Deloitte GDC: When AI Changed the Link Between Hours and Value

Executive summary

Starting around mid-2024, Deloitte’s Chongqing Global Delivery Center (GDC) began experimenting with AI coding tools for its professionals who support consulting engagements. Some employees reported speed gains, but for more than a year management saw no comparable improvement in project cycle time, revenue per employee, or profit. One bottleneck lay in how employees were evaluated and rewarded. Appraisal had depended almost entirely on utilization, the share of available working time charged to billable projects. This had been a practical, low-cost proxy for contribution. But AI allowed employees to accomplish very different things within the same number of hours. Utilization became less informative about the effort, skills, and contributions GDC wanted to encourage.

GDC reduced utilization’s weight in appraisal and gave more weight to project feedback, demonstrated capabilities, and reusable AI innovations. These measures recognized contributions that time records alone could not reveal: project work, the development of relevant skills, and improvements that benefited other teams. GDC also piloted a gradual move toward an 80/20 fixed-variable pay mix for selected senior staff, refined its career grades, and introduced an expert track alongside the management path. Changes to pay and career progression made the broader assessments consequential. Together, these changes aimed to reward the value employees created, not just the hours they logged.

Background

Deloitte China's Global Delivery Center in Chongqing supports more than 4,000 consulting engagements each year. Client-facing teams originate the engagements, shape the solutions, and manage client relationships. GDC turns the design into working systems and supports them after launch. Its more than 2,300 employees include developers, business analysts, enterprise-system consultants, testers, and data specialists.

Like most professional-services firms, Deloitte prices projects in person-days. At the proposal stage, teams estimate how many days each role will need and apply the relevant daily rates. During delivery, employees charge their actual hours to a project code. Managers compare planned and actual effort to monitor budgets and margins.

The same time records also generate utilization, the share of an employee’s standard available working time charged to billable projects. Historically, utilization offered a practical proxy for contribution and a common basis for assessment without requiring detailed judgments about every employee’s work. Utilization drove almost the entire individual appraisal.

The first intervention: better tools

Starting around mid-2024, GDC began experimenting with AI coding tools. Some employees completed tasks faster. But over the following year, management saw no clear improvement in project cycle time, revenue per employee, or profit. The tools were creating pockets of individual productivity, but not yet a more productive organization. Why did faster employees not make a faster organization?

One reason was that the performance system could not distinguish the resulting contributions. Two employees could each charge eight hours to the same project, yet one might complete more work, produce a better result, or develop a method that saved time for the whole team. Because utilization gave both the same credit, some employees felt that existing measures did not fully recognize differences in contribution and impact.

The problem was therefore not only one of measuring productivity but also one of incentives. Utilization had been attractive because the firm could observe it across thousands of projects. However, it did little to reward the additional care needed to assess and improve AI-assisted work, the acquisition of relevant skills, or the sharing of useful methods. These activities could improve project delivery without increasing hours charged. As hours became less informative about contribution, rewarding them left valuable differences in effort and capability insufficiently recognized. GDC therefore needed a broader way to evaluate and reward contribution.

The second intervention: rebuilding the incentive system

GDC broadened its performance assessment, increased the share of performance-based pay for selected employees, and revised the basis for advancement. The new system aimed to encourage better project delivery, skill development, and the sharing of useful methods.

Measure contribution

In its initial FY26 redesign, GDC reduced utilization’s weight from almost 100 percent to 30 percent. Project Feedback accounted for another 30 percent, while Competency Excellence and GAI Innovation made up the remaining 40 percent.

Project Feedback came from the partner or manager in charge and was collected quarterly by the competency team. Their assessment added information about the employee’s quality of work, collaboration, and responsibility that time records and project-level numbers alone could not reveal. This gave employees an incentive to care about the result delivered to the project, not only the hours charged to it.

Competency Excellence made skill development count. Assessments covered technical, professional, and leadership capabilities, including AI-assisted development, quality assurance, business problem-solving, and project management. Training, examinations, and project experience supplied evidence of progress. These measures complemented project feedback by recognizing capabilities that utilization did not distinguish.

GAI Innovation recognized the sharing of useful methods. Documenting a useful method and helping others adopt it took time, while much of the benefit went to other teams. Making these contributions part of quarterly evaluation, with stronger recognition for contributions that reached beyond the employee’s own project, gave employees a reason to turn personal improvements into shared gains.

Recognition for sharing was accompanied by a process for putting useful ideas to work. For example, practitioners built an AI-assisted method for SAP development. Competency teams and the central transformation task force evaluated the method. GDC then expected comparable work to require about 70 units of effort rather than 100, unless teams could justify an exception. The contribution extended beyond the developers’ own work: they had created a method that other teams could use and a new basis for planning future projects.

Make performance matter more for pay

In FY26, GDC piloted a gradual move toward an 80/20 fixed-variable pay mix with senior staff. Existing fixed salaries were not cut; salary increases were directed into a variable component. Strong performers could receive an uplift through performance multipliers.

The pilot was intended to encourage senior staff to lead from the front: develop better methods, contribute their own knowledge as well as experience, tackle difficult work, and demonstrate what AI-enabled delivery could achieve. A planned expansion would bring project leaders and experienced frontline staff into the arrangement.

Rewards also extended beyond individual performance. GDC had paid a Service Delivery & Transformation incentive linked to organization-wide profitability improvements. It was also designing a project incentive to share part of the gains from AI-enabled delivery with project teams. The proposed incentive would give teams a financial reason to turn improvements in how they worked into better project results.

Link career standing to demonstrated capability

GDC made two changes to its career structure: it expanded the ladder from seven broad levels to twelve and introduced an expert track alongside the management track at senior levels.

The finer grades allowed it to distinguish employees who shared a title but differed in capability and contribution. Analyst, Consultant, and Senior Consultant, for example, were each divided into two grades. The objective was to match career position more closely to demonstrated capability and give employees clearer goals for development. Placement and advancement depended on demonstrated capabilities rather than legacy titles.

The expert track gave technical professionals a route to senior positions without becoming conventional people managers. GDC increasingly needed people who could decompose complex projects, direct humans and AI, judge their outputs, and take responsibility for the results. The new path made developing these capabilities relevant to career progression, not only to the next performance review.

Consequences

Following the combined changes in technology, management, and incentives, GDC reported that productivity per employee rose by close to 7 percent in the year from June 2025 through May 2026, while time spent correcting quality problems fell by more than 30 percent. On some projects using established AI methods, reductions in recorded delivery costs appeared as additional project margin in Deloitte’s monthly accounts.

There were also early signs that the new performance assessments were affecting rewards. Managers reported that most senior employees in the initial pay pilot performed sufficiently well to qualify for performance multipliers on their variable awards.

These results reflect a combination of changes in tools, incentives, training, working methods, staffing, and quality control. GDC changed the broader operating system. The tool-only phase generated scattered individual gains; the wider organizational redesign helped convert them into lower project costs, higher margins, and measurable company performance.

Takeaway

A performance measure should be judged not only by how easily it can be collected, but also by what it reveals about contribution and the behavior it rewards. Time had been a practical proxy for contribution, but AI made hours less informative about careful effort, relevant skills, and the value employees created. If the incentive system remains unchanged, the time saved through AI may simply stay with the individual rather than being converted into better project outcomes or organizational value.

The lesson is that AI can change which performance measures are most informative and which behaviors they reward. Each measure should influence pay, promotion, and status according to how much it reveals about the contribution the firm wants to encourage. This may require combining simple quantitative measures with richer qualitative assessments. Redesigning measurement and incentives help ensure that AI-enabled productivity gains translate into organizational outcomes.

AI creates the possibility of productivity; deliberate organizational redesign turns it into organizational value.