What are the benefits of leveraging AI in your customer success function?

These benefits are significant now, and may evolve over time – things are changing fast with AI but even as the landscape changes, the need for strategic leadership and informed decision-making will remain.

• Increased margins – leading to better profitability. AI can streamline your customer success organization by automating routine tasks and processes and providing data-driven insights to contribute to a more efficient and profitable operation.
• The potential to increase customer retention – smart use of AI and automation can help CS better service customers and increase retention, enhancing the profitability of the CS department.
• Reducing the time spent on mundane tasks – AI can help automate repetitive and time-consuming tasks, allowing CSMs to focus on more strategic, high-value activities. For example, automating CS follow-ups can lead to substantial savings of 30 minutes a day for CSMs. That creates significant capacity over the course of the year.

What are some of the specific AI tools that are currently in the market for CS use cases?

AI-powered chatbots and virtual assistants
Tools	• BotPress • VoiceFlow • Gleen • Intercom  • Zendesk  • Maven AGI • BerryApp
What They Do	Serving a variety of use cases– no code tools like BotPress and VoiceFlow offer a lot of configuration and automation that may not be necessary for everyone. Legacy tools like Zendesk and Intercom offer robust ticketing and chatbot features with some new AI capabilities. Gleen and Maven are newer AI-first companies that have unique capabilities. Gleen promises to be hallucination-free, comprehensive knowledge ingestion (pdfs, URLs, etc.) and can read and generate images. Maven provides ticket summaries, chatbots, and insights. Berry focuses on onboarding and training while also providing a chatbot. Find a balance between customization and efficiency and consider your business needs and resources before deciding on the best approach.

Personalized customer interactions
Tools	• Gainsight PX • WalkMe • Pendo • Matik • CastApp
What They Do	In-app interactions – are where you engage with your customers directly within your application. This could include guides, tutorials, or other forms of user assistance. For this, consider point solutions like Gainsight PX, WalkMe, or Pendo. These tools are designed specifically for platform analytics and in-app guidance, making them an ideal choice for situations where you want to provide in-app tutorials and guides. Out-of-app interactions – refer to personalized customer engagement that happens outside of your application, such as through email. For these types of interactions, tools like Matik or CastApp can be useful. These applications are designed for Quarterly Business Review style interactions. CastApp provides more interactive Avatar-led onboarding, personalized DBRs (Digital Business Reviews), training, etc. Matik focuses on creating personalized decks and PDFs

AI-generated self-service tools/documentation
Tools	• Intercom • Zendesk • AriGlad • MavenAGI
What They Do	Offer chatbot functionality and ticketing systems – Intercom and Zendesk allow you to handle customer queries and help in managing and tracking customer interactions. Create custom materials from historical data – New AI-first tools like Ariglad can use past Zendesk/Intercom tickets to create self-service documentation. Smaller companies that may not have the time or resources to create extensive enablement material may find this feature to be particularly beneficial.

Category	Tools
Chatbots	• BotPress • VoiceFlow • Gleen • Intercom • Zendesk • Maven AGI
CSP	• Gainsight CX • ClientSuccess • ChurnZero • Catalyst/Totango • Vitally.io • Custify
Enablement Creation	• Ariglad • Everflow
Digital CS	• ThenaAI • onereach.ai • Everflow • WalkMe • Pendo • Gainsight PX • Funnelstory • Magnify • CastApp • Matik
Predictive	• Obviously.ai • reef.ai • Parative • DISQO • Pendo • Gainsight PX • Funnelstory • Magnify
Personalization	• Funnelstory • Magnify • DISQO
DBR	• CastApp • Matik
Support Ticketing	• Intercom • Zendesk • Salesforce • Hubspot • Maven AGI
Call Summarization	• Gong • Fathom • Chorus • UpdateAI

Example use case: Task Creation and Execution

How can AI be applied to help customer success teams create tasks and execute them?

AI can extract actionable tasks from customer calls – using call recording summaries or manually entered notes from customer success managers. A customer might report a platform bug, provide feedback, and share a positive story, all in one session. Each of those categories necessitates its own response — AI can recognize categories that can help generate corresponding tasks, such as filing support tickets or presenting issues to the product team. 

Creating account headlines – by having access to call summaries, platform usage, support inquiries, emails, etc. we can use AI to generate summaries for each account for leadership. These headlines can be updated monthly and pushed to slack channels to further streamline efficiency so leadership isn’t pinging ICs for updates and ICs aren’t scrambling to provide them. 

Generating Talking Points- We can use AI to pull relevant talking points like usage snapshot (from platform data), challenges (from call summaries or support tickets), etc. into slide decks.

ML can be used to detect trends in platform usage to gauge health – it can identify behaviors that lead to expansion in revenue and behaviors that lead to contraction in revenue. While there are tools out there that can help in this area, it is recommended that someone with a data science or stats background review the data input to ensure you’re coming to the right conclusions. Once healthy or risky behavioral patterns are identified, they can be addressed with manual tasks by the customer success manager or by automated digital paths (also known as Digital Customer Success).

This helps create a Digital Customer Success org, and builds a one-to-many approachto CS – with ML, you can use your historic data to identify risky behaviors or capitalize on healthy ones. AI can help categorize customers, put them on specific drip campaigns, and send tailored follow-up emails based on customer responses to assessments. This approach relies heavily on the quality of the data and the assumptions made from it.

Digital Customer Success Case Study

Say that your data indicates if a customer who’s within one week of signing up goes three days in a row without logging back into the tool, then they have a 99% chance of never logging back in. With that knowledge, if they go one day without logging in, you might want to contact them to make sure they get back on the platform. If they go two days without logging in, maybe you consider some other type of outreach to re-engage them. Once you establish how you want to re-engage. You can automate this outreach as soon as the customer begins to go down an unhealthy pipeline.

Once you have it set up, you can monitor user activity and suggest improvements – AI connected to your data will help you analyze your product or playbook based on customer activity within the tool. This can help get in front of potential issues before they become problematic.

What tools do you need in place to help you create and execute tasks for CS? What does a task creation workflow look like?

Example Task Creation Workflow:

Tool	Purpose		Example in Workflow
Relational database and automation	Form the foundation of a lightweight customer success platform.		A common, effective combination of no-code tools is Airtable and Make. Airtable provides the relational database and interface layers of software creation, while Make offers a robust automation layer. Together, they can handle a significant amount of work.
Call Recording Software (optional)	Record calls and generate summaries, which can then be sent back to the CRM for analysis.		Gong or Update AI are examples of call recording software. Update AI is recommended for its better pricing options and overall suitability for most Customer Success teams.
GPT 4 (or equivalent) API	Used to analyze call summaries for action items and export them in JSON format.		Could be Open AI or Perplexity. Rather than just having a single call note record, the JSON data format allows the creation of individual tasks with automatic due dates applied to them

Once call summaries are available, GPT 4 can identify action items and export those as JSON – this is then passed into Airtable as an individual task. From there each task is categorized using AI. Based on the categorization we can use Make to build out logic and route those tasks (with additional CSM added context) into whatever existing tools your support, product, and marketing teams use.

Why is it important to implement an AI strategy across all of your data?

Data feeds AI – in which you can take data from multiple sources, aggregate that data, and then summarize that data is extremely powerful 

Creating accurate and robust account headlines – is possible when all this data is summarized. For instance, you might find that while a customer has regular positive conversations, they are not utilizing some of the sophisticated features of your tool. This could be indicated by the basic questions they ask in emails and support tickets, suggesting a lack of training on the tool. It’s critical (now more than ever) to factor all data points to assess next steps.

Track specific platform usage and trigger different automations – when customers aren’t using the tool as much as you would hope. This doesn’t necessarily have to be AI-powered automations, but simply automations that happen across multiple datasets, not just one singular dataset.

How can you aggregate data for customer success across your organization?

Centralize data in a CRM – tools like HubSpot and Salesforce are capable of aggregating vast amounts of data, functioning as data warehouses. While nearly every tool on the market uses AI in one form or another, the data is often sent back to the CRM, which serves as a central repository that can then be used to extract relevant information for specific teams. For CS this can include platform usage, email interactions, support tickets, call summaries, and more. 

Incorporating a trust metric into your AI Summaries or Health Scores – can provide a more accurate picture of customer interactions. For example, the sentiment data from ten customer calls is likely to be more reliable than that from just two calls. Similarly, a customer’s usage stats will differ depending on their stage in the customer journey. 

Trigger alerts for certain topics that may indicate a potential issue – such as a company acquisition or product cancellation. This can trigger automations to notify leadership, trigger playbook tasks, and help businesses respond more quickly to potential problems.

Incorporate all collected data – and be wary of ignoring one set of data in favor of another. For example, while call data can be used to create sentiment, other data such as platform usage, support tickets filed, and email interactions with the CSM and sales rep should also be considered. 

Overall

How do you handle data privacy and security when using AI to analyze customer interactions?

Proper handling of data privacy and security is up to interpretation – there is not a one-size-fits-all situation. Your security team should define your organization’s specific needs and regulations to tailor your approach.

APIs play a significant role in data privacy and security – and how the APIs interact with data is constantly changing. For instance, OpenAI at one point used the data that passed through their API to train their model. However, they have since changed this practice. Now, when using their API, the data is not used to train their models, hence making it compliant for most orgs. 

Many tools today utilize the OpenAI API in some form – and its widespread use has led to a consensus that it should be fine for most organizations from a security standpoint. However, it’s always essential to ensure that your use of such tools aligns with your organization’s specific data privacy and security guidelines.

Data privacy concerns are different in different industries – certain industries, such as financial or medical institutions may have more stringent data privacy and security concerns. However, it’s becoming increasingly clear that the use of AI and related tools is crucial. The key is to adapt and evolve with the technology or risk becoming obsolete.

How should you think about splitting up customer interactions between Human-only vs. AI-assisted vs. AI-only?

Incorporate some human input where it makes sense – for example, customers may provide information that triggers the AI to generate a task, but the AI might lack the full context behind that specific task as of today. Therefore, allowing the human to provide additional details, perhaps a sentence or two, before executing the task can ensure higher accuracy and less back and forth with cross-functional partners. This ensures that the task is routed to the appropriate parties.

Consider scenarios where automations don’t work exactly as expected – using the same example, when wording differs slightly or if context is missing. In such cases, a support or product employee might receive a ticket without context, leading to frustration and inefficiency as they reach out for clarification via Slack. This undermines the efficiency gains expected from automation.

What are the most important things to get right?

Data centralization – allows you to have all your data in one place, making it easier to analyze and draw insights from. Once your data is centralized, you can move from a reactive to an informed state. 

Data hygiene – if you feed your AI system with poor-quality data, it will not be able to produce quality outputs. The principle of “garbage in, garbage out” applies here. 

Use AI to create insights and power automations – based on your centralized, clean data. This allows you to move from an informed state to a predictive state. 

Consider fringe use cases – and consider the situations where AI may not apply and the level of friction it may cause. For instance, if an AI system works 95% of the time, you need to think about what happens the other 5% of the time. 

What are common pitfalls?

Expecting tools to solve organizational problems – remember that tools are only as effective as the people using them and the processes they support. 

Neglecting change management – implementing new tools often requires significant changes in how people work. Organizations must use proper change management best practices to ensure a smooth transition and to maximize the effectiveness of the new tools.

Over-reliance on a single tool – no singular tool is going to be able to solve every problem. By its nature, implementing AI requires flexibility and movement. 

Poor data hygiene – faulty underlying data that moves through the automation pipeline unchecked can have large consequences. This will magnify problems rather than help solve them.