Service · Two weeks · Fixed scope
Find out what’s wrong with your data before your AI agent does
The Salesforce Data Health Check is a two-week audit of the data your teams rely on. You get a score, a fix list in priority order, and a 90-day plan you could start on Monday.
You know this one
Someone senior asks, in a meeting, whether the data is “good enough” for the AI project. Or the new forecast. Or the migration.
Everyone turns to the person who runs Salesforce. That person knows there are duplicates, because there are always duplicates. They know a third of the contacts have no title. They suspect the “Industry” field stopped meaning anything in 2022.
What they don’t have is a number. So the answer comes out as “it’s okay, mostly,” and the project goes ahead on a feeling.
A Salesforce data health check turns the feeling into a list
We scan the objects your go-to-market teams depend on, usually Accounts, Contacts, Leads, Opportunities and Cases, and measure the things that decide whether data can be trusted:
- Duplicates. How many records describe the same customer.
- Completeness. Whether the fields you call “key” are filled in.
- Staleness. Records nobody has touched in a long time.
- Orphans. Contacts without accounts, opportunities without owners who still work here.
- Ownership. Records owned by inactive users or a catch-all queue.
- Agent readiness. Whether an AI agent would find one record per customer, fields that say what they hold, a history of changes, and enough text to ground an answer.
Every figure links to the records behind it. Nothing is changed. The scan reads your data and writes only its own results.
What you get
| The scorecard | The fix list | The 90-day plan |
|---|---|---|
| A score per object and per check, with the records behind every number. | Every finding, ranked by how much it hurts and how hard it is to fix. | What to do in weeks 1 to 4, 5 to 8 and 9 to 12, and who should do it. |
Plus a one-hour read-out with your team, and a short note at the end: what we’d fix first, what we’d leave alone, and what’s already handled by something you own.
How the two weeks run
- Day 1: kick-off (45 minutes). You tell us which objects matter and which fields you consider key. We agree who gets access and how.
- Days 2 to 4: the scan. We run Sounding, our data health app, in a sandbox or in production. It’s read-only and runs in the background.
- Days 5 to 8: reading the results. We go through the findings, open the records behind the odd ones, and separate real problems from quirks of how your org works.
- Day 9: draft review (30 minutes). You see the draft and tell us what we’ve misread.
- Day 10: read-out (one hour). Scorecard, fix list, plan. Recorded, so the people who couldn’t make it can watch.
What we need from you
- One person who knows the org, for about three hours across the two weeks.
- A user with read access to the objects in scope. The score reflects what that user can see.
- A decision on sandbox or production. A recent full sandbox is fine. Production gives truer numbers.
What it isn’t
We’d rather say this now.
- It doesn’t fix anything. It measures and recommends. Fixing is a separate piece of work, which you can do yourselves, with our apps, or with us.
- It isn’t a security audit. It looks at data quality, not at threats, logins or sharing rules. (For who-can-do-what, see Watchbill.)
- Duplicate detection is exact matching on normalized names, domains and emails. It finds the clear cases and gives you the scale. It won’t find every fuzzy match.
- Very large objects are sampled, and the report says so wherever a figure is a sample.
What usually happens next
Some teams take the plan and run with it. Good. That’s what it’s for.
Others want help with the first fixes. The common ones: merging duplicates with an undo, loading corrected data with a preview, and tracking who changes what from here on. If an AI agent is on the roadmap, the natural follow-on is the Agentforce Readiness Sprint.
And quite often, the most useful finding is a process problem, not a data one: records arrive incomplete because a form doesn’t ask, or a required field isn’t required. The plan says so.
Questions people ask
How do you check data quality in Salesforce?
You measure a handful of things per object: duplicate records, how complete the key fields are, how stale the records are, orphans, and ownership by inactive users. Reports can do some of this by hand. A data health check runs all of it together and ranks the results so you know what to fix first.
Will the Data Health Check change any of our data?
No. The scan reads your records and writes only its own run and result records. No business record is created, edited or deleted, and you can remove the results afterwards.
Does our data leave Salesforce?
No. The scan runs inside your org and the results are stored as records in your org. We read the results with the access you give us, and we don’t copy your data out.
How do I know if my data is ready for Agentforce?
An agent needs four things from your data: one record per customer, fields that mean what their labels say, a history of changes, and enough text on a record to ground an answer. The Health Check measures each and tells you which gaps would trip an agent first.
Can we run it ourselves?
You can make a solid start. The data health scorecard walks you through a 20-minute self-assessment using reports you can build today. The Health Check goes deeper, opens the records behind each number, and gives you the plan.
How long does it take, really?
Ten working days from kick-off to read-out, provided the access is ready on day one. The scan itself runs in the background, and its length depends on how many records you have.
Find out what the number is
Thirty minutes to see if it’s a fit. Two weeks to know where you stand.