At the Salesforce DC World Tour, we saw a clear change in the go to market message, summed up by the phrase “AI plus Data plus CRM is where the magic happens”. This makes sense based on the current acceleration of the development of AI tools. Salesforce is always looking for ways to innovate the platform and enable customers to drive more value – and AI is going to make a step change.
The announcements of Einstein GPT and Flow are clear indications of the direction that Salesforce is going. So let’s unpack the AI → Data → Platform message and what it means to run your org ‘AI-first’.
AI’s ability to make sense of data and make recommendations is staggering. And it is only going to get more powerful.
What is also evolving is the experience of generating the prompts to get right answers. So expect apps that are optimized for a certain set of use cases, and these apps will teach and fine tune the AI models.
Quick AI Primer
AI is far more than ChatGPT. ChatGPT is simply an interface on top of the OpenAI davinci LLM (Large Language Model). You type a ‘prompt’ into ChatGPT which stores the context, provides a result, and uses the prompts to improve the results. This is clearly an issue when you want decisions on confidential information, as the recent Samsung debacle highlighted. OpenAI only stores data for 30 days when accessed via the API.
Any app can access the LLM models developed by OpenAI, Google, and others. The app will help formulate the prompts and use an API to get the results from one or more LLMs and present the results in context.
The LLMs can now be run on the app provider’s servers so they can be constantly optimized to provide better answers for the app’s tight use case – plus, there is no risk of confidential data being leaked. Finally, larger LLMs are being made available which improves the quality of the results.
The results from AI are only as good as the data and the prompts that you give it. Therefore the value of clean and current data is critical.
This may seem obvious, but with AI it has become even more important. Without AI, the data was presented for executives to make decisions. They could probably sense-check the data, spotting and highlighting any anomalies, so poor decisions could be avoided. AI cannot sense check – it takes the data and drives actions, often without any human intervention.
The end user is largely uncontrollable, even with lots of validation rules. If a user is confronted by hundreds of fields on a page, some of which are picklist fields with multiple items, users will take the path of least resistance. They hit enter and see which fields are mandatory and enter values to satisfy the validation rules. This generates poor data.
The data in a field in Salesforce could be populated in a number of ways: a user entering on a screen, a Flow, a formula, Apex, Mulesoft integration, or from an external system. This will be used in a number of different places, so you need to be confident in the source and target of the data to make sure that it is clean.
A scary and common example of this in Salesforce, is changing the use, or currency, of the Amount field on Opportunities. Salesforce and Reports & Dashboards were designed when the field was used to capture a single product value in US Dollars.
But now, multiple product types could be included (e.g. SaaS and professional services), and the values could be made up of USD, EUR, and GBP, plus there is an extra field denoting the currency.
If the use of this field is changed, nothing will break, Dashboards, Tableau, EinsteinGPT, and automations will all work – but the results will be wrong, even if they are only 10-15% off.
What Are the Implications of ‘AI-First’?
Without solid foundations then your AI strategy is built on sand. To build a strong AI base, these are the things to consider…
“By far, the greatest danger of Artificial Intelligence is that people conclude too early that they understand it.”– Eliezer Yudkowsky, American artificial intelligence researcher and writer on decision theory and ethics.
Org analysis will help you understand how fields are populated, whether it is a user entering on a screen, a Flow, a formula, Apex, Mulesoft integration, or from an external system, and where it is being used.
The data could be used by Einstein GPT, in dashboards, reports, Tableau, Flows, formulas, Apex or external systems. Start with the target and work back to the source of the data to understand where it comes from.
A simple question is “Are all the fields in this dashboard the correct ones, and are they being populated correctly?”.
You need thorough and sophisticated org dependency analysis. This needs to cover all the key metadata items for all fields. This is more than the “Where is this used?” button in Salesforce or relying on the Dependency API which is still in beta (sadly, this API has had no investment in nearly 3 years and there is none planned for FY24).
Another task you can do is simplifying the pages with the critical source data fields. This is done by removing fields or using dynamic forms, and also consider adding help text so that users understand what to enter. This will improve the quality of data, reduce user frustration, and save the users time. The org analysis will tell you where the fields and picklists are used and how populated they are by record type.
Once you are confident that you understand how the data is populated, write some documentation so that if these fields are ever changed then the implications are understood.
After optimizing your platform, you can focus on the next level: the current data. Based on what you discovered in the org analysis, you can understand how much data clean up is required before letting AI loose on it.
There are some great data validation and clean-up apps for the Salesforce ecosystem, as well as external services that can clean up or validate your data. At the same time, you can start putting in place compliance rules (such as GDPR).
Bear in mind that approximately 20% of your contact database churns each year. People move jobs and they change personal emails – you should feel good about deleting dead data.
If data is being fed from external systems or forms on websites, you need to address these. You need to be ruthless in cleaning up the data and then eliminating the sources that are polluting your pool of data. Crack down on users that are clearly adding poor data and celebrate those who are doing a good job. This will build a culture of good data entry.
Once you have the foundations in place and are sure your data is good, you can focus on AI. What do you want AI to support you on? How does it apply to your industry and your use cases? Think about what guardrails and checks you want to put in place.
These may be quite stringent to start with, but as your confidence with the results improves you can relax the oversight. Make sure to bear in mind how fast the AI models are evolving – this could be changing the results. It might be necessary to keep the controls in place for longer than you expect.
Tools to Support Your Org Analysis
The scale of Salesforce orgs and the pace that they change has highlighted the importance of tools to automate the analysis of orgs and support a more rigorous development cycle. That tooling is called a Change Intelligence Platform.
Any change, no matter how small, can significantly impact systems downtime, company reputation, and regulatory compliance. A Change Intelligence Platform gives you back control.
Invisory, an industry analyst, wrote a report evaluating the top Change Intelligence Platforms. A complimentary copy of the report is available here.
The new notion of “AI plus Data plus CRM is where the magic happens” is good in principle, but you need to make sure that your platform is ready. To run your org ‘AI-first’, make sure you consider your CRM and your data before even thinking about using AI. This will give you a strong base to move forward.