From Development to Production: Ask ServiceNow's AI Experts Anything

[ryanyong]

Hey everyone, I'm Ryan, an engineer on ServiceNow's Forward Deployed Engineering team. We're a team of experts and builders who've gained deep AI deployment expertise at tier 1 technology companies and research labs. We're certified experts on the ServiceNow AI Platform, with direct access to Product and the ServiceNow C-suite.

What we do
We partner directly with customers facing the most complex challenges to co-innovate agentic AI use cases and ship them to production on the ServiceNow AI Platform.

Why this thread
If you're a ServiceNow customer sitting on AI entitlement you haven't fully put to work, this is your chance to ask the Forward Deployed Engineers building these solutions in the field. Bring your hardest questions about moving from unused credits to AI running in production, and getting the most value out of your existing ServiceNow investment.

Ask us anything. We are the subject matter experts, and we will get you an answer.

We'll be answering live from August 3rd to 7th. Post your question anytime between now and 8/7, upvote the ones you want answered first.

[Dustin Gannon]

I'm going to push back on the premise a little bit.

I think our customers are sleeping on the fact that ServiceNow is an extraordinary AI transformation platform because it covers both traditional automation and transformation and agentic AI innovation and transformation.

And for existing ServiceNow customers, it's already a blessed enterprise platform inside the organization.

That matters.

You don't have to begin every AI transformation by solving all of the hardest platform questions from scratch around things like security, governance, enterprise data, workflows, and systems of record.

So my challenge to customers is...

Why aren't you bringing your biggest, hairiest, most audacious transformation goals to ServiceNow?

And please stop bringing us only CMDB-shaped and ITSM-shaped problems.

We can talk about those forever, and they're important.

But the opportunity represented by agentic AI and AI-driven business transformation is 100 times, 1,000 times, maybe a million times larger than the traditional set of CMDB and ITSM use cases we've historically associated with the platform.

I want the big problems.

Bring them to me!

[rneville]

We have just implemented ServiceNow CSM and ITSM and would like to know the best steps to start using Agentic AI within the platform. Where should we start?

[Dustin Gannon]

Step one: you're probably already using some AI if you're using AI Search.

If you're not using AI Search yet, that's one of the first things I would get going.

Step two: look for small, high-volume opportunities where something like AI Skill Kit can solve a discrete, single-turn problem.

For example, we worked with an insurance company where the first call with a customer could involve roughly five to seven minutes of preparation because the person answering the phone had to read through pages of notes before they could meaningfully engage.

We were able to create an AI skill that summarized all of that into the most valuable information.

When you're dealing with thousands or millions of calls, replacing five to seven minutes with something closer to seconds creates enormous aggregate value.

Little improvements become very big improvements at scale.

Third: recognize that not everything needs to become an AI agent—but don't shy away from agentic use cases either.

AI agents are especially useful when the branching logic of the process is sufficiently ambiguous that you need an intelligent orchestrator to reason about what should happen next.

But that doesn't mean you need to rebuild an entire business process as one giant agent.

Imagine you already have a deterministic process, but it needs a “brain” at steps two, four, and eight.

Great.

Keep the deterministic process and use AI to provide decision automation at steps two, four, and eight.

Then use tools like process mining to understand where the real time sinks are.

If you discover that steps four and eight account for 60% of the time it takes to complete the process, you don't necessarily need to transform everything.

Fix those two ugly parts.

The performance of the entire process may improve dramatically because you applied AI precisely where it created the most leverage.

So don't try to boil the ocean.

Find the painful part of the process, make that part dramatically better, measure what happened, and expand from there.

You can make a humongous impact on people's lives without rebuilding everything at once.

[ChristinaLy]

CSM Now Assist Skill  'Suggested Steps' - won't work if you have encrypted fields, the clustering fails. How can we can we get this skill to work when you have encrypted fields (short description & description). Thanks

[GreysonS]

Christina — this isn't a config issue with a workaround hiding somewhere, it's a structural conflict, so I'd redirect the effort rather than keep fighting the skill.

Suggested Steps runs on GAF, which clusters resolved cases by the semantic content of short_description. Clustering has to recognize that "can't log in" and "locked out of my account" belong together. Encryption breaks that at the math level — the platform is comparing ciphertext, and no encryption mode preserves the similarity relationships clustering depends on.

First, confirm what's actually failing. Clustering output lands in sn_gaf_record_group and sn_gaf_record_group_detail. If those are empty, activation never completed and you have a different problem. If they're populated but the groups are nonsense, that's the encryption signature.

If it's encryption, two real options:

1. Change the control, not the goal. If the concern is sensitive values appearing in free text, PII detection and anonymization protects the sensitive tokens while leaving the semantic content clustering needs. Encryption is a heavy instrument for "there might be a customer name in here" — and short_description and description are the highest-leverage fields on the platform for AI. Encrypting them opts you out of AI Search, summarization, and similarity too, not just this skill.

2. Build your own grouping layer. You can't repoint the OOTB Suggested Steps skill; those GAF implementations aren't modifiable today. But the generic GAF grouping skills are — look for "GAF CSM grouping" in sn_nowassist_skill_config. Those support clustering on custom fields, and cloning one lets you cluster on a custom table. So you could populate an unencrypted, scrubbed derived field and cluster on that. More work than out-of-the-box, but a genuine path.

If encryption on those two fields is non-negotiable for regulatory reasons, then Suggested Steps is off the table for that scope today — and that's worth raising formally, because encryption-compatible AI is a legitimate ask.