ServiceNow Now Assist rollout across the service operations desk at a global insurance group. Twenty two percent productivity uplift, credit cost discipline below budget, and a controlled three wave deployment across thirty thousand customer service agents.
A global insurer rolled out ServiceNow Now Assist to its service desk and gained 22 percent in agent productivity while holding generative AI credit costs flat through a disciplined consumption model.
Key takeaways
The insurer ran a high volume service desk with long handle times on routine cases. The goal was faster resolution without adding headcount.
ServiceNow proposed Now Assist for the platform, with generative AI summarization and resolution assistance billed on assist credits.
Measure before you commit. The team ran a scoped pilot to establish credit burn per resolved case, using the platform ServiceNow pricing structure as the commercial frame.
The pilot produced a hard number for credits per resolved case. That benchmark, not the vendor forecast, became the commitment basis.
Three controls held the burn down. Each was a configuration choice backed by the platform ServiceNow product documentation.
Now Assist pilot versus vendor forecast, illustrative
| Metric | Vendor forecast | Measured pilot | Gap |
|---|---|---|---|
| Credits per resolved case | Higher baseline | Tuned baseline | 20 to 40% lower |
| Monthly credit burn | Aggressive curve | Held to benchmark | Flat to forecast |
| Productivity gain | Estimated | 22% measured | Confirmed on covered queues |
The standard rollout advice is to enable Now Assist broadly and let adoption drive value, so buyers commit credits to the vendor growth curve. We disagree. In roughly half of the ServiceNow generative AI engagements we supported in 2024 and 2025, broad enablement drove 2 to 3 times the credit burn of a tuned configuration with no extra productivity. The buyer side move is to run a scoped pilot first, measure credits per resolved case, and commit only to that benchmark times your real case volume. Enable skills deliberately, queue by queue, so consumption tracks value rather than the forecast.
Source: Redress Compliance advisory engagement file, 2024 to 2025.
Measure credits per resolved case in a pilot, then commit to that number. The forecast is not your benchmark.
Resolution time on the covered queues fell 22 percent within the first quarter. Agent satisfaction rose as routine summarization moved off their plate.
Credit burn held flat against the pilot benchmark even as usage scaled, because the controls kept consumption tied to resolved cases.
The renewal committed credits to measured demand plus a modest margin, not the vendor growth forecast. The insurer kept the productivity gain without the open ended bill.
The pattern is repeatable. Pilot, measure, scope, then commit to evidence.
Agents needed clear guidance on when to trust the generative summary and when to override it. A short enablement program kept quality high while handle time fell.
A monthly credit review tied burn to resolved cases and flagged drift early. The platform updates are tracked through the ServiceNow newsroom.
Get the buyer side framework our advisors use on live ServiceNow engagements. The credit cost model, the rollout sequence, and the renewal levers.
Download the Renewal ToolkitThe insurer achieved a 22 percent reduction in resolution time on the covered service desk queues within the first quarter of the Now Assist rollout, driven mainly by automated summarization and resolution assistance on routine cases.
Now Assist is billed on assist credits that accrue per generative action rather than per seat. Because consumption drives the bill, measuring credit burn per resolved case is the central cost control for a rollout.
A scoped pilot produced a measured figure for credits per resolved case, which ran 20 to 40 percent below the vendor forecast. That benchmark, not the forecast, became the defensible basis for the credit commitment.
Three controls kept burn flat against the pilot benchmark even as usage scaled: tuned prompt scoping to cut tokens per action, deliberate skill limits at launch, and a queue by queue rollout targeting the highest volume routine work first.
The forecast trap is committing credits to the vendor adoption curve rather than measured demand. Broad enablement can drive two to three times the credit burn of a tuned configuration with no extra productivity, inflating the commitment.
The renewal committed credits to measured demand plus a modest margin rather than the vendor growth forecast, so the insurer retained the productivity gain without an open ended generative AI bill.
Yes. The pilot, measure, scope, and commit pattern applies to any Now Assist use case, because the underlying control is tying credit consumption to a measured unit of delivered value rather than to a forecast.
Committing to measured pilot consumption rather than the vendor forecast typically holds renewal credit spend flat while usage scales, avoiding the 20 to 40 percent overcommitment that an unmeasured forecast tends to carry.
