How AI works in Salesforce: Realistic application examples

Artificial intelligence (AI) is omnipresent these days and is attracting a great deal of attention worldwide. However, despite the widespread hype, for many it remains an abstract concept that lacks clear and specific use cases. Decision makers are under pressure to implement AI, but often lack precise information on capabilities, goals and suitable tools. This article is intended to provide practical support and help them to successfully use the numerous AI functions of Salesforce.

There are two types of AI tools in Salesforce: predictive and generative AI.

Predictive AI solutions are advanced software solutions that use artificial intelligence to make predictions about future events or trends. They do this by analyzing large amounts of data to identify patterns and relationships. The process begins by collecting data in Salesforce or other systems. Mathematical and statistical analyses are then used to identify patterns and relationships in this data. A data model is created based on these findings. When new data enters the system or existing data is changed, the tool uses this model to make predictions. For example, predictive AI can predict which customers are likely to churn or which products will sell well in the future. This allows companies to take measures at an early stage, such as personalized offers or improved customer service.

Generative AI tools use language models based on previous communication with customers and a knowledge database to automatically generate texts. For example, Salesforce can automatically generate answers to customer questions via sales and service channels such as chat and email. This speeds up communication with customers and makes your team more efficient.

The AI tools in Salesforce are geared towards specific business use cases. Customers need to provide the necessary data, inputs and basic configurations to use the tools in a meaningful way. These tools can be used by administrators or consultants, even without deep knowledge of data science or computer science. Some of the tools are even free or included in the Salesforce Cloud licenses up to a certain usage limit.

We will be happy to advise you on your specific application and show you the right options.

The introduction of AI systems in companies follows a clear pattern that comprises various phases:

• Target definition: Clearly defined goals are the cornerstone of success. What do you want to achieve with AI? Which business problems should be solved or processes optimized?

• Data strategy: Identify the relevant data sources and ensure that you have sufficient data volumes and quality to train and validate the AI models.

• Selection of the AI solution: Choose the right AI solution that meets your requirements and goals. Take into account the type of data you have and the specific use cases you want to address.

• Data preparation: Cleanse, transform and integrate your data to ensure it is suitable for AI analysis.

• Model development and training: Develop and train AI models that are tailored to your specific use cases.

• Model evaluation and optimization: Test and evaluate the performance of the models and optimize them if necessary.

• Implementation and integration: Integrate the AI models into your existing systems and processes.

• Monitoring and maintenance: Continuously monitor the performance of AI solutions and adjust them if necessary.

These are the requirements for your data for a successful implementation

Below, we'll take an in-depth look at the specific requirements and best practices for data preparation in Salesforce to ensure your AI implementation is built on a solid foundation.

The scope of the available data plays a major role.

AI models work better the more data is available. This applies to both predictive and generative tools. Note that most Salesforce AI solutions have minimum requirements for the number of data records. Check your data volume before deciding to implement a particular Salesforce AI tool. Estimate your expected future data volume in different objects in Salesforce to see which AI solutions are a good choice for you. The minimum requirements vary depending on the AI tool and usage. We have therefore added the respective requirements to the application examples below.

The quality of your data is also crucial for the use of AI tools

The quality of your data also plays a major role. Don't be alarmed if not all fields are always filled in or if your entries are incomplete in some cases. Individual and relatively random data quality problems in a large data set will not lead to distortions in your data. However, if there are systemic (large-scale, repeated) inconsistencies with the data quality, you should clean these up first before training your model on "bad" data. You should ensure that data quality remains high in the future to avoid new problems. This can be achieved by introducing standard procedures such as validation rules, default values and regular checks as well as cleansing or archiving data.

An application example of how you can optimize your sales and lead management processes with the help of AI.

The initial situation for sales

The sales team is overloaded with capacity due to the high number of leads and customers with sales opportunities. It is difficult to prioritize the incoming leads due to a lack of forecasts. Nevertheless, you want the team to focus on leads with a high probability of purchase.

Use of AI: Sales Cloud Einstein Opportunity Scoring and Einstein Lead Scoring

These two forecasting tools are similar: they predict the likelihood of an opportunity being successfully closed or a lead being converted. They also give you important insights into the opportunity or lead, such as which attributes of the data set contributed to the score. A score is calculated for the potential customer's record, which you can use to filter and sort so that the team can work on high-probability, high-priority customers.

Data requirements

For opportunities, there should be at least 200 won and 200 lost data records in the last two years (with a lifetime of at least two days). For optimal results, it is recommended to consider at least 500 closed (won and lost) sales opportunities. For leads, more than 1,000 lead records should have been created, of which at least 120 have been converted in the last six months.

The initial situation in service

Your company offers support via chat, messaging or email and receives numerous inquiries. You want to ensure that all customer inquiries are answered quickly and reliably, regardless of the time, date and capacity of your teams. Your teams also need a lot of time to answer individual customer inquiries. This manual effort should be minimized.

Use of AI: Einstein Service Replies & Einstein Reply Recommendations

Generative AI tools increase the efficiency of your service team by automatically responding to customer questions in chat or by email. They also provide service staff with suggested response messages that can be sent to customers after review and customization. This saves time and resources and enables standardized responses and processes.

Data requirements

Your organization must have set up a chat or messaging service and have processed at least 1,000 chat cases. These generative AI tools work best when a knowledge base is set up and articles are attached to the cases. This allows the generative AI to learn not only from the chat responses, but also from the knowledge articles on similar cases.

Initial situation in marketing

Your company generates many leads through various marketing campaigns and wants to ensure that the most promising leads are prioritized. The aim is to increase the conversion rate by addressing the leads with the highest probability of purchase first.

Use of AI: Einstein Lead Scoring & Einstein Campaign Insights

These AI tools help your marketing team to increase efficiency by automatically scoring leads and providing valuable insights into campaign performance. Leads are ranked according to their likelihood to buy and data-based recommendations are made to optimize future campaigns.

Data requirements

To use the Einstein Lead Scoring and Einstein Campaign Insights AI tools in the Marketing Cloud, your company must have an extensive database of leads and their interactions. The Marketing Cloud should be set up and used regularly, and at least 1,000 leads should already have been generated and tracked. A well-maintained CRM database is also required in order to be able to make accurate predictions.

The Einstein Prediction Builder can be used in the various Salesforce clouds and thus offers functions for different departments and teams. We have listed the possible application examples for you:

Use cases for the Einstein Prediction Builder:

1. Contract extension: Your company has fixed-term contracts with customers. Your teams should know how likely it is that a customer will not renew their contract and leave (churn prediction). Your teams should also know when the best time is to take action to retain the customer.

2. System maintenance: You sell products (e.g. systems) to customers and receive regular updates on the status of the systems via IoT. You want to predict when a system could fail so that you can carry out planned maintenance in good time.

3. Loyalty programs: You have started a loyalty program for your customers and want to predict the lifetime value of a customer. This helps you to categorize customers into loyalty levels and offer them corresponding benefits.

4. SLA compliance: Your company has strict service level agreements (SLAs) with customers. You want to predict whether a new customer case could lead to a breach of the SLA so that you can take action in good time.

5. Sales planning: You want to predict the closing date or the value of an opportunity in order to use this data in your sales or production planning.

How the Einstein Prediction Builder works

The Einstein Prediction Builder is a general prediction tool that uses historical data to predict the probability of future events as described above. Salesforce provides an active prediction model free of charge, making this tool an ideal, cost-effective first step towards AI.

Data requirements

All input data for a model must be in the same Salesforce object. If relevant data is located in other objects, they must be integrated into the predicted object. Example: Defects of machines sold must be aggregated in the asset object (e.g. "Number of defects in the last 3 months").

• For binary predictions (e.g. will the customer churn? yes/no), you need at least 400 sample data records, of which at least 100 yes and 100 no examples.

• For numerical predictions (e.g. chance value), you also need at least 400 sample data records. In our experience, at least 1,000 data records are required for a strong model.

• These data requirements apply to each data segment, not to the entire object.

The integration of artificial intelligence in Salesforce offers companies the opportunity to optimize their business processes and make data-driven decisions. Predictive and generative AI tools enable more efficient management of sales opportunities and customer inquiries as well as personalized customer interactions.

Tailor-made solutions can be developed by preparing Salesforce data correctly and selecting suitable AI tools. Salesfive supports you as a competent partner. Our experienced consultants, organized in special Competence Centers, know the AI functions of the respective clouds and offer comprehensive know-how in the field of business intelligence and AI.

Rely on Salesfive to successfully launch your AI integration and ensure long-term business success.