One of your best agents from last year is now in the bottom half of the handle-time ranking. Her performance may be exactly the same, but her queue has changed.
Since the bot went live, password resets, order status questions, and balance inquiries are resolved before they reach an agent. The calls that remain are angry escalations and billing disputes that touch three systems. The rest come from customers who already tried self-service twice.
If your bot now resolves the simple contacts, every agent metric needs a new baseline built on the calls that still reach people.
Why does AHT go up after a chatbot goes live?
KPI drift is the gap that opens between a contact center's performance targets and the work agents actually handle once automation changes the contact mix. A bot that resolves simple contacts removes the short calls that used to pull average handle time (AHT) down. AHT goes up even when every agent performs exactly as before. The drift reaches past AHT. Adherence thresholds, coaching triggers, bonus baselines and WFM staffing forecasts were all set for the old mix. In a billing queue, for example, the bot takes the balance inquiries and agents keep the disputes.
I tell clients to beware of averages. If the bot takes the two-minute calls out of a queue that averaged five minutes, AHT goes up. It goes up even if every agent works exactly as fast as before. Keep the old target and you penalize agents for handling the work you designed the bot to send them.
KPI drift does damage beyond the scorecard. Agents land on coaching plans for missing an outdated target, and bonuses depend on baselines that no longer match the work. Forecasting models also underestimate staffing because they still assume the old contact mix.
Does AI make the work left for agents harder?
AI often makes the work left for contact center agents harder. Automation takes the short, repeatable contacts, so agents spend more of each shift on escalations, multi-system billing disputes and customers who already tried self-service. Research inside and outside the contact center points the same way.
In February 2026, Harvard Business Review published a study by Aruna Ranganathan and Xingqi Maggie Ye of UC Berkeley's Haas School of Business. They followed a technology company of about 200 employees for eight months. Employees using AI worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day.
In contact centers, Omdia's 2025 Digital CX Survey found that 75% of North American contact center leaders believe their AI investments may be increasing agent stress.
Turnover points the same way. Metrigy tracked contact center turnover rising from 21.8% in 2022 to 28.1% in 2023, with a projection of 31.2% by the end of 2024. Separately, Metrigy research cited by Five9 found that agent assist tools cut average handle time by 27%.
These findings belong in any AI ROI business case. The tools make each contact faster, and the contacts that remain are harder on the people who take them. If those agents are still measured against pre-bot standards, the scorecard describes a job that no longer exists.
Technology alone won't improve your contact center. Execution matters. With AI, execution includes changing how you measure the people who work next to it.
How should you change agent scorecards after AI deflection?
Start with the call mix. If the bot now handles a large share of simple contacts, review the agent scorecard using the contacts that still reach agents. Give complex queues more weight on resolution and quality. Re-baseline speed metrics on the post-bot call mix, and split them by contact type where the mix varies. WFM teams should also review adherence thresholds and staffing assumptions for queues that now take more complex or emotionally difficult calls.
Pay-for-performance deserves the same review. A bonus target set when the queue was full of short, simple contacts no longer measures the performance you meant to pay for.
Metrics to check first:
Average handle time and other speed targets, split by contact type
Resolution and quality assurance weighting in complex queues
Schedule adherence thresholds for queues with more escalations
Bonus and pay-for-performance baselines set before the bot went live
Handle-time inputs in your WFM platform's forecast and staffing model
Put one meeting on the calendar. Sit down with your WFM and QA leads and list every agent metric that still assumes the pre-bot call mix. Then decide which baselines change, based on the work agents handle today. If an AI implementation is still ahead of you, build that review into the plan.
Melissa Copeland is Founder & Principal of Blue Orbit Consulting, a contact center and CX technology consulting firm. She is an ICMI Top 25 Thought Leader 2026, and Forbes named her America's Top Contact Center Expert.