How the score is calculated
Each statement is answered on a scale from 1 (strongly disagree) to 5 (strongly agree). A dimension scores the average of its statements. The overall score is the weighted average of the dimensions; the weights are shown below and total 6. A band is assigned only when every statement is answered, and an answer outside the scale counts as unanswered.
Every statement answered at 2 or below adds its action to your checklist, weakest dimension first. The next diagnostic is the one named below for your lowest-scoring dimension.
Bands
| Overall score | Band | What it means |
|---|---|---|
| 1 to below 1.8 | Not Ready | Your answers show significant gaps in data, workflows and governance. AI deployed on this foundation is likely to underperform or create risk. Priority: build the data and workflow foundation before investing in AI tools. Start by assessing data quality, documenting workflows and writing the governance policy. Hold off on buying AI tools until those are in place. |
| 1.8 to below 2.6 | Early Stage | Some foundations are in place, but critical gaps remain. A small AI pilot may work in a narrow, well-defined use case. Priority: close the biggest gaps in data access, integration and governance before expanding. Pilot AI in one high-volume, low-complexity use case. At the same time, invest in data quality, API readiness and governance policy. |
| 2.6 to below 3.4 | Foundation Set | The core is ready for structured AI deployments. Data access, workflows and governance work, though some areas lack depth. Priority: expand AI one use case at a time while you strengthen the weak dimensions. Roll out agent assist (AI that suggests answers and steps to agents during the contact) and automated summaries broadly. Start IVA (intelligent virtual agent) pilots for your top 3 contact types. Invest in your weakest dimension. |
| 3.4 to below 4.2 | AI Capable | Readiness is strong on most dimensions. The organization can support substantial AI deployments, including AI that resolves contacts on its own, agent assist and predictive analytics. Priority: refine and scale. Scale AI resolution for Tier 1 contacts (the simplest, most common requests). Roll out real-time agent assist across voice and digital. Build AI governance into standard operating procedures. |
| 4.2 and above | AI Advanced | Strong readiness on every dimension of this rubric: data, architecture, governance and talent are in place to operate AI as a core part of the service model. Priority: move toward agentic AI (AI that carries out multi-step tasks across systems) and orchestration of the whole customer journey. Test agentic workflows, proactive service automation and AI-driven orchestration on well-governed contact types first, and measure each against its baseline. |
Automation pattern the rubric maps to
| 1 to below 2 | Era 1: Rules-Based | Start with rules-based automation, such as a better IVR (the phone menu) or a simple chatbot. Build the data and workflow foundations before investing in AI. |
| 2 to below 3 | Era 2: Intent-Based | Start with intent-based automation, which recognizes a set list of customer requests. Large language models (LLMs) need data quality and governance in place first. |
| 3 to below 4 | Era 3: Hybrid (Intent + LLM) | These answers fit a hybrid virtual agent: intent matching first, with an LLM behind it for what the intents miss. Fully autonomous AI needs the data and governance gaps closed first. |
| 4 and above | Era 4: LLM-Native | On these answers the rubric maps to LLM-native automation, including autonomous agents on well-governed contact types, measured against a baseline. |
Dimensions, statements and actions
Data Quality & Access
Weight 1 of 6. If lowest, next diagnostic: CX IT Alignment| Statement | Action if answered 2 or below |
|---|---|
| Customer interaction data (calls, chats, emails) is captured, stored and retrievable in a structured format. | Confirm every channel's interactions are recorded or logged, retained and retrievable in a structured, queryable form. |
| CRM (customer relationship management) and case data is reliably linked to interaction records, so AI can see the customer's context. | Link each interaction record to its customer and case ID, and measure the share of interactions that link. |
| Knowledge base content is current, well organized and readable by software as well as people. | Audit the knowledge base for stale and duplicate articles, assign article owners, and set a review date on each article. |
| Data quality problems (duplicates, missing fields, stale records) are measured and someone works them down. | Pick three data quality measures, such as duplicate rate, missing fields and record age, and report them monthly with an owner. |
Workflow Readiness
Weight 1 of 6. If lowest, next diagnostic: AHT Decomposition| Statement | Action if answered 2 or below |
|---|---|
| Common contact types (order status, billing, scheduling) follow written, repeatable workflows. | Document the step-by-step workflow for your five highest-volume contact types. |
| Escalation paths and exception handling are clearly defined and followed the same way every time. | Write down each escalation path and exception rule, and check that agents follow them in QA (quality assurance) reviews. |
| Agent desktop workflows are standardized: agents follow the same steps for the same type of issue. | Standardize the desktop steps for each common issue type and remove local workarounds. |
| You know which contact types are high in volume and low in complexity, the usual first candidates for automation. | Rank contact types by volume and complexity to produce a short list of automation candidates. |
Integration Architecture
Weight 1 of 6. If lowest, next diagnostic: Platform Decision| Statement | Action if answered 2 or below |
|---|---|
| Core systems (CRM, the contact center platform, knowledge base, billing) have documented APIs that are actively maintained. | List the APIs of each core system, confirm which are documented and supported, and flag the gaps. |
| The CCaaS (contact center as a service) platform can send events as they happen, through event streams or webhooks, to other systems. | Confirm in writing which real-time events and webhooks your contact center platform exposes. |
| An automated interaction can call your identity and authentication systems to verify a customer. | Test whether an automated interaction can verify a customer's identity without handing off to an agent. |
| A named owner or team keeps the connections between systems working. | Name an owner for cross-system integrations and give them a monitored list of every connection. |
AI Governance & Policy
Weight 1 of 6. If lowest, next diagnostic: Governance Model| Statement | Action if answered 2 or below |
|---|---|
| A written policy says what AI may and may not do when it deals with customers. | Write a policy that lists what AI may and may not do with customers, and require it for every deployment. |
| AI outputs are reviewed, tested and approved before they reach customers. | Add a test and approval gate before any AI output reaches customers, with a named approver. |
| A named person owns AI quality, model performance and the design of escalations from AI to people. | Name an owner for AI quality, model performance and escalation design. |
| Compliance, privacy and consent requirements are written down and applied to interactions that AI handles. | Document the consent, privacy and compliance rules that apply to AI interactions and test each deployment against them. |
Talent & Change Readiness
Weight 1 of 6. If lowest, next diagnostic: Attrition Cost| Statement | Action if answered 2 or below |
|---|---|
| You have people who can configure, tune and maintain AI tools, or a plan to hire or train them. | Identify who will configure and tune AI tools, and fund training or hiring before deployment. |
| Frontline agents and supervisors understand how AI will change their roles and daily work. | Brief agents and supervisors on how their work changes, and collect their questions before launch. |
| Leadership has realistic expectations for how long AI takes to deploy and what it will deliver. | Set written targets for AI timelines and outcomes with a baseline, and agree how they will be measured. |
| A change management plan covers how agents adopt the tools, whether they trust them, and how their feedback gets back to the team tuning them. | Write a change plan that covers adoption measures, agent feedback channels and how feedback changes the tool. |
Measurement & Iteration
Weight 1 of 6. If lowest, next diagnostic: AI Deflection Reality Check| Statement | Action if answered 2 or below |
|---|---|
| AI performance has defined KPIs (key performance indicators), such as containment rate (the share of contacts the AI finishes without a person), handoff quality, answer accuracy and resolution time. | Define AI KPIs before the pilot starts, including whether the issue was resolved after automation and whether the customer contacted you again. |
| AI interactions are monitored as closely as human ones, with QA scoring and compliance checks. | Sample AI interactions into QA and score them against the same criteria as human interactions. |
| AI performance data feeds a regular cycle of tuning and improvement. | Schedule a recurring tuning review where AI performance data leads to specific changes. |
| You can measure the economic effect of AI: the change in cost per contact, agent time freed, and deflection rate (the share of contacts that never reach an agent). | Set up the baseline cost per contact and resolution rate now, so the economic effect of AI can be measured later. |
What this assessment cannot tell you
- It records how the respondent, the person answering, sees the organization. Nothing here tests your data, your APIs or your models.
- The score is a position on this rubric. It does not rank you against other organizations or give a percentile, because no sourced distribution of scores exists to compare against.
- All six dimensions carry equal weight. A single critical gap, such as no consent handling, can matter more than the average suggests.
- It recommends actions and a next diagnostic. It never recommends a vendor or a product.