A voice agent should pause on an uncertain street name, read back what it heard and let the caller correct it through speech, spelling or text. It should then confirm the complete address and preserve any remaining uncertainty before dispatch preparation. A plausible street name is still a guess until the caller has confirmed the intended destination.
For a South African service business, the useful design decision is what happens between hearing an address and preparing a job. The workflow below is a proposed approach for AI chatbots with a voice intake channel, rather than a claim that address confirmation comes built into a particular product.
Ask about the part that is uncertain
Avoid making callers repeat everything when only the street name is unclear. Separate the address into property number, street name, street type, suburb or locality, town or city, and any unit or entrance information needed for the visit.
A useful prompt is: “I have the property number and suburb. I’m unsure about the street name. Could you say just that name again?” This explains the problem and gives the caller a manageable task.
If there is a tentative reading, make its status audible: “I may have heard Mhlabeni Road. Is that correct?” Do not silently replace an unfamiliar name with a familiar one. If there is no defensible reading, ask for repetition without suggesting a candidate.
Treat an interruption, correction or inconsistent answer as a reason to reopen the affected field. A clear-sounding transcript does not establish that the destination is correct. Likewise, a confident delivery by the agent should never substitute for the caller’s confirmation.
Use a short recovery sequence
Start with repetition, then change the method if uncertainty remains. Ask the caller to spell the name in manageable chunks, using their own letter examples where helpful. Read those chunks back before assembling the complete name. Do not force a caller to use an unfamiliar spelling alphabet.
After recovering the name, confirm the street type separately if necessary. “Road” and “Street” should remain distinct fields when the business needs that distinction to locate the property.
Offer text confirmation through a channel the business actually supports: “Would you prefer to send the address in writing?” Explain where to send it and how staff will associate it with the enquiry. Do not promise an SMS or messaging link unless that connection exists and works.
OpenAI’s audio documentation describes speech-to-speech workflows and voice interfaces built around transcription, a text agent and speech generation. These are building options; the confirmation sequence must be designed by the business. Source: OpenAI audio workflow guide
A proposed stopping rule is to offer another method after an unsuccessful repeat and spelling attempt. If the caller prefers help sooner, transfer sooner.
Confirm the destination after correcting the name
A street-name correction can leave the rest of the address wrong. Read back the complete destination after the correction: property number, street name and type, locality, town or city, and relevant unit details.
Ask a concrete question: “Is that the address where the technician should attend?” This helps distinguish a service address from a billing address or the caller’s current location.
An address lookup, if connected, should supply candidates for confirmation. A matching street name does not establish the correct property or entrance. Ask for the missing locality before choosing between duplicate names. Keep a landmark as supporting directions rather than silently converting it into a street address.
Function calling allows a model to request access to application tools; the application executes those requests. Connecting an address search or saving a confirmation therefore requires an implemented integration and checks on its inputs. Source: OpenAI function calling guide
The broader AI agents versus automation comparison helps explain this division: conversation can recover information, while explicit workflow rules decide whether the record can progress.
Keep uncertainty visible in the intake record
Store the initial captured wording separately from the latest caller-confirmed version. Label the first value as an initial capture, since transcription may itself be mistaken. Keep alternatives as alternatives, rather than joining them into one invented name.
Useful fields include the confirmed address components, confirmation method, unresolved issue, next action and person responsible for follow-up. Use readable statuses such as “awaiting spelling”, “awaiting text”, “caller confirmed” and “human review needed”. These are proposed business states.
For example, “caller confirmed street name; property number missing” is more useful than “address confidence: high”. It tells the dispatcher exactly what remains unfinished. Avoid asking a language model to invent a numerical confidence score as evidence of accuracy.
Structured Outputs can constrain a response to a supplied schema. That supports consistent fields, but schema conformity alone cannot establish that a street name was heard correctly. The caller’s confirmation and unresolved issues still need their own checks. Source: OpenAI Structured Outputs guide
Retain the address evidence needed for follow-up without copying unrelated conversation into the job record. A custom AI agent needs these business-specific instructions as well as a conversational interface.
Reusable street-name confirmation checklist
Use this proposed checklist for each service-address intake. Adapt the required fields to the actual service before use.
- Capture the initial address wording and identify the uncertain component.
- Ask the caller to repeat only the unclear name or detail.
- Read back any candidate as tentative; ask for an explicit correction or confirmation.
- If uncertainty remains, offer spelling in chunks or confirmation through an existing text channel.
- Record the caller’s spelling or text separately from the initial capture.
- Confirm property number, street name, street type, locality, town or city, and relevant unit or entrance details.
- Read back the complete revised address and ask whether it is the service destination.
- Record the confirmation method and any detail still missing or disputed.
- Treat lookup matches as candidates until the caller confirms the intended destination.
- Check whether this enquiry duplicates an existing job; preserve conflicting addresses for human review.
- Assign unresolved cases to a named person or monitored queue, with a clear follow-up action.
- Keep dispatch preparation on hold when a required destination detail remains unresolved.
Completion rule: the record may enter dispatch preparation when the required destination fields are present, the caller has confirmed the complete address, and no unresolved conflict remains. A human dispatcher still applies the business’s normal release checks.
Worked cases: ordinary, missing, ambiguous and duplicate
All addresses, names and numbers in these examples are hypothetical. They illustrate handling rules, not verified properties.
Ordinary correction: A caller gives “18 Mhlabeni Road, Example Gardens, Example Town”. The agent captures “Mhlaveni”. The caller spells the name, the agent reads back “Mhlabeni”, then confirms the complete address. Expected handling: staff receive the corrected address, the spelling confirmation and no unresolved address issue. The initial capture remains distinguishable from the corrected value.
Missing property number: The street name is confirmed, but the caller says, “I don’t know the number; it’s behind the shop.” Expected handling: keep the number missing, capture the directions and assign a follow-up. A person contacts the caller to establish an identifiable service destination under the business’s accepted address rules. The agent must not manufacture a number.
Ambiguous destination: A connected lookup returns hypothetical candidates in Example East and Example West. The caller cannot identify the locality. Expected handling: retain both candidates, ask for another distinguishing detail and route the unresolved case to staff. The human reviewer confirms the locality with the caller; the first lookup result does not win by default.
Duplicate enquiry: An existing hypothetical job contains “18 Mhlabeni Road”, while a new call gives “18 Mhlabeni Street”. Expected handling: flag a possible duplicate and preserve the discrepancy. Staff establish whether this is a correction or another destination before updating the job. Do not merge solely because the caller’s name matches.
FAQ: handling street-name confirmation calls
What if the caller says yes before the read-back is finished?
Treat that answer cautiously when it is unclear which detail they accepted. Finish the address read-back and ask for confirmation of the whole destination. If the caller interrupts with a correction, apply it and repeat the revised address. Keep the prompt brief so confirmation remains practical.
What if spelling and text confirmation disagree?
Record the disagreement and ask which version should be used for the visit. Do not automatically prefer text: it can contain a typing error or a different address. If the caller has left the call, assign a follow-up and keep dispatch preparation on hold until a person resolves the conflict.
Who handles the address when confirmation fails?
Assign a named role or monitored queue, rather than an unspecified “human”. Give that reviewer the captured wording, alternatives, missing fields and contact reference. Where n8n is used, its documented human-review mechanism can pause selected tool calls for approval or denial; the address process still needs configuring. Source: n8n human review for AI tools
Before deployment, check the chosen product’s current account, plan and region eligibility, supported connections and fallback behaviour. The custom AI agents workflow guide can help frame that implementation work.
If your business needs help connecting voice intake to a reviewable address record, explore AI automation and get in touch to discuss the confirmation steps and human handover your dispatch team needs.