AI agents for the work your team handles every day.

1. 1

   Identify what matters

   Use the budget and laptop size already supplied. Ask one focused follow-up only if another requirement would change the recommendation.

2. 2

   Compare approved product details

   Check product details, currency, and the date of the approved catalog. Rule out the pack without a sleeve; do not recommend a near-match as an exact fit.

3. 3

   Explain the choice and the return terms

   Explain the match and cite the product and policy sources. If nothing meets the requirements or a source is missing, say so and offer a human follow-up.

Sample inputs

Sample catalog reviewed September 8: Daypack 24 costs US$89 with a 15-inch laptop sleeve; Weekend 32 costs US$119 without one. The approved sample policy accepts unused returns within 30 days. Live stock is not supplied.

What you’ll need

Adapt the Customer Support starter for product selection, then add an approved product sheet and return policy. Connect a read-only tool separately if live stock is needed. Verified order help needs an authorized connection and identity checks; it stays unavailable until configured. Keep checkout and refunds with your store.

1. 1

   Answer from the service page

   Confirm that the studio offers on-site team sessions. Explain that this project needs a custom quote.

2. 2

   Qualify the project, one question at a time

   Reuse the headcount, location, and timing already supplied. Ask only for missing project details, then invite a preferred contact method for follow-up.

3. 3

   Prepare the sales follow-up

   Confirm the brief with the buyer. Include the question, source used, known details, and what the studio still needs to confirm. Send existing-customer support needs down the support path.

Sample inputs

The sample service page offers on-site corporate headshots with custom pricing. The visitor supplies the headcount, city, and timing. No calendar availability or price is provided.

What you’ll need

Add service pages, pricing guidance, and a follow-up route. Enable lead capture for contact details. The agent answers basic questions before requesting contact information and leaves unsupported prices or dates for your team to confirm.

1. 1

   Check the service and location

   Check wood-fence replacement and Cambridge against approved services and coverage before collecting a full intake. Recognize urgent hazards and offer the configured human contact path.

2. 2

   Collect the details needed for an estimate

   Ask about property access, the existing fence, and preferred callback times. Keep the requested timeline separate from confirmed availability.

3. 3

   Confirm a useful handoff

   Read back the job details and ask for corrections. Preserve the requested timeline, missing access details, and preferred callback method for the estimator; do not promise a visit is booked.

Sample inputs

The sample contractor replaces wood fences in Cambridge and requires a site visit before a final quote. The visitor provides the approximate length and asks for work this month.

What you’ll need

Provide services, coverage areas, estimate requirements, and a human follow-up route with its hours. Identify the agent as AI and allow a request for a person at any point. Enable lead capture for callback details. Your team confirms quotes and appointments.

1. 1

   Read the relevant variants

   Use an authorized inventory export for the selected location and variants. State its timestamp and quantity type; available-to-sell stock differs from total stock on hand.

2. 2

   Apply the team’s review rule

   Compare available-to-sell units against the supplied 10-unit threshold. Exclude committed, unavailable, and incoming units. Flag missing or stale data instead of estimating it.

3. 3

   Give the operator a decision to review

   Flag the low-stock variant and suggest checking inbound stock before changing the promotion. Keep the analysis read-only.

Sample inputs

Sample export: September 8, 09:00 UTC, East warehouse. Red/small has 8 available, 4 committed, and 2 unavailable units (14 on hand); black/medium has 42 available. The review threshold is 10 available units. Inbound-stock and forecast data are not supplied.

What you’ll need

Adapt the Data Analysis starter using an approved inventory export and your review rules. Include location, snapshot time, and quantity definitions. Keep the agent internal and read-only; the template analyzes supplied data and does not query your store or change stock, prices, or campaigns.

1. 1

   Confirm the first deliverable

   Agree on the first deliverable, target date, answer approver, and a testable definition of ready: approved answers and a verified human follow-up route.

2. 2

   Check the launch essentials

   Record the website and logo as ready. Treat the unapproved returns FAQ as a blocker, not a source the agent can use yet.

3. 3

   Assign the missing work

   Confirm the FAQ owner, approver, and due date. Keep the target launch date distinct from readiness; finish with the blocking item and the next approval checkpoint.

Sample inputs

Sample intake targets October 1. The website and logo are supplied; the client agrees that marketing will submit the FAQ by September 24 and operations will approve it by September 26. Readiness requires approved FAQ answers and a tested human follow-up route.

What you’ll need

Add your onboarding checklist, required knowledge sources, and approval process. Confirm owners with the client and request access through your approved sharing process. Your team reviews the brief and decides when the agent is ready to publish.

1. 1

   Map the trigger and destination

   Identify required fields, contact matching, and deterministic owner-routing rules. Use AI only for unstructured summaries or classification that the team actually needs.

2. 2

   Ask about the exceptions

   Separate duplicate contacts from repeated submissions. Give each submission a stable ID and record each completed action; define what happens on missing email, missing owner, or a failed delivery.

3. 3

   Propose the smallest useful build

   Plan email-based contact matching and deterministic routing, with bounded retries and manual review. If a send times out ambiguously, check its status before resending. Test repeated submissions before launch.

Sample inputs

The sample team chooses email-based matching in HubSpot, rule-based owner routing, and a stable submission ID. Each action records completion so retries can resume safely. An uncertain Slack send goes to review before any resend. These are proposed implementation rules.

What you’ll need

Bring the current process, field requirements, routing rules, and an implementation owner. This starter produces a plan; no CRM records or Slack messages are created. Execution requires a separately configured and tested integration.