The GenAI Problem Nobody Advertises
ServiceNow's Now Assist features - text summarisation, case deflection, incident context, change risk prediction - are genuinely useful capabilities. In the right environment, they reduce handle time, improve first-contact resolution, and surface information that agents previously had to hunt for manually.
The problem is that "the right environment" has prerequisites that most organisations haven't met. And ServiceNow's marketing materials are not particularly forthcoming about what happens when you enable Now Assist on a platform with poor data quality.
What happens is this: the AI works exactly as designed, producing outputs that faithfully reflect the data available to it. If your incident records have sparse descriptions, your change records lack meaningful impact fields, and your knowledge base articles are outdated, Now Assist will summarise, suggest, and surface all of that - accurately, at speed, and to no useful effect.
The Four Data Readiness Dimensions
We use a four-dimension readiness framework when assessing organisations for Now Assist and GenAI deployments:
1. Incident and case data quality
Now Assist's incident summarisation and agent assist features depend on structured, consistent incident data. Specifically: short descriptions that accurately reflect the issue, work notes that document resolution steps, close codes that are meaningful rather than defaulted, and category/subcategory classifications that reflect actual workload patterns.
Most organisations have inconsistent short descriptions, work notes written for the resolver rather than for future analysis, and close codes that were defaulted to "Resolved" by a service desk team that found the dropdown more trouble than it was worth.
Before enabling AI summarisation, run a data quality assessment on your last 12 months of incident data. The percentage of incidents with meaningful short descriptions and documented resolution steps will tell you more about your AI readiness than any maturity model.
2. Knowledge base completeness and currency
AI-assisted deflection and knowledge suggestions are among the highest-value Now Assist features - and the most dependent on knowledge base quality. A knowledge base with articles that were accurate two ServiceNow releases ago, written by people who left the organisation, and never reviewed after publication will actively harm deflection rates when AI surfaces them to end users.
Knowledge base readiness requires: article coverage mapped to your actual incident volume by category, currency review (when was this last validated?), format standardisation (AI models perform better on consistently structured articles), and gap analysis (which high-volume incident categories have no knowledge coverage?).
3. CMDB and service graph completeness
Now Assist features that involve service context - impact analysis, change risk assessment, outage notifications - depend on a service graph that accurately represents how your infrastructure relates to your business services. This is the CSDM model.
Without a populated service graph, change risk predictions have no dependency data to reason over. Incident context cannot surface related CIs. Proactive customer notifications cannot identify which customers are affected by a CI failure.
This is the most common blocker we see for Now Assist deployments: organisations have licensed the features and are waiting for value, while their service graph has no Business Service records and Application Services defined as flat lists of servers.
4. Process instrumentation
Predictive Intelligence and AI-driven routing depend on historical data from instrumented processes. If your incident categorisation is inconsistent, your priority assignments are manual and subjective, or your assignment groups change faster than the training data accumulates, AI models trained on this history will produce unreliable predictions.
Process instrumentation means: consistent categorisation applied at the point of logging (not retrospectively), assignment group structures that are stable over at least 6 months, and priority matrices that are applied consistently rather than left to individual judgment.
GenAI doesn't fix data quality problems. It amplifies them. Good data produces useful AI outputs at speed. Poor data produces confident-sounding noise at the same speed.
The Readiness Assessment Process
When we engage with organisations planning Now Assist deployments, the assessment process covers:
- Data quality scoring: automated analysis of incident, change, and knowledge data against readiness criteria
- CSDM coverage map: how much of your service graph exists, and which Now Assist features it enables or blocks
- Feature prioritisation: which Now Assist features are ready to deploy now, which require data remediation, and what that remediation involves
- Quick-win identification: what can be done in 4 to 6 weeks to unblock the highest-value features
- Roadmap to full readiness: a phased plan that delivers AI value at each stage rather than waiting for perfect data
What You Can Deploy Today
Most organisations are further from full AI readiness than they expect - but closer to partial value than they realise. Some Now Assist features have lower data quality dependencies than others:
- Text summarisation for agents: useful even with imperfect incident data - it saves reading time even if the summary isn't perfect
- Suggested knowledge articles: valuable if your knowledge base has reasonable coverage for your top 10 incident categories
- Email-to-case classification: works well if your case categories are consistent and the training data is clean
Deploy these first, measure the impact, and use the experience to build the case for the data quality investments that unlock the higher-value features.
The SN Architect Connection
Our SN Architect platform is itself an AI-assisted delivery tool - built on the same principle that AI performs well when the data and context are structured correctly. Every SN Architect engagement starts with a structured discovery phase that produces a requirements baseline before any design work begins. This is the same principle applied to Now Assist readiness: structure the data and context first, then apply the AI.
MainStack runs Now Assist and GenAI readiness assessments as standalone engagements. If you want to understand where you are and what it would take to deploy specific Now Assist features, we can run that diagnostic in a working session.