Blog

A Practical Approach to Generative AI

Author: Janet Elliott

There’s been so much hype about AI that it’s leading to fatigue and confusion — and the majority of content out there doesn’t make it easy to understand the value it can realistically deliver for an organization.

If you're struggling to cut through the noise, you’re not alone.

Generative AI takes on challenges and provides automation in a way that’s completely different to previous generations of AI. Our team here at Kicksaw has found that the possibilities are so endless that unless you narrow things down, defining organizational value can seem just as abstract — which is where we come in.

This blog will help you zero in on a practical approach to generative AI use for your organization, with information about:

The types of AI commonly used (predictive and generative)
What’s going on with AI development
Salesforce AI tools and use cases — with two HLS ‘day in the life’ scenarios
Value-focused adoption
Next steps and additional resources

Predictive vs. Generative AI

Both predictive and generative AI can improve processes and operations in ways that positively impact a business — but generative AI is a big leap forward in technology (as demonstrated by its constant presence in the media). It’s capable of assisting — and sometimes reinventing — the ways that processes are handled in the workplace.

The AI that’s become standard over the past decade (traditional, predictive, ‘deep-learning’, or ‘big-data’ AI) works by using machine-learning to analyze past patterns and data to predict future outcomes. For example, looking at appointment volume to provide recommendations about resource allocation.

Predictive learning use cases and tooling:

It’s great for analytics, forecasting, and planning
Salesforce’s Einstein Prediction Builder and Einstein Next Best Action are examples of tools that use this type of AI
It’s the AI most people are familiar with and runs mostly in the background

The AI you’re hearing about ad nauseam (generative AI) uses large language models (LLMs) and natural language processing (NLP) to generalize patterns within data, in a process called vectorization — and use those vectors to produce new content and ideas. For example, generating care summaries for referrals based on requested parameters.

Generative AI use cases and tooling:

It’s learning, human language, and generative-focused
It relies on prompts entered by the user
It can complete tasks using the pattern and relationship data it learns from
Salesforce’s Einstein Copilot is an example of generative AI; it utilizes common LLMs, including GPT by OpenAI, Claude by Anthropic, Gemini by Google

Generative AI development is still in a very ‘wild west’ state. There’s been such a massive push to create ‘faster, better’ versions of both the generative AI technology and the hardware used to support it that it's been referred to as an "AI arms race."

Although generative AI and its capabilities will continue to improve — to get a better handle on what generative AI can do and ways it can add value — narrow your scope:

Start with generative AI solution vendors
Look at vendor-specific tools
And assess what those tools can do for individual users in a typical day

Salesforce AI Solutions

Kicksaw leverages Salesforce solutions to tackle client challenges, so we’ll be providing Salesforce-specific AI information and generative AI use cases to illustrate value.

Salesforce is positioning their generative AI as “human at the helm” technology and AI + Data + CRM — referencing the human interaction element of its functionality and how it works with other solution components and data.

Einstein 1 Studio (Copilot Builder, Prompt Builder, and Model Builder), along with Einstein Copilot, are the foundation of their current generative AI offering
Their solutions have a Trust layer, which helps with security and governance
They work well with a number of LLMs

Let’s look at two of these tools for clearer examples of how generative AI can add value — with solution information below and use case information in the next section of the blog.

Salesforce Prompt Builder

If you’ve ever played around with tools such as ChatGPT, you know a well-configured prompt is the key to effective use — because LLMs rely on these prompts to set parameters for task completion.

A prompt is a series of statements which help provide context to the model you wish to interact with
For example, if you ask an LLM to explain what the stock market is, you’ll get vastly different results if you make the distinction between your audience: 5th graders vs industry experts

Salesforce’s Prompt Builder lets users set these types of distinctions in the form of prompt templates. Admins build, refine and test prompts to accomplish a particular generative AI goal to help create the template.

Templates are grounded in your CRM’s data
Prompt template formats available at the time this article was published include: Field Generation, Flex, Record Summary, and Sales Emails.
In Salesforce AI terms, “grounding” means that the model relies on not only the LLM, but factual data and context in order to generate output specific to your unique ask
They can be triggered and used in multiple contexts including Flow, Apex, and Einstein Copilot Actions

Think of prompt templates as prompt “cheat sheets” for your users, so they don’t have to start prompts from a blank slate. End users can leverage the templates to generate responses, but also have the ability to further refine the responses with additional prompt inputs.

Salesforce Einstein Copilot

Einstein Copilot is an assistant sidebar for your Salesforce org that was announced as GA (generally available) April 25th, 2024. Using natural language prompts it can be used to complete simple tasks, such as find a particular record, summarize it, and find related records.

You can also take actions, including updating or creating a record, or sending an email — as well as use customer actions to create your own action — all from your sidebar.

It uses Copilot Actions to build a plan
It comes with Standard Actions, such as querying a record by name and summarizing a record
You can also build Custom Actions to extend the functionality
Einstein Copilot Custom Actions can leverage a prompt template, invoke a flow and leverage Apex
All of the metadata around the actions (including the all important Description field!) help Einstein Copilot inform it’s actions

Einstein Copilot (as with other AI assistants) retains the context from the prior response so you only need to provide follow-up prompts.

Salesforce AI Use Cases

The two solutions outlined above work together in a complimentary way, combining automation, generative functionality, and improved understanding of next steps. While they’re useful for ‘one-off’ use cases, their real value is demonstrated by transforming the way a user goes about their day.

Because the overall value for any generative AI solution will vary depending on a number of factors, including the solution, industry, organization, and role — to better illustrate the benefits of AI — we’ll be reviewing how Salesforce’s Prompt Builder and Einstein Copilot can be used for for an HLS Care Coordinator and a luxury Med Spa Concierge.

These examples are imagined based on actual ‘day in the life’ scenarios from Kicksaw HLS projects.

Care Coordinator

Responsibilities include:

Taking in referrals
Coordinating care
Making referrals
Communication

Processes include:

Reviewing a referral, searching for appropriate providers based on specialty and location, communicating with the providers, assessing the patient’s care plan, reviewing clinical notes.

Processes using Prompt Builder and Einstein Copilot:

A coordinator can start their day by prompting Einstein Copilot for all new referrals received and ask it to find the right therapists.

The logic that Einstein Copilot performs on the backend would be based on a custom Einstein Copilot action that triggers the flow with the therapist matching logic
Matches are provided to the Care Coordinator in the AI Assistant sidebar

A prompt template can be used to draft a referral email for the therapist that is grounded in the patient’s data and includes a summary of the patient’s case.

The coordinator can then review the draft and press send, with the email being logged automatically

For patients whose care is complete, prompt templates can be used to automate the review of post evaluation write ups and call out action items for additional services needed, for the coordinator to act on.

This is done based on custom actions defined in Einstein Copilot and because Einstein Copilot retains the context of prior requests
The coordinator can act on those action items — and if they’re common, Einstein Copilot Custom Actions could be built to make the task more efficient

If a patient needs a third-party coordinator for specialty care, the AI Assistant can easily identify the appropriate service, and use a prompt template to draft a referral with the relevant patient information and a summary of the request.

Value and benefits:

Using Prompt Builder and Einstein Copilot, the care coordinator has the tools that are crafted to support their daily flow of work. This is a snapshot of one aspect of their role, but you can see how these tools can be used to achieve things like:

Faster time-to-referral
Greater referral volume processing
Better provider and patient relationship management
Increased revenue potential through follow-up service identification

Identifying goals (like the benefits listed above) with your SI (systems integration) partner — and working backwards to figure out if and how generative AI can add value — is how we recommend approaching generative AI solution adoption.

Luxury Med Spa Concierge

Responsibilities include:

Proactive, high-touch, resort-like customer service
Detailed and personalized client knowledge
Scheduling of treatments that include individual appointments, outings, and activities
Coordination of services with clients’ travel

Clients fly in and have treatments scheduled, so the concierge needs to be aware of everything happening with their clients to ensure satisfaction — and proactively adapt plans accordingly.

Processes include:

Accessing various areas of Salesforce to identify where the client is, review whether readiness steps have been completed for the appointment, check flight information separately to coordinate airport pickup, and reschedule treatments if flights are late.

Processes using Prompt Builder and Einstein Copilot:

The concierge can open Einstein Copilot’s AI assistant and use a prompt template to summarize the client’s treatment plan.

They can identify if the client is ready for their appointment
If a pre-appointment checklist hasn’t been completed, an Einstein Custom action can assess the remaining checklist items and offer follow-up actions

At any time, the concierge can prompt the AI Assistant to identify where the client is, because their treatment plan and check-in status are tracked in the system.

If the client is flying in, a custom Einstein Copilot Action can make a callout to a flight status API to determine if the flight is on time. If the flight is delayed, it can provide the car service and reservation information that is stored in Salesforce.

Einstein Copilot can use a prompt template to draft a message to the car service and let them know the updated flight details
The concierge can review the communication and send it out via email or SMS
An Einstein Copilot custom action could advise if any of the client’s treatments will need to be rescheduled and suggest the best available new times

A proactive notification based on a prompt template can be sent to the client, who is still in the air, confirming that the car service has been rescheduled and their treatment plan has been updated accordingly.

Additionally, in the spirit of great customer service, the med spa concierge can assess a sentiment score that has been calculated and proactively provide communication or offers to increase customer satisfaction and loyalty.

Value and benefits:

This role and type of organization offer a very clear picture of how generative AI can eliminate manual processes and lower the chance of human error — with a focus on improving the client experience. We can see in this scenario that AI is:

Speeding up processes and reducing delays
Automating communication and coordination
Keeping the concierge more informed and in a better position to strengthen client relationships

Tips for Value-Focused AI Adoption

Start with your data and data governance or no matter what generative AI you end up using, you won’t end up with the ROI you’re looking for.

From there:

Determine your pain points, challenges, and goals
Look at solution vendors and vendor-specific tools
Assess what those tools can do for individual users at your organization in a typical day

If you’re still stuck on how to get started, try analyzing how different users are interacting with your Salesforce org. or solution:

Are they clicking on a record and then having to review all related records to get up to speed on an Opportunity or a Case?
Are they manually typing out emails, trying to figure out what information to include and what tone to use?
Are your sales people not sure what Opportunity to focus on next or developing a plan to close the opportunity?
Are your customer service agents typing up case summaries or manually creating Knowledge Articles?
Are there multi-step processes that could leverage a generative AI tool to make the interaction easier and more intuitive for an end user?

Remember that the ultimate goal with generative AI, just like any other solution, is better adoption of your CRM, increasing user efficiency, and moving your business forward.

Next Steps

If you’re looking for your own practical approach to AI — we can get you AI ready and help you determine its organizational value, as well as where it fits into your solution roadmap. Contact us.

Additional resources:

‍

Back to Knowledge Center