What Leaves the Instance
Datum CMDB Pro reads the shape of the CMDB, not what's in it. What stays in ServiceNow, what leaves, where it goes, and the compliance answers.
The short answer for security, compliance and plant-systems teams — including manufacturers with OT and regulated infrastructure in the CMDB.
Three things to know. That’s the whole story.
1
The records stay.
Server names, IP addresses, serial numbers, owners, locations and people never leave ServiceNow. Datum counts records. It does not read them.
2
Only the structure leaves.
Table and field definitions, rules, counts and health scores go to an AWS environment set up for that one customer. They are not used to train AI.
3
Need more control? There are options.
For customers with stricter requirements, running the analysis side in their own AWS account is available on request.
Datum never changes the CMDB. It writes results to its own tables and creates tasks for the customer’s team to pick up.
What stays, what leaves, and where it goes.
Customer’s ServiceNow instance
Stays here
- Every CI record: names, IPs, serial numbers, locations, owners
- Users, groups and any personal data
- Credentials and secrets
- Logs and attachments
- Script code (unless an admin opts in)
Leaves, once per run, over HTTPS
Only the structure
- Table and field definitions
- Identification and reconciliation rules
- Record counts per class
- CMDB Health scores
- Names and settings of business rules and scripts
Comes back: the plan and its tasks, written to Datum’s own tables.
Dedicated AWS environment
Where it goes
- Set up for that one customer, in AWS US East
- Analysed by Claude through Amazon Bedrock
- Not used to train AI models; not shared with the model provider (AWS policy)
- Kept 90 days, then deleted. Sooner on request
One server. What Datum sees of it.
In the customer’s CMDB · a server record
| Name | PLANT-07-MES-DB01 | STAYS |
|---|---|---|
| IP address | 10.42.7.18 | STAYS |
| Serial number | SN-4KX82201 | STAYS |
| Owner | Plant Systems team | STAYS |
| Location | Greenville, Line 3 | STAYS |
What Datum receives · the shape of the table
| Table | Linux Server |
|---|---|
| Fields defined | name, ip_address, serial_number, owned_by, location |
| Record count | 1,906 |
| Matching rule | Name only |
Enough to see that 1,906 servers are matched on name alone, so duplicates are waiting to happen. Not enough to know which servers they are.
Sample data for illustration. The server record shown is invented.
Nothing happens unless the customer starts it.
01
Off until an admin turns it on
Nothing leaves until an admin enters the endpoint and key.
02
Runs only on request
Someone with the role clicks Run, or makes an authenticated API call. No background collection.
03
One customer per environment
Each customer gets a dedicated AWS environment. One customer’s data never sits with another’s.
04
Not used to train AI
AWS states that Bedrock inputs are not used to train models and are not shared with the model provider.
05
Deleted after 90 days
Kept 90 days, then deleted. Sooner on request.
06
Read-only on the CMDB
Datum never writes to the CMDB. Results land in its own tables. Every payload carries a record of what was collected.
For the customer’s compliance team: the questions they ask, answered.
Is it self-hosted?
By default, no. The app runs in the customer’s instance; the analysis runs in an AWS environment EFS sets up for that customer alone.
Does plant or asset detail leave?
No. CI tables are only counted. No hostnames, IPs, serial numbers or locations.
Does any personal data leave?
No.
Can anything sensitive ride along?
Custom table, field and rule names do leave. If a table is named after a program or a client, that name is visible to the analysis. Script code leaves only if an admin turns that option on.
Is the data used to train AI?
No. AWS states that Bedrock inputs are not used to train models and are not shared with the model provider.
Does Datum change the CMDB?
No. It writes results to its own tables and creates tasks for people to pick up.
Export-controlled or defence work?
The same facts apply: no record data leaves, and hosting is commercial AWS in US East. Whether structure-level information is in scope for a program is the customer’s decision.
Need more control? Run it in the customer’s AWS account.
Customer-hosted option · available on request
The analysis side runs inside the customer’s own AWS account.
- Available on request: set up with the customer’s team
- The customer’s account, the customer’s controls and audit trail
- Datum Pro was built to support this from the start
- Launch through AWS Marketplace is on the roadmap
Sensitive options are off by default
Script code is only sent if an admin switches that on.
Stricter requirements?
We work through them with the customer’s security team, and share the field-level list of what is collected.
What a health score would never tell you.
Sample data. Each of these becomes an ordered task in ServiceNow, not just a finding.
01
The shadow CMDB
The team’s real work lives in custom tables, so scores on the standard tables mislead.
02
The rule pile-up
A dozen overlapping business rules on servers, left behind by past projects.
03
Two sources fighting
Two feeds write to servers at the same priority. Fix this first, or the cleanup won’t stick.
04
Servers on the generic class
Thousands of servers never classified. The plan sets prerequisites, then works in batches.
05
Nobody is keeping score
Health jobs were switched off. First action: turn them on and take a baseline.
06
Stale records, sorted
Records untouched in a year are split into “validate or retire” and “fix”.
07
Duplicates waiting to happen
Servers matched on name only, and custom tables with no matching rule.
Next step: see it on a real CMDB. A no-cost pilot, run together.
01
Start with one CMDB
One where ITOM, Service Mapping or AI is licensed but held up by the data.
02
A read-only assessment
Results the same day. No CMDB writes. Only the structure described here leaves.
03
A sequenced plan
Findings turned into ordered ServiceNow tasks the customer’s team can run.
Demo and figures use sample data and estimates. ServiceNow is a trademark of ServiceNow, Inc.
Let's talk about what you're building.
Our team brings over two decades of experience to every engagement. Tell us about your project and we'll show you what's possible.
Let's talk about what you're building.
Our team brings over two decades of experience to every engagement. Tell us about your project and we'll show you what's possible.