Governing AI Readiness: Business Value With People at the Center

Most organizations use AI now. Far fewer can point to what it has earned them. In McKinsey’s 2026 survey, nearly nine in ten respondents said their organizations use AI regularly in at least one business function, and eight in ten said it had improved their own productivity. Only 37% said AI had added to their organization’s EBIT, about the same share as the year before.¹

A study of 5,172 customer-support agents found that an AI assistant raised the number of issues resolved per hour by 15% on average. The least experienced agents gained the most: they got faster, and their answers got better. Customers in those chats were more polite and less likely to ask for a manager.² At the other end, Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027 because of escalating costs, unclear business value, or inadequate risk controls.³

Most of the difference between those two outcomes is settled before anything goes live. Did someone measure the starting point? Did anyone decide who owns the tool and what it’s allowed to touch? Were the people who do the work part of the design? This paper sets out a model in three parts:

1. Assess. Score readiness across eight dimensions, from data and knowledge to governance and how outcomes will be measured, before any money is committed.
2. Govern. Give each use case an owner, guardrails, and a review schedule, following the NIST AI Risk Management Framework and ISO/IEC 42001.
3. Empower. Use AI to take low-value work off people’s plates and bring every employee closer to the performance of your best, with people keeping the judgment calls.

Done in that order, every AI dollar is tied to a measured outcome you can explain to regulators and customers, and employees treat the tools as something that helps them.

AI is already inside most organizations, whether or not anyone approved it. In Microsoft and LinkedIn’s 2024 survey of 31,000 people in 31 countries, 75% of knowledge workers said they use AI at work. Of those users, 78% bring their own tools, and only 39% have had AI training from their employer.⁴ When leadership hasn’t set the rules, employees set their own. Customer records, pricing, and internal policy end up in whichever tool someone signed up for.

The cost shows up in three places.

Spending without a baseline. In McKinsey’s 2025 survey, fewer than one in five respondents said their organizations track KPIs for their generative AI tools.⁵ A project with no measured starting point can’t show what it changed, so it’s the first thing cut when budgets tighten.

Unreviewed output. In the same survey, 27% of organizations using generative AI said employees review everything it produces before it’s used.⁵ The rest review less than all of it.

Rules with penalties attached. Texas’s Responsible Artificial Intelligence Governance Act took effect on January 1, 2026. Government agencies must tell people when they’re interacting with an AI system, health care providers must disclose their use of AI in certain settings, and certain uses are banned outright. The Attorney General enforces it, with civil penalties of $10,000 to $200,000 per violation and up to $40,000 a day for violations that continue.⁶ California followed in September 2026 with AB 1609. Starting January 1, 2027, it bars large businesses from presenting a customer service chatbot as a human and requires them to let customers reach a person during business hours.⁷

Governance is also where the value comes from. Of the 25 attributes McKinsey tested in 2025, redesigning workflows had the biggest effect on whether an organization saw EBIT impact from generative AI. Only 21% of organizations using it had fundamentally redesigned even some of their workflows.⁵ In the 2026 survey, the 6% of respondents McKinsey counts as high performers stood out for the same habit, along with human-in-the-loop design, impact measurement, and risk management.¹ Redesigning a workflow takes someone with the authority to decide what the work should look like, who owns it, and how success will be measured.

AI pays off in two ways. It handles work that never needed a person, and it helps the person who handles everything else work faster and get it right. Customer service shows both: routine calls can be answered without an agent, and in the study above, agents with an assistant beside them resolved more issues per hour.²

The same split holds in other departments.

The last column is where most organizations fall short: fewer than one in five track KPIs for their generative AI tools.⁵ The ones that get results measure a baseline first, redesign the workflow around the tool, and track the outcome afterward. A readiness assessment gives you that baseline.

The most useful finding in the customer-support study is who benefited. The assistant helped newer agents most, because it put the know-how of the best agents in front of everyone.² The goal is AI that brings every employee closer to the performance of your best, rather than AI that replaces them or watches them.

Service leaders are already planning this way. In a Gartner survey of 321 customer service and support leaders, 85% said they’re adding new responsibilities to frontline roles as AI cuts contact volume, while only 31% had carried out or planned AI-driven layoffs through early 2027.⁸ Customers want a person within reach, too: 87% told Gartner it’s essential for a company using generative AI in customer service to offer a way to reach a human.⁹

Six principles put this into practice.

AI takes the task, and people keep the judgment. Routine work is automated. Exceptions, complaints, and decisions that affect a customer’s money, health, or rights stay with a person.

The people who do the work help design it. They point out what slows them down and test the tools before anyone else has to use them.

Training comes before rollout. In 2024, only 39% of AI users had been trained by their employer.⁴ Everyone whose role touches AI learns what it does, where it gets things wrong, and when to override it.

Employees know what’s being measured. They know what AI records about their work and how it’s used. Coaching data is used for coaching and nothing else.

Freed-up time has a plan. Hours AI gives back go to complex cases, retention calls, outreach, and skills development, as agreed before rollout.

Anyone can raise a concern. An employee who spots a wrong answer, an unfair outcome, or a risky use can flag it, and the flag reaches someone who can act on it.

These principles decide whether employees use the tools you approved or keep bringing their own, and whether the productivity gain ever shows up in your numbers.

The assessment produces a readiness score for each dimension, along with the evidence behind it. It starts with an inventory of every system that touches a customer or employee interaction: the phone system and IVR, call routing, workforce management, quality management, CRM, call recording, transcription, digital channels, and any outsourcer platforms. Each dimension below is then scored on a four-stage scale.

The four stages, from first steps to routine:

You come away with a scored report for each dimension, a list of use cases ranked by value and readiness, and the governance steps each use case needs before it goes live.

Governance answers four questions for every AI use case: who decides, what the rules are, how it’s measured, and how often it’s reviewed. The model below follows the four functions of the NIST AI Risk Management Framework (Govern, Map, Measure, and Manage).¹¹ It can be extended to meet ISO/IEC 42001, the first international standard for AI management systems.¹²

Decision rights

Executive sponsor. Sets AI strategy, budget, and how much risk the organization will accept, and answers for business outcomes and final escalations.

AI governance council. Operations, IT security, legal, HR, and a frontline representative. The council approves each use case and its guardrails, and owns policy, the AI inventory, and risk review.

Use-case owner. Runs one AI use case day to day and answers for its KPIs, incidents, and user feedback.

Data and knowledge owners. Decide what data and content AI may use, and keep it accurate, current, and limited to the people who need it.

Frontline champions. Help shape design, testing, and training, and flag problems early.

Review schedule

Each use-case owner reviews KPIs against the baseline, incidents, and user feedback every month. The governance council meets quarterly to approve new use cases, update policy, review risk, and check the AI inventory. Once a year, or after a major platform change, the readiness assessment is repeated and priorities are reset.

The assessment is the first phase of the work. Each phase has to pass a gate, a measured result, before the next one starts and before more money goes in.

Computer Instruments has built self-service voice systems for public sector and enterprise organizations since 1979. Our team designs and integrates them, adding AI to the phone infrastructure an organization already has without disrupting it. For many organizations, the phone line is a sensible first use case. The routine questions are easy to name, the call volumes are already counted, and the baseline is sitting in the call data.

The AI Concierge answers callers’ routine questions in their own words, on the phone system you already run. It can tell callers they’re speaking with AI when the greeting is set up to say so, and it can hand a caller to a person when they ask.¹³ That supports two of the guardrails in the checklist above.

If you’d like to start with the assessment, we’ll give you a scored report for each dimension, a ranked list of use cases, and the governance steps each one needs before it goes live.