AI
One goal for the reply.
The goal is a short line: send the booking link, answer the shipping question, or ask for the email. The product requires a goal when AI is on.
3 minute read
Examples
Keep the goal to one action.
- Send the menu link and the hours.
- Ask for an email, then send the guide.
- Answer fit questions from the size note. Do not offer a discount.
Write a goal you can judge from one answer
A useful goal names the intended outcome and its boundary. Answer a shipping question from the note and provide the shipping-policy link is more testable than maximize sales. DripDM requires a goal when AI is enabled and allows up to 500 characters. The goal steers generated wording; it is not a separate workflow engine that automatically books, charges, or qualifies a customer. Do not import competitor goal types or automatic success counters that the implementation does not establish.
A workflow to try
For a coach, set the goal to explain the listed packages and provide the booking destination when the person asks about working together. Keep package facts in knowledge and tone in instructions. Preview a price question, a fit question, and a request for guaranteed results. Judge whether the answer advances the selected next step without promising an outcome or inventing a discount. A booking-link click is evidence of interest, not proof that a call was scheduled.
Checks before you change the setup
If replies wander, narrow the goal rather than adding several competing objectives. Asking the same response to sell, collect an email, grow followers, and solve support disputes creates unclear priorities. Keep the fallback appropriate when the answer cannot be established. Live generation may include recent conversation history, but a preview uses the supplied sample without that history. Review both the preview and real reply context before deciding that the goal reliably handles every thread.
Separate objective, facts, and tone in the settings
Write the outcome in the goal, the prices and policies in knowledge, and the preferred style and boundaries in instructions. If a test answer is too long, adjust the length setting rather than deleting essential facts from the note. If the next step is unclear, simplify the goal rather than adding several new destinations. Use the fallback to handle uncertainty instead of telling the model to guess. This separation makes later maintenance easier: a new price changes the factual note, a new booking route changes the destination, and a new campaign objective changes what a useful answer is trying to accomplish.
Questions
Up to 500 characters. A single sentence is easier to test.
No. It can steer the generated message toward the configured booking link or next step. Confirmation still belongs to the booking system. Measure a click separately from a booking, and avoid wording that says a time is reserved when no reservation action occurred.
Prefer one testable next step. Competing instructions to sell, collect data, grow followers, and resolve support disputes make evaluation harder. Put facts in knowledge, style in instructions, and the intended next action in the goal.