Imagine you’re in the market for some new business software. Your first port of call is an obvious but sensible choice: A vendor that provides another product your organization is already using.
But before you can move forward, you need to clarify some important details about the app’s security and compliance credentials that aren’t available on the website.
To discuss this, you’re directed to a new AI sales agent that the vendor has just started using. At first, things seem to go well: It identifies you by name and clearly recognizes some basic details about the problem you’re trying to solve.
But things quickly start to go downhill. The agent doesn’t know you’re an existing customer, meaning its suggestions and advice are… a bit off. Worse still, it insists on explaining details you already know, asking questions it should already have the answers to, and generally just making it unnecessarily difficult to get the answers you need.
Sooner or later, you’re probably going to try a different vendor…
The AI Data Problem
It’s easy to think that AI is the culprit here. In truth, the issue is often much more fundamental: Bad data.
Ultimately, the best AI agents in the world can only perform well if they’re powered by reliable, structured, and effective data. Without it, the performance of your agents will be noticeably reduced. The result? An increasing number of frustrating customer interactions like the one described above.
It’s easy to see how this happens. With so much pressure to bring AI products to market, it’s tempting to quickly spin up a new agent with your existing Salesforce data. But if the underlying data is flawed, the resulting products risk doing more harm than good.
Effective AI, therefore, starts with high-quality data. But before we can enforce data quality in Salesforce, we need to define it. In general, there are three principles for what good data should be:
- Complete: Without the full picture, AI models can easily miss vital context, like what stage the contact is at or what previous interactions they’ve had. Giving AI models access to the full range of data is the best way to avoid these issues. This may involve bringing in data that isn’t available in Salesforce by default.
- Structured: The relationship between different objects in Salesforce is also hugely vital. There’s no point having a mountain of data if your AI model can’t work out which individuals, leads, or organizations it belongs to. Therefore, data should follow consistent formats and field types, so AI can effectively parse and understand it at scale.
- Objective: Avoid conflicting information wherever possible. This tends to arise when individuals or organizations are split across multiple records – or when information from outside Salesforce is siloed from data within it.
Across the rest of this piece, we’ll set out a series of recommendations for how you can enforce these principles in Salesforce.
1. Collecting the Right Data
The first step is to consider what data you’re actually training your AI models on. There are several different options that could be relevant here – and not all of them are available in Salesforce by default:
- Interaction data: Customer interaction data (i.e. emails, calls, and meeting notes) can give AI models a wealth of useful information for sentiment analysis, churn predictions, and more.
- Pipeline history: This includes information about how/when deals progressed to the next stage (or didn’t) – and what events in Salesforce were recorded at the same time. This can give AI models rich insights for win/loss predictions and forecasting.
- Event data: Events like ‘contract renewal’, ‘product milestone’, or ‘deal closed’ are recorded in Salesforce. Like pipeline history, this provides more detail into how customers are interacting with your company or products.
- Product usage data: If you’re building your AI model on top of an existing app/product, there should be a wealth of additional data you can call on. This includes logins, usage, session frequency, and more. By default, this exists outside of Salesforce.
In all cases, the logic for recording this information is simple. The more data available to the AI – the better it will be at whatever job you’re training it to do. But as ever in Salesforce, the structure of that data is at least as important as the quantity. There’s no use training an AI model on all your historical pipeline data if it can’t distinguish your smallest customers from the enterprises.
Therefore, it’s important to ensure that all this data flows into a single, unified data model in Salesforce.
2. Define and Enforce Data Hygiene
Next, it’s helpful to define a set of hygiene standards across your CRM. The goal here is to build a foundation of structure into your CRM, through several key practices:
- Clear account hierarchies: First, it’s important to ensure clear relationships between different objects in Salesforce. Some common examples include ruling out contacts that aren’t associated with accounts (‘orphaned contacts’) and enabling contacts with multiple accounts. Collectively, these ensure contacts are always associated with accounts, while making it easier to see when different contact records belong to the same individual.
- Customer segmentation: The next step is to create clear segmentation categories for your accounts and contacts. This could include fields like industry, revenue, or organization size. You may need to define custom fields if the out-of-the-box options don’t match your sales processes. Consider making these required fields, so they’re always collected.
- Clean stage definitions: Now, consider defining the steps of your sales process in Salesforce. Common examples could include 1) Prospecting, 2) Qualification, 3) Proposal, 4) Contract, and 5) Closed (won/lost). This will allow you to set more granular rules for what information is collected at each stage (see next section).
3. Define Information Requirements and Formats
Now that we’ve built a foundation of structure in Salesforce, we can define more granular rules around what information should be stored in Salesforce and how. This helps to reduce the risk of different Salesforce users entering whatever information they consider relevant, in whichever format they decide.
Establish Clear Duplicate Rules
Duplicate contacts cause some of the most persistent and common data quality issues in Salesforce. Using duplicate rules and matching rules, you can define policies to identify these issues as information is being entered into Salesforce. You can configure these rules to block the input, notify the user, or log the duplicate.
However, to manage or merge existing duplicates at scale, you’ll need a third-party tool like Plauti.
Define Information Requirements
Next, consider defining what information is required and what isn’t. Essential details like name and contact details will generally be required fields in Salesforce already. But realistically, this is the bare minimum.
Therefore, it’s probably a good idea to define specific additional information that’s particularly relevant for your sales processes or AI models. For accounts, this might be industry or company size. For contacts, it could be the first name, email, or job title. This ensures key information is captured in a consistent and comparable way.
Define Field Structure
Next, you need to decide the format in which information should be stored. Wherever possible, you should avoid blank text fields (‘text blobs’), as they resist structure and metadata by design.
Salesforce already enforces formats for basic details like phone numbers or email addresses – but you may want to add further structured fields as well. An example could be requiring a figure in digits for ‘monthly recurring revenue’. This makes it harder for Salesforce users to circumvent required fields with dummy or placeholder text like ‘Unknown’ or ‘N/A’.
Picklists and Dependent Picklists
Picklists play a huge role in keeping your Salesforce data clean, structured, and effective by requiring users to choose from pre-defined options. This essentially avoids the issue of having multiple variations of the same basic information (i.e. ‘price’, vs. ‘price too high’, vs. ‘budget’).
You can also introduce more dynamic options via dependent picklists. These prompt more detailed follow-up questions based on the user’s previous input. For instance, answering ‘USA’ for the ‘country’ picklist would prompt the dependent picklist to offer US states.
Define and Require Closed/Lost Reasons
If deals are ultimately lost, it’s helpful to understand why. Whether it’s ‘chose a competitor’ or ‘insufficient budget’, collecting this information helps you to focus your efforts on the best leads and improve your overall sales processes.
To do this, you can simply require the Salesforce user to input a reason when marking deals as closed/lost. Again, you can use required fields, picklists, and dependent picklists to ensure this data is collected in a way that’s consistent and easily comparable.
Plauti: Your Ticket to AI-Ready Data
The advice in this blog can go a long way to building a consistent, quantitative, and structured foundation for your AI agents and products. But when it comes to guaranteeing data quality, this is really just the tip of the iceberg.
That’s why Plauti exists, offering a range of tools to help define and enforce effective data quality control at scale. Here’s what that involves:
- Plauti Deduplicate: Salesforce offers some rudimentary rules to identify duplicates, but it struggles to resolve them automatically. At scale, this creates a huge manual bottleneck. Instead, Plauti Deduplicate enables you to find duplicates (across leads, contacts, and accounts), then create custom rules to decide how they’ll be merged or resolved.
- Plauti Verify: Validate the accuracy of your Salesforce data by verifying email, phone numbers, and address fields.
- Plauti Agentforce: Extend Plauti Verify’s functionality to Agentforce and verify contact details in real time.
Together with the principles we outlined in this piece, Plauti ensures your CRM data is clean, consistent, and detailed. This gives AI tools and human agents the best possible chance of delivering on their expectations and converting your customers.
Want to dig deeper into data quality in Salesforce? Our recent e-guide offers a comprehensive look into the root causes of poor data and how to improve data quality.
Check out the e-guide to find out more.
