Now Assist · Consumption Forecasting

A 6-Step Framework for Now Assist Consumption Forecasting

A structured approach to planning AI consumption with confidence — from entitlement baseline to growth modeling

By Neha Agrawal · Product Manager – AI Platform · February 2026

Why Forecasting Matters Now

AI adoption in the enterprise is accelerating faster than most organizations anticipated. As teams move from pilots to production — and from one use case to many — the ability to plan and forecast AI consumption confidently has become one of the most valuable skills a partner or platform team can offer.

Across customer deployments, four themes consistently surface when consumption doesn't go as planned:

| 🔁 Agentic Workflow Misconfiguration

A single AI Agent with a bad trigger configuration can consume hundreds of assists in minutes. | 🧪 Testing Without Consumption Monitoring

Teams build and iterate in sub-production without realizing every test burns real assists from the same pool as production. |
| 🗄️ Cloned Production Data + Scheduled Jobs

Cloning an instance for testing can expose an entire incident backlog to a scheduled GenAI job. | ✅ Unmonitored Proof-of-Concept

Someone activates a skill to explore it — and nobody's watching the consumption dashboard. |

The common thread: a forecasting gap. None of these situations are the result of bad intent — they happen when teams are moving fast and AI capabilities are expanding quickly. The 6-step framework in this article helps close that gap, giving customers, partners, and platform teams a structured approach to plan consumption with confidence.

Who This Is For

Customers, partners, implementation consultants, architects, and platform owners working with Now Assist. Whether scoping an initial deployment or expanding existing AI coverage, this framework applies at every stage.

The 6-Step Forecasting Framework

Six steps, six words. Each builds on the previous to produce a consumption forecast tied to business value — something you can bring to stakeholders with confidence.

VALUE → BASELINE → MAPPING → ESTIMATE → BUFFER → SCALE

Click any step below to expand details.

VALUE

The GenAI Value Equation
|

Start every forecasting engagement here — not with assist counts, but with the value question. This framework, adapted from MIT Sloan professor Rama Ramakrishnan, anchors the conversation on what AI actually delivers.

AI delivers value when:

Time & effort doing the task without AI > time to set up + time using AI + time to review outputs
AND the quality level meets business requirements
AND risk tolerance aligns with the use case

The practical output is an efficiency ratio — a way to reframe assist consumption as time recovered:

Efficiency Ratio	What It Means	Recommendation
10:1 +	Strong ROI — AI dramatically reduces effort	Prioritize for early adoption
5:1 – 10:1	Solid ROI — clear time savings, reasonable review	Good candidate for rollout
2:1 – 5:1	Moderate ROI — review effort is more significant	Optimize prompts/quality first

Example — Incident Summarization (8,000 incidents/month)
- Without AI: 8,000 × 5 min manual effort = 667 hours/month
- With AI: 95% quick review (30 sec) + 5% light edit (2–3 min) = ~76 hours/month
- Time recovered: ~590 hours/month | Efficiency ratio: 8:1

Adapted from: Ramakrishnan, R. "A Practical Guide to Gaining Value From LLMs." MIT Sloan Management Review, Winter 2025.

BASELINE

Establish Your Entitlement Baseline
|

Know your starting point before building any forecast. Three questions to answer:

How many assists did the customer purchase?
How many have they already consumed?
When is the contract anniversary date?

Where to look	What it shows
Subscription Management → Account Level Entitlements → Now Assist Usage	Total purchased vs. allocated assists across all instances (account level)
AI Agent Analytics in AI Agent Studio → Assist Consumption tab	Instance-level agentic workflow consumption, near real-time

January 2025 Update — Burn-Down Model
- Now Assist measurement transitioned from a rolling 365-day look-back to a 365-day burn-down model tied to the contract anniversary date
- Unused assists are forfeited at reset — no rollover
- The contract anniversary date is now the anchor for all consumption planning

MAPPING

Map Current & Planned AI Use Cases
|

Inventory every AI skill and agent active in production, plus planned use cases for upcoming rollouts, and map each to its consumption category.

Category	Example Skills	Assists
GenAI Skills	Summarization (incident, case, etc.) / Resolution Note Generation	1
Knowledge Article Generation	10
Conversational AI (NAVA)	Now Assist Panel Conversation	5
Virtual Agent Topic	10
Voice Agent Conversation	50
AI Agents	Small workflow (<4 actions)	25
Medium workflow (5–8 actions)	50
Large workflow (9–20 actions)	150
Builder Tools	Experience (UI) Generation	1,000
App / Playbook Generation	2,500

December 2025 Update — Skills On By Default
- Select Now Assist skills now activate automatically when installing or upgrading plugins
- Check Now Assist Admin Console after any upgrade to review which skills were auto-activated
- Verify scheduled job filter conditions before upgrading sub-production environments with cloned data

Reference the published Now Assist FAQs for the latest assists-per-action values across all skill types.

ESTIMATE

Calculate Baseline Consumption
|

Apply the formula to each use case separately, then sum:

Monthly Consumption = Volume × Adoption Rate × Assists per Action

Three things to keep in mind:

Volume: Pull from ServiceNow Performance Analytics — don't guess. Estimate projected use cases from source systems or stakeholder input.
Adoption: Model as a range, not a single number. Adoption depends on enablement, not just availability.
Approach: Calculate GenAI Skills and AI Agents separately — they have fundamentally different consumption patterns.

Example — ITSM Implementation (Month 1 Baseline)
- GenAI Skills: 5,000 incidents × 70% × 1 = 3,500 | 1,000 problems × 50% × 1 = 500 | 500 KB articles × 30% × 10 = 1,500 → Subtotal: 5,500
- AI Agents: 5,000 incidents × 20% routed × 25 assists → Subtotal: 25,000
- Total baseline: 30,500 assists/month — AI Agents represent 82% of consumption despite handling only 20% of volume

This pattern — where AI Agents drive the majority of consumption even at lower volume — is consistent across implementations. The single most important question to ask: "What percentage of tickets do you want AI to handle autonomously versus assist humans?"

BUFFER

Budget for Development & Testing
|

This is the most commonly missed factor in forecasting. Sub-production environments draw from the same account-level assist pool as production.

Common sources of sub-production consumption:

Prompt Refinement Iterations
Agentic Evaluations
UAT Cycles with Realistic Data
Developer Experimentation
Scheduled Job Testing

The right buffer depends on implementation complexity. Use your own trending data as the best guide — measure during the initial pilot, compare sub-prod vs. production ratios, and refine continuously.

Sub-Production Planning Checklist
- Review which skills will auto-activate before upgrading
- Verify filter conditions exclude historical/cloned data (e.g., Created >= activation date)
- Consider reducing scheduled job frequency in sub-prod
- Monitor sys_gen_ai_usage_log daily during initial testing phases
- Check Now Assist Admin Console after any upgrade for auto-activated skills

SCALE

Model Growth Over Time
|

Don't use a flat multiplier. Consumption grows as adoption increases and use cases expand. Think in adoption phases — and build three scenarios with stakeholders:

Phase	Characteristics	Consumption Pattern
1 · Focused Rollout	Limited user groups, controlled use cases, heavy testing	May spike unpredictably; sub-production often dominates
2 · Planned Expansion	Broader rollout, established guardrails, AI Agents added incrementally	Steady, measurable growth aligned to change management
3 · Full Scale	Org-wide access, multiple AI Agents, AI embedded in daily workflows	High sustained consumption with seasonal variations

Plan Your Checkpoints — Align to Contract Anniversary Date
- 50% consumed: Check in — review adoption trajectory and upcoming use cases
- 75% consumed: Plan ahead — assess whether current pace aligns with entitlement
- 90% consumed: Begin expansion conversations with your account team

Framework in Action: Sentiment Analysis Scenario

The following example applies the full 6-step framework to a single use case. Numbers below are illustrative — your actual results will depend on your specific configuration, entitlement, and rollout approach. In practice, you would calculate each skill separately and combine for a total forecast.

Large Healthcare System

Applying the 6-step framework to Now Assist Sentiment Analysis for ITSM

Monthly incidents
| |
300

Service desk agents
| |
1.5M

Annual assist entitlement
|

VALUE

Sentiment Analysis Value Equation

Without Now Assist: Manual sentiment review + escalation reviews + callbacks from missed frustration = 1,025 hours/month
With Now Assist (optimized): 4 hrs one-time setup + automatic analysis + 5-sec indicator review per incident = 35 hours/month

990 hours recovered/month · 29:1 efficiency ratio · Quality and risk conditions met
|
|
2
|
BASELINE

Entitlement Context

Total annual entitlement: 1,500,000 assists
Contract anniversary: July 1, 2026
8 months in — current consumption: 620,000 assists (41%)
Remaining: 880,000 assists with 4 months until reset
Current run rate projects to ~930K by anniversary — well within entitlement. Adding Sentiment Analysis at 25K/month fits comfortably.

|
|
3
|
MAPPING

Use Case Identification

Skill: Sentiment Analysis (ITSM) — 1 assist per analysis, triggered by a scheduled job (batch)
Default filter scope: all active incidents (includes the full backlog)
Recommended filter scope: updated in last 24 hours — same analytical value, predictable consumption

Configuration opportunity: adjust filter scope from "all active" to "updated last 24 hrs"
|
|
4
|
ESTIMATE

Consumption Calculation

Optimized: 25,000 incidents × 100% adoption × 1 assist, filtered to last 24 hours, once daily = 25,000 assists/month

Configuration	Filter	Schedule	% of Entitlement
Default (OOTB)	All active incidents	Every 15 min	Optimize for best results
Optimized ✓	Last 24 hrs	Once daily	20%
Balanced	Last 12 hrs	Twice daily	40%

|
|
5
|
BUFFER

Sub-Production Allocation

Production baseline: 25,000 assists/month
Sub-production buffer (10%): +2,500 assists
Sentiment Analysis is a scheduled job with no prompt refinement — the buffer accounts for sub-prod instances processing cloned data
Tip: Configure sub-prod filters to a narrow date range (e.g., last 7 days) to minimize sub-prod consumption while still validating functionality

Adjusted monthly baseline: 27,500 assists
|
|
6
|
SCALE

Growth Projections

Growth comes from incident volume increases, expanding to CSM cases, or building custom sentiment skills for other record types.

Scenario	Growth	Assists/Year	% of Entitlement
Steady State	0% MoM	~330K	22%
Focused Rollout	10% MoM	~465K	31%
Planned Expansion	20% MoM	~800K	53%
Full Scale + More Skills	40% MoM	~2.0M	Plan for expansion

Summary: 29:1 efficiency ratio · 25K monthly production · 27.5K with buffer · 300K annual (steady state) · 20% of 1.5M entitlement
|

Key Takeaways

| The biggest risk isn't overconsumption — it's underconsumption Under the burn-down model, unused assists are forfeited at the contract anniversary. A well-structured forecast helps customers use what they've purchased and realize the value they've invested in. | | The framework's value isn't the forecast — it's the knowledge Adoption will evolve. New use cases will emerge. The six steps give you a diagnostic tool to understand consumption patterns and course-correct with confidence. | | Your judgment is the added value The platform gives customers powerful tools. What they need from you is judgment — which use cases to prioritize, how to configure them, and when to scale. That's the trusted advisor role. |

Resources

Views are my own, and do not represent my team, employer, partners, or customers.