MedeAnalytics, a healthcare performance improvement company, cut its annual employee goal review from a full day of manual spreadsheet work to about ten minutes using ChatGPT Enterprise. The HR team at the company, which has roughly 1,000 employees, now runs its annual goal export through a two-prompt AI system that checks whether each goal aligns with executive objectives - a process that previously required the entire HR team to read every goal statement individually.
The change matters because goal attainment feeds directly into the company's bonus plan. Previously, employees in bonus-eligible roles were paid out based on overall company performance, regardless of whether their individual goals were meaningful or measurable. Chief people officer Lisa King said that wasn't fair for employees who were exceeding expectations.
The setup: A normalization prompt and an evaluation prompt
King's team pulls the same ADP goal export they always did - roughly 20 fields per goal including employee position, manager, department, goal titles, descriptions, and timestamps. The raw data goes through a first prompt that normalizes it into a consistent format. A second prompt then evaluates each goal against the management by objectives (MBOs) set by each divisional executive.
The second prompt applies what King calls a "custom semantic alignment framework." Instead of matching keywords, it compares employee goals to executive MBOs through business-outcome reasoning. Each goal gets sorted into one of three buckets: it directly owns an outcome, it contributes to that outcome through enablement or operational support, or it isn't aligned at all. The tool generates an individualized rationale for every goal, quality-audits its own reasoning, and produces a final evaluation in Excel.
"[In one performance review cycle], it actually said, 'All these people are okay, these people are not. These people have two goals that are aligned, but these two don't align to the MBOs' - and we did that within ten minutes," King said.
Because the two-prompt structure is reusable, the team doesn't rebuild it each cycle. The same normalization-then-evaluation method runs again when goals need auditing. Ambiguous cases - particularly for technical, specialized goals - get kicked back to the relevant executive for a direct call rather than left to the AI.
What worked and what didn't
The clearest gain was speed. Instead of a team spending a workday or more combing through a spreadsheet, the review happens almost immediately. That freed King's team to focus on coaching managers whose teams' goals needed rework, rather than on manual detection work. The process also became more consistent, since the same prompt evaluates every goal against the same criteria regardless of which team member happened to read a particular entry.
King was candid that the tool's usefulness depends entirely on the person reviewing its output - and that this is where she thinks a lot of HR teams could get into trouble.
"Someone who's never been in HR could look at this and say, 'Oh, that looks correct, I guess I've got to go with it,'" she said. "To be able to validate that it's real or not takes someone who already knows the subject matter, not just someone who knows how to run the prompt."
The cultural and language gap didn't disappear either. AI could flag a goal as poorly formed, but King's team still had to have conversations with international employees about what a SMART goal is and why it's different from a development goal. That work stayed human.
Because MedeAnalytics handles sensitive employee and health-related data, the company requires staff to use enterprise-licensed AI tools - Microsoft Copilot and ChatGPT Enterprise - rather than free or personal accounts. The company is high-trust certified and handles personally identifiable information and health data as part of its core business.
Advice for HR teams
King's approach offers a template for other HR teams looking to apply AI to performance data. She started with data her team already had structured in their HRIS - she didn't build a new system. She wrote the prompts from her team's own expertise, building them around the company's actual corporate objectives rather than a generic goal-quality checklist. And she kept a human in the loop who knows what "good" looks like.
The payoff was concrete: saved time and a fairer bonus process. "It really has transformed our work," King said. What used to take days of manual review now happens in minutes, giving the team more time for the part of the job that still requires a person: the conversation with a manager or employee about what a good goal actually looks like.
For HR teams considering a similar approach, King's experience points to a broader lesson about AI for Human Resources: the tool is only as useful as the criteria you give it, and someone with real HR judgment still has to validate the output. The same principle applies to AI for Management more broadly - AI lets you apply judgment to more people and faster, rather than replacing judgment itself.
Why this matters for management professionals
Performance management is a recurring pain point for managers at every level, and this case shows what AI can actually do about it: not automate the judgment, but compress the administrative work around it. The MedeAnalytics example demonstrates that AI quality-checking works best when it's built on structured data you already have, tied to a concrete business outcome like bonus fairness, and paired with a reviewer who can tell the difference between a valid flag and a false positive. Managers responsible for goal-setting processes should note that the bottleneck in performance management was never reading the goals - it was deciding what to do about the ones that don't measure up. AI can surface those cases in minutes, but the conversation about fixing them still belongs to people.
Your membership also unlocks:
