Immediate Action Drill: Four Data Fixes to Make Before Data 360

I wrote the first version of this in December 2024. Two years of Data 360 projects later, the advice held up, but I can be a lot more specific about why. This is the rewrite.

An immediate action drill is what you do without thinking when something goes wrong: steps rehearsed so many times they’re automatic. Data prep for Data 360 (formerly Data Cloud) and Agentforce deserves the same treatment. Not a transformation program. A short list of checks you run first, because every one of them fails loudly later if you skip it.

Here’s the drill, and what breaks without each step.

1. Fix the formats your match rules will compare

Identity resolution compares field values. It doesn’t understand them. “(555) 555-0142” and “5555550142” are the same phone to you. To a match rule, they only line up once both are normalized. Same story for full names crammed into one field, dates in three formats, and emails with stray spaces.

Start with the fields a match rule will touch: first name, last name, email, phone, address. For each one, ask:

• Is it populated? Blank values never match. If 40% of your phone fields are empty, a phone rule only works on the other 60%.
• Is it one format? Pick one, and either fix it at the source or plan the transform before ingestion.
• Is it split the way the data model expects? A full name in one field has to become first and last before a name rule can do anything with it.

You can watch this happen in the data harmonization visualizer: turn off “split names” or “standardize dates” and see what stops working downstream.

2. Find the key that exists in more than one system

The most reliable match in Data 360 isn’t fuzzy at all. It’s a shared identifier: a loyalty number or customer ID that shows up in the CRM, the POS, and the web store. Data 360 stores those as a Party Identification, and an exact match on one is about as close to certain as matching gets.

Inventory the systems that hold customer data (ERP, commerce, loyalty, service desk, whatever applies) and write down, for each:

• What identifier it uses for a person
• Whether that identifier appears in any other system
• How the data gets out: a connector, an API, a file drop, or zero copy

If no key crosses systems, that’s the finding. Your matching will lean on email and phone, which puts even more weight on step 1. To see the difference a shared key makes, add a loyalty ID rule in the identity resolution simulator and compare it with the name and address rules.

3. Clean up contact points before you need to send to them

A segment counts people. An activation needs somewhere to send. The gap between the two is behind a lot of “why is my audience smaller than my segment” questions.

Check:

• Do emails and phone numbers live somewhere you can tell which one is current, or are they scattered across records?
• Are phone numbers in E.164 format with a country code? Marketing Cloud activations need them that way.
• Is consent captured, and does it live somewhere Data 360 can read it?

The activation visualizer walks one segment of 40 people down to the ones who actually get the message. Every person who drops out drops out for a reason you could have fixed in the source system.

4. Simplify before you integrate

Technical debt compounds when you layer AI on top of it. Every unused field still shows up when you map data. Every overlapping automation is one more thing writing a value you didn’t expect. Before you connect anything:

• Deactivate fields nobody uses
• Consolidate old Workflows and Process Builders into Flows
• Standardize picklist values that mean the same thing, since they feed segmentation
• Make page layouts support good data entry, because the cheapest bad data to fix is the data nobody entered

Then bring the team along. Data quality starts decaying the week a cleanup project ends unless the people entering data know why it matters. Show them their own data in a demo. It lands differently than a slide.

The point

None of this requires Data 360 to start. Every step is something you can do this week, in the org you already have, and each one removes a failure you’d otherwise find in production.

The foundation you build today is what AI will amplify tomorrow.

When you’re ready for the next step, planning for identity resolution picks up where this leaves off. And once the data is in shape, here’s how I decide which AI use cases are worth building.