This guide is for partners who maintain custom instructions on behalf of a client. When an AI Employee — the AI-powered assistant that handles your business's calls, chats, and bookings — reads a policy number, account number, order number, or address incorrectly, you can state the grouping and spelling you want it to use, then confirm the result in a test call. Those instructions live in the Workflow Builder, where you configure an AI Employee's conversation flows, or in the context you send with an outbound call.
Note: Standard phone-number scenarios need no extra setup. Newo prepares pronunciation formatting for the customer's phone number and the primary business phone number by default. Review custom instructions only when you run into a pronunciation problem.
Give the agent an exact spoken format
Keep the original value and add a pronunciation instruction alongside it. Split a long uninterrupted sequence into short groups separated by commas, which indicate brief pauses, instead of leaving the grouping entirely to the voice model.
Original value:
account number: 9300298546
The same value with explicit grouping:
account number: 9300298546 (pronounce it seriatim as: "930, 029, 85, 46")
Every digit keeps its original position, including the leading zero in the group 029. The commas describe where the agent pauses; they don't change the identifier.
Grouping and exact spoken wording are two different things. 930, 029, 85, 46 fixes where the pauses fall. It does not decide whether the first group is spoken as "nine three zero" or "nine hundred thirty". If the wording matters as much as the pauses, write the words you want. For individual digits inside each group, that would be:
account number: 9300298546 (say as: "nine three zero, zero two nine, eight five, four six")
Use this spelled-out version when you need the wording pinned down.
Match the rule to the type of number
Preserve the digits and speak them in the caller's language. Different kinds of number take different groupings:
Street number 8816: group as 88, 16 — "eighty-eight, sixteen" in English.
ZIP code 25356: speak each digit on its own as 2, 5, 3, 5, 6 — "two, five, three, five, six" in English.
"Seriatim" means "in sequence". State the intended wording rather than relying on the term.
Include pronunciation in outbound call context
When you send a number in outbound call context, include both the original value and a version that shows the intended pronunciation:
policy_number: 9300298546 policy_number_spelled_version: 930, 029, 85, 46
Alternatively, put the instruction beside the original value:
policy_number: 9300298546 (say as: "930, 029, 85, 46")
Spell letters and pause at separators
For identifiers that mix digits and letters, spell each letter with a clear word example, such as "D as in David" or "K as in Kilo". At spaces and hyphens, pause briefly and carry on, without saying "space" or "dash".
For example, a mixed identifier:
Original: 240 9991-D26-30 Say as: "240, 9991, D as in David, 26, 30."
An instruction in this shape covers all three cases for that value:
Always say the order number 240 9991-D26-30 aloud in full.
Say it as: "240, 9991, D as in David, 26, 30."
DIGITS: Pause briefly at each comma, and keep every digit in the order shown, including any zeros. LETTERS: Spell each letter phonetically using a clear word example, such as "D as in David" or "K as in Kilo". SPACES / HYPHENS: Pause briefly and continue. Never say the words "space" or "dash".
To reuse this example, replace both the identifier and its complete spoken form with the actual value, and write that spoken form out in full rather than listing the numeric groups on their own — the letter cue has to stay in its original position, which is between 9991 and 26 in this example. Preserve every digit and letter in order, including leading zeros.
Note that the spoken form still fixes grouping rather than word choice: "two four zero" and "two hundred forty" both match 240. Where that matters, write the words, then confirm the result on a call.
Handle repeated corrections
If a caller corrects a number more than once, use the latest correction and read the number back in parts. Use fallback wording like this:
Let me try to pronounce the number in parts. The first digits are ..., the next three digits are ..., the next three digits are ..., and the next four digits are .... Is that correct?
Fill the gaps with the digits from the caller's most recent correction. The three-digit and four-digit groups here are only an example, not a universal pattern — match the grouping language to the value you actually have, and never add, drop, or reorder digits to fit a pattern.
Check the result with a test call
Place a test call and run the affected scenario with the original value and your revised instruction.
Listen for missing or reordered digits, and confirm leading zeros are still spoken.
Confirm each letter is announced with its word example, in its original position within the value.
Confirm the agent pauses at spaces and hyphens and never says "space" or "dash".
Correct the number twice during the call and confirm the next readback uses your latest correction.
The test passes when the readback preserves the whole value in order and matches the spoken form you intended. If it doesn't, revise the instruction — tighten the grouping or write the words out — and test again.
Note: These are prompting recommendations and do not guarantee correct pronunciation. Follow the configuration process for your own account, and confirm behavior on a call before you rely on any example here.