Amazon Connect Evaluation Form Templates: One Scorecard for Human and AI Agents
Amazon Connect now ships ready-made evaluation forms for service, sales and self-service that score human and AI agents on the same criteria — including outcomes like satisfaction and attrition risk.
Quality programs usually stall at the same place: somebody has to design the scorecard before anyone can be scored. Amazon Connect's November 9 addition removes that first obstacle by shipping ready-made evaluation form templates — and it quietly does something more interesting at the same time. The same templates score human agents and AI agents, on the same criteria, including the business outcomes that actually matter. If your quality team has been waiting for a reason to start automated evaluation, or for a fair way to compare an AI agent against the humans it works alongside, this is both.
What changed
AWS added evaluation form templates for three common scenarios: customer service, inbound sales and self-service. Each template assesses both human and AI agent behavior — for example, whether the agent showed empathy for the customer's issue — and includes questions about business outcomes such as customer satisfaction and attrition risk. Managers can deploy a template as-is or customize it first, which shortens the setup for automated evaluations.
The templates are available in N. Virginia, Oregon, Canada (Central), Frankfurt, London, Singapore, Sydney and Tokyo. Pricing follows the Amazon Connect Customer pricing page.
For the full release record, see the Amazon Connect release intelligence page and the official AWS What's New announcement (opens in a new tab). The update behind this article is Evaluation form templates cover service, sales and self-service.
Pattern 1: Start automated QA from a template, not a blank form
The most common reason automated evaluation programs never start is that building the first form takes weeks of workshops: what counts as a good answer, how much empathy weighs against resolution, who signs off on the rubric. A template turns that into an editing job. Pick the template closest to the work your team actually does — service, inbound sales or self-service — deploy it as-is on a small sample, and let the first round of scores tell you what to change.
That first round is a calibration exercise, and it matters more than the template itself. Score a batch of interactions with the template, then have experienced evaluators score the same interactions by hand and compare. Where the automated scores and the human scores disagree, you have found the questions that need rewording, reweighting or a better definition — and you have found them on a dozen interactions, not after a quarter of production scoring. Only once the template agrees with your evaluators on the samples should it start scoring live work, and even then, keep a human sample running alongside it so drift shows up early.
Resist the temptation to replace a calibrated form overnight. If your team already has evaluation forms that have been tuned over months, treat the templates as a reference point: borrow the outcome questions where they fill gaps — satisfaction and attrition risk are the two most commonly missing — and leave the scoring your team trusts in place. A template is a starting point, not a mandate, and the fastest way to lose a quality team's trust is to swap their rubric without warning.
Pattern 2: Score AI and human agents on the same rubric
The genuinely new thing here is not that forms exist, but that one form covers both kinds of agent. Until now, AI agents were usually judged on task completion — did the bot resolve the issue, did it escalate correctly — while human agents were judged on behavior — empathy, tone, ownership. Two scorecards meant two definitions of good, and no honest way to compare them.
A shared rubric changes the conversation. Run the same template over a week's worth of AI-handled self-service interactions and human-handled service interactions, and the comparison becomes concrete: where does the AI agent show less empathy than a human, where does it outperform on consistency, where do both struggle with the same kind of issue. That last finding is usually the most valuable — if humans and the AI agent both score poorly on the same outcome question, the problem is probably the process, the knowledge or the policy, not the agent.
Use the shared scorecard to draw the line between the two deliberately. If the AI agent scores well on routine self-service but poorly on empathy in emotionally charged cases, that is evidence for routing those cases to humans sooner rather than evidence that the AI agent is failing. If a human agent scores well on empathy but the outcome questions show customers still leaving dissatisfied, the coaching conversation is about resolution, not warmth. The template gives both kinds of agent the same mirror, and the routing and coaching decisions get sharper because of it.
Pattern 3: Let the outcome questions drive the coaching loop
The behavior questions in a template tell you how an interaction went. The outcome questions — satisfaction, attrition risk — tell you whether it mattered. Most quality programs collect the first and ignore the second, which is how a team can score 95 percent on its rubric while its customers quietly leave. The templates put both on the same form, and the pattern worth adopting is to treat the outcome score as the trigger for everything else.
Set it up as a loop. When an interaction scores low on an outcome question, that is the signal to look at the behavior questions on the same form: did the agent miss empathy, did it skip a step, did the self-service flow end without confirming the customer's problem was solved. The behavior answers become the diagnosis, and the diagnosis picks the fix — a coaching conversation with the agent, a knowledge article that needs rewriting, a flow step that needs reordering, a policy that needs clarifying. Then the next round of evaluations measures whether the fix moved the outcome score.
That loop gets more useful the more systems it touches, so think of the evaluation form as one input among several rather than a standalone report. Pair the outcome signal with what your survey tool already hears — a post-interaction survey from a tool like Qualtrics or Medallia — and with what your CRM knows about the customer — a Salesforce or ServiceNow record showing whether this was the third contact about the same issue. An agent who scores well on behavior but whose customers keep calling back about the same problem is not a coaching problem; it is a routing or knowledge problem, and you only see that when the evaluation score sits next to the CRM history. Similarly, feed the themes from low outcome scores back into the places that create them: the knowledge base the AI agent answers from, the scripts the human agents follow, the hiring and onboarding plan in your HRIS. The template is the starting point; the loop around it is the program.
Summary
Evaluation form templates are a small feature with a large second-order effect: they make starting automated quality work cheap, and they make comparing human and AI agents honest by scoring both on the same form. Deploy a template on a sample, calibrate it against human scorers, borrow the outcome questions into your existing forms where they fill gaps, and let low outcome scores drive the coaching and process fixes that follow. The scorecard is ready-made; the program you build around it is the part that moves customer outcomes.
