World Model Readiness
Engraved team-functions instrument

For your team

Teams by Function

Module · When the bot answers first

The Support Team AI Check

A support bot has one honest job: resolve the request, or hand it to someone who can, fast. The failure mode is quieter than an outage. The bot closes tickets it never solved, the customer gives up, and your dashboard calls that a win. This module checks the five things that separate real automation from expensive deflection: whether you measure resolution or just containment, how fast a stuck customer reaches a human, whether the agent-assist your team leans on is worth trusting, whether the knowledge behind the answers is current, and whether satisfaction survives contact with the bot.

Question 1 of 5 · Resolution, not deflection

Do you measure whether the bot actually resolved the issue, or just that it ended the chat?

Containment counts chats the bot handled without a human. Resolution counts problems the customer no longer has. They are not the same number, and the gap between them is customers who gave up. If your headline metric is containment, you are rewarding the bot for wearing people down.

Question 2 of 5 · Escape hatch is fast

When the bot cannot help, how quickly does the customer reach a human?

Every bot hits its limit. What matters is what happens next: an instant, obvious route to a person, or a loop that keeps offering articles the customer has already rejected. A slow or hidden escape hatch turns a minor failure into a furious customer, and the anger lands on the human who finally picks up.

Question 3 of 5 · Agent-assist is trustworthy

Can your agents trust what the AI drafts and suggests, or do they have to double-check it?

Agent-assist that drafts replies and surfaces answers speeds up good teams and quietly poisons careless ones. If the suggestions are often wrong, agents either waste time correcting them or, worse, paste them through unread. The tool is only a help if your team knows when to trust it and when not to.

Question 4 of 5 · Knowledge is current

Is the knowledge base behind your AI answers kept current, or quietly out of date?

A support bot is only as good as the content it reads from. Old prices, retired products, superseded policies: the bot will state them with total confidence. Nobody notices stale knowledge until a customer acts on a wrong answer, and by then it is a complaint, not a content-management problem.

Question 5 of 5 · Satisfaction survives the bot

Do customers who went through the bot end up as satisfied as those who reached a human?

This is the number that catches everything the others miss. If bot-handled customers rate their experience worse than human-handled ones, the automation is buying you cost savings against loyalty you cannot see leaving. Segment your satisfaction score by channel, or you are averaging away the damage.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

What share of your support contacts does an AI bot handle first?

None
Under a quarter
A quarter to sixty percent
Over sixty percent
We do not track it

Which number does your team report as the bot's headline result?

Containment or deflection
Tickets handled
Confirmed resolution
Customer satisfaction
No headline metric

How does a stuck customer reach a human from the bot?

No human route
Only after several tries
On a clear request
Instantly on intent
No bot in use

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

Resolution, not deflection

  1. 1Containment only
  2. 2Ticket volume
  3. 3Ended-chat rate
  4. 4Resolution tracked
  5. 5Resolution plus follow-up

At the low end: A containment number with no resolution behind it flatters the bot and hides abandoned customers. Start measuring whether the issue actually got solved, by follow-up survey or repeat-contact rate, before you trust the automation. What good looks like: Measuring true resolution plus repeat contact is what lets you scale the bot without hollowing out service. Keep watching the follow-up rate; a resolution number can drift the moment the knowledge behind it goes stale.

Escape hatch is fast

  1. 1No human route
  2. 2Buried deep
  3. 3After several tries
  4. 4One clear request
  5. 5Instant on intent

At the low end: A bot with no fast route to a human traps your most frustrated customers exactly when they need help most. Add a visible, one-step handoff to a person and make it work on the first ask. What good looks like: An instant handoff the moment the customer signals they are stuck keeps a bot failure from becoming a service failure. Watch the handoff volume; a rising rate is the bot telling you where it is weak.

Agent-assist is trustworthy

  1. 1Blindly pasted
  2. 2Often wrong
  3. 3Sometimes useful
  4. 4Reliable, checked
  5. 5Reliable, sourced

At the low end: Suggestions your agents paste without reading turn one AI error into a wrong answer sent to a customer. Require a quick check on every AI draft, and measure how often agents have to correct it. What good looks like: Reliable suggestions with a visible source let agents verify in a glance instead of retyping from scratch. Keep sampling the drafts for accuracy; assist quality tracks the knowledge base and decays with it.

Knowledge is current

  1. 1Never reviewed
  2. 2Ad hoc edits
  3. 3Reviewed yearly
  4. 4Owned and scheduled
  5. 5Owned, fresh, audited

At the low end: A knowledge base nobody maintains feeds your bot confident, out-of-date answers. Assign an owner and set a review cadence before the next product or price change makes the bot lie for you. What good looks like: Owned, scheduled, audited content is what makes an AI answer trustworthy at all. Tie the review cycle to your product and pricing changes so the knowledge updates the day reality does.

Satisfaction survives the bot

  1. 1Not measured
  2. 2Bot rates far worse
  3. 3Bot rates worse
  4. 4Roughly equal
  5. 5Bot equal or better

At the low end: If you never compare satisfaction by channel, the bot can be shedding loyal customers while the average score holds steady. Split your satisfaction metric into bot-handled and human-handled and look at the gap. What good looks like: Bot-handled satisfaction that matches or beats human contact means the automation is genuinely serving customers, not just deflecting them. Keep the comparison live; the moment knowledge or intent handling slips, this gap opens first.