Yes, you can design a chatbot to switch between English and isiZulu while preserving a quote, provided both replies use the same approved commercial facts. Translation should change how the quote is explained, while amounts, quantities, scope and conditions remain anchored to that record. Any unresolved difference needs human review before the customer receives a firm quotation.
For a business considering AI chatbots, the useful starting point is a bilingual quotation test. It should establish whether customers receive the same offer in either language, including when they switch languages halfway through the conversation.
Define what must stay the same
A quotation means more than its total. Two replies can show the same price while promising different work. One might include installation; another might imply delivery and installation. A translation can also weaken a condition, making an estimated completion date sound guaranteed.
Create one approved quotation record containing the reference, version, currency, quantities, unit prices, total, tax treatment as confirmed by the business, included work, exclusions, validity and relevant conditions. Keep missing values visibly unresolved. An empty delivery field must not become “delivery included”.
Separate fixed identifiers from wording that needs translation. Product codes and quote references should remain recognisable. Scope descriptions need equivalent meaning, although their sentence structure may differ. Dates should be displayed unambiguously in both languages.
The proposed rule is simple: switching language reuses the approved record. It does not authorise a discount, recalculate a charge or create a new quotation. A customer requesting different work starts a separate clarification and commercial review.
Separate quote facts from translated explanation
Build the proposed workflow around retrieval, presentation and checking. First retrieve the approved quote. Then display its verified amounts and identifiers through a controlled template. Finally generate or select the explanation in the customer’s preferred language, using reviewed terminology for commercial conditions.
OpenAI’s Structured Outputs documentation describes responses constrained to a supplied schema. This can organise fields such as quote reference, language and explanation. Schema adherence alone does not establish that the selected price or translated promise is correct; those need comparison with the approved record. Source: OpenAI Structured Outputs
When connecting a chatbot to quotation records, distinguish requesting information from deciding what may be sent. OpenAI’s function-calling guide describes a flow in which the model requests a tool and the application executes it. Your application must therefore implement the retrieval checks and sending rules; a tool request does not itself prove that the correct quote was found. Source: OpenAI function calling
That division helps when deciding between AI agents and automation: fixed comparisons suit ordinary automation, while conversational clarification may need language interpretation.
Review isiZulu wording for commercial meaning
Google’s Cloud Translation overview describes several translation options, including glossary support in its Advanced API. That establishes an available terminology mechanism, rather than proof that your chosen configuration supports every English–isiZulu requirement. Check the exact language pair, model, glossary compatibility and current account or region eligibility before deployment. Source: Google Cloud Translation overview
Prepare a terminology sheet for your own quotations. Include the English phrase, reviewer-approved isiZulu wording, intended meaning, context and phrases to avoid. Prioritise “excluded”, “subject to confirmation”, “estimated”, “per item” and “once-off”. These terms affect what the customer believes they are buying.
Use local bilingual speakers who understand the service. Ask one reviewer to check the translation against the approved offer, then ask another person to explain the offer back without seeing the English version. Compare their understanding of inclusions, exclusions and conditions.
A translation back into English can reveal possible discrepancies, but should be supporting evidence. It cannot replace checking whether an isiZulu-speaking customer understands the intended offer. Record the disputed phrase and revise that wording before repeating the test.
Work through ordinary and difficult cases
All figures and quote records below are hypothetical. The handling rules are proposed controls, not product defaults.
Ordinary language switch. Quote Q-104, version A, covers two replacement components at R600 each and R300 labour, totalling R1 500. The approved record excludes travel and says scheduling requires confirmation. The customer first asks in English, then requests isiZulu. Both replies should retain Q-104, version A, the same quantities and amounts, the travel exclusion and the scheduling condition. A bilingual reviewer checks whether the translated reply could suggest travel is included or an appointment is booked.
Missing information. The same hypothetical record has no confirmed tax treatment. The chatbot should not infer it from other quotes or add a familiar phrase. It should explain that the detail needs confirmation and pass the record to the quotation owner. That person supplies the approved information before a complete quote is sent in either language.
Ambiguous scope. A customer asks whether “fitting” includes removing an old component. The approved record does not define removal. The chatbot should surface that gap rather than translate “fitting” into a broader promise. A service adviser confirms whether removal is included, excluded or requires a revised quote, then approves equivalent wording for both languages.
Duplicate or conflicting records. A search returns Q-104 version A at R1 500 and version B at R1 700. Neither has a clear current approval status. The chatbot should not choose the cheaper amount or assume the latest timestamp settles the question. A quotation owner identifies the valid version, resolves the competing record and checks which version the customer already received.
Test those cases with language changes before and after the price appears. Include mixed-language questions and a request to return to English. Each switch should preserve the same resolved quote version.
Bilingual quotation test checklist
Use this checklist for each quotation template and each material change to terminology or the translation setup. These are proposed acceptance rules.
- Record the quote reference, approved version, test date, reviewer and translation configuration.
- Confirm the selected product, account and region currently support the required language pair and terminology approach.
- Identify the approved source for amounts, quantities, currency, tax treatment, scope, exclusions, validity and conditions.
- Check that English and isiZulu replies reproduce every fixed value from that source.
- Have a local bilingual reviewer confirm equivalent inclusions, exclusions and conditional wording.
- Ask a second speaker to explain what they would receive, pay and still need confirmed; compare this with the approved offer.
- Switch English → isiZulu → English within one conversation; confirm the quote reference and version stay unchanged.
- Test missing information; require clarification rather than an invented value or condition.
- Test ambiguous scope; require the quotation owner to resolve it before a firm reply.
- Test duplicate and conflicting versions; require an authoritative version decision.
- Confirm language changes cannot approve discounts, alter scope, record acceptance or trigger payment.
- Keep the source record, both replies, discrepancies and reviewer decision together. Release the tested template only after discrepancies are resolved.
Decide when the chatbot should hand over
A proposed handover rule should name the problem and the person who can resolve it. “Ask a human” is insufficient if the reviewer receives only the translated sentence. Send the quote reference, conflicting fields, relevant customer question and both language versions together.
Let the quotation owner decide commercial facts and the bilingual reviewer decide whether the explanation preserves meaning. If a tool will send the quote externally, place review before that action where your workflow requires it. n8n documents configurable human approval for selected AI tool calls; this is an implementation option, not an automatic safeguard in every chatbot. Source: n8n human review for AI tool calls
If these connections need tailoring, the custom AI agents workflow guide and custom AI agent definition help frame the work. Specify the quote source, checks and handover before choosing tools.
FAQ: English–isiZulu quotation handling
Should customers choose a language or should the chatbot detect it?
Offer an explicit choice and allow switching. Detection may inform a suggestion, but a mixed-language message should not silently change the quote. Retain the selected language until the customer requests another, and preserve the commercial record throughout.
Can we send a translated quote before a bilingual reviewer checks it?
For a new template, use review before external sending as a proposed starting rule. After approval, reuse the reviewed wording within its tested scope. New exclusions, unfamiliar terminology or changed conditions should return to review rather than inherit approval from an unrelated quote.
What should happen if the customer disputes the isiZulu wording?
Pause any step that depends on the disputed meaning. Preserve the reply they received and ask a bilingual service adviser to resolve the wording against the approved quote. If the offer changes, issue an approved revision and explain the change in the customer’s chosen language.
If your business needs help connecting quotation records to bilingual conversations, explore AI automation and get in touch with a sample quote and the wording your team needs to preserve.