UnderstandStrengths & Limitations

Strengths & Limitations

Honest expectations save the most time. What works well today:

  • Recurring conversations with a clear sequence (appointment, enquiry, status check, qualification).
  • Answering frequent questions when the answer has been stored.
  • Capturing data and passing it on in structured form.
  • High volumes of similar calls — including at night, including in parallel.
  • Live lookups in your own systems when an integration is connected.

Some situations still need a deliberate fix. Here’s what to do when they come up:

  • If a spelled-out name, street, or email address might be wrong — have the agent read it back for confirmation before it’s used. Build this into the flow rather than assuming it heard correctly; see Writing the Prompt → Examples for how to script a read-back.
  • If accents, background noise, or overlapping speakers cause mishearing — have the agent confirm anything it’s unsure about instead of guessing, and ask the caller to repeat if needed. If this happens systematically rather than occasionally, it’s a platform-level issue, not something your prompt can fix — see Roles & Responsibilities and report it to support.
  • If a request is long, tangled, or has no clear call reason — don’t try to handle it with a longer prompt. Set up a transfer to a human for exactly this case.
  • If the caller is upset, escalating, or trying to negotiate — transfer immediately rather than letting the agent attempt it. Add this as an explicit transfer condition; see Transfer to a Human.
  • If a call feels like it has awkward gaps — keep the agent’s answers short; see Writing the Prompt → Response Rules. A voice agent takes its turn the same way a phone call does — a brief beat between one side finishing and the other starting — so it will never feel exactly like two people talking face to face. Short, direct answers are what make that turn-taking feel natural instead of sluggish.

Rule of thumb: the clearer the call reason, the better the result. An agent built for three precise call reasons beats an agent built for “everything” every single time.

Continue to Roles & Responsibilities to see who owns what when something isn’t working.