading to more personalized marketing and sales strategies.
Emails contain detailed customer questions, specific objections and information requests. By using NLP or even AI-driven analytics, marketing teams can extract patterns from emails to improve the messaging and tailor follow-up content approaches.
This approach allows us to understand more about individual customer’s preferred tone and content formats. By applying NLP techniques, organizations can analyze the emotional tone of incoming emails and classify them as positive, negative or neutral. This sentiment analysis can be combined with the intent of message sequences — such as “demo request” or “pricing question” — to help determine deal momentum or blockers.
Improved marketing-sales-customer collaboration
A centralized analysis of email activity enhances collaboration among teams, going beyond the stored history of customer interactions. Extracting insights was limited in the past due to sheer volume.
Recognizing when additional stakeholders get added into email replies could lead to more insight and automation around buying roles and decision makers. Matching this information to additional meetings tracked for RFPs, requirements and more helps drive correlation analysis to overall funnel stages, lead statuses and deal pipeline stages and improve those processes.
Data-driven decision making
Using an LLM to analyze emails potentially informs decisions, allowing teams to respond quickly to trends and customer needs.
Email sentiment analysis helps inform target accounts and contact scoring beyond simple personas. Traditional lead scoring models relied too heavily on structured data like job titles, which alone are also unstructured data.
Rather than waiting for milestones and pre-programmed actions to trigger surveys and extract sentiment, we can now monitor sentiment along the journey.
The cons: Key concerns and challenges of analyzing unstructured email
Email conversations are much more casual than other communication forms. We are likely more cognizant of what we say in a meeting that’s being recorded, than what we quickly fire back in an email. That creates some level of risk.
Privacy and consent
Teams will need to revisit customer privacy and appropriate consent measures, especially when dealing with sensitive information.
We need to revisit email footers created long before the latest genAI LLM capabilities were released. Although these were generally structured as confidentiality disclosures that protected against risks, they may not include the appropriate language to indicate that they will now be fed into an LLM model for additional predictive analysis.
Teams may need to consider new processes, like aggregating insights without connecting them to specific individuals. Similarly, another concern that may arise in typical B2B use-cases when contacts move across accounts regularly. Does an individual’s email from their prior employer still represent their views now that they’re at a new employer? Can you count on this to be their consistent position, or do you need to reset their sentiment toward you?
How does this change when teams enable the CRM connectors with ChatGPT and have broader implied use cases?
Regulatory compliance
Hubspot made news last year when it released its sensitive data and HIPAA compliance. However, because it’s likely that Hubspot customers had enabled the centralized email tracking prior to then, organizations may need to revisit their email policies and processes.
These new capabilities also illustrate the nuances of structured vs. unstructured data. Sensitive data classifications were typically structured around defined form fields that could be categorized as such. It’s not as clear how those guidelines apply if sensitive data elements are included in the body of unstructured emails.
Access controls and internal policies
All of these newly embedded capabilities may lead to organizations adjusting access controls so only authorized, trained users can view email activity. This is a significant change, as centrally captured emails were previously available with company-wide access. By definition, it’s impossible to predict what’s going to be part of an email conversation. Some organizations may choose to take a conservative approach rather than allowing the LLMs to analyze all email.
Platform considerations
Marketers need to ask their platform vendors for more help navigating these risks. New features roll out quickly.
Vendors can also help users understand what should be activated by default. For example, Hubspot’s Breeze AI Assistant is still labeled as beta. But when HubSpot released its new AI capabilities, the customer conversation data was on by default — on the broadest possible set of data for call transcripts, emails and more.
New opportunities, weighed against new concerns
Leveraging genAI capabilities to extract insights from email conversations has tremendous potential for every organization. However, the everyday prevalence of email, along with its unstructured ability to be used for almost everything in day-to-day business, carries significant risk.
Having these capabilities on by default does not mean organizations will immediately benefit. We need to structure our processes and weigh the pros and cons, all at a newly unprecedented pace.
Dig deeper: Using AI to build a DIY customer sentiment analysis solution
The post Customer sentiment — and risk — are hidden in the emails in your CRM appeared first on MarTech.