For a small or mid-sized exporter, is a monthly subscription or a one-off project more common for a multilingual AI inquiry system, and what budget is reasonable?
Subscriptions (monthly or annual) are more common for small and mid-sized exporters. Project-based builds are typical of larger companies that need ERP integration or private deployment. In between sits done-for-you: a build plus an annual service. Public vendor selection guides in China (2025, for reference only) put SME SaaS at about RMB 5,000–30,000/year, advanced SaaS at RMB 100,000–300,000/year, and private or custom builds at RMB 100,000–500,000+. A practical check is to compare the cost with one sales rep’s annual cost.
What are subscription, project and done-for-you models?
Subscription: pay monthly or yearly for software
You open a SaaS account and pay by seat, tier or usage. It is quick and cheap to start, but data prep, scripts and daily tuning are yours. For example, SaleSmartly lists Pro from about $15.9/month and Max from about $199/month (vendor site).
Project: one-off development plus annual maintenance
A vendor builds a system to your spec, often with ERP integration or on-premise data. Upfront costs are high—about RMB 100,000–500,000+ plus 10%–20% yearly maintenance (public vendor guides, 2025, reference only)—and changes usually go through change requests.
Done-for-you: build plus annual service
The vendor builds the system from your product data and quoting rules and keeps tuning it. You buy a service, not just software. It suits factories without an in-house AI operator that need several languages and channels covered.
How do the three compare?
| Subscription SaaS | Project | Done-for-you | |
|---|---|---|---|
| Upfront | Low | High, long build | Medium, build included or itemized |
| Ongoing | Monthly/annual, grows with seats or usage | ~10%–20% maintenance/yr (ref.) | Annual service by scope |
| Who maintains | You | Vendor fixes, you operate | Vendor |
| Who updates data | You | Change requests | Vendor, you confirm |
| Exit cost | Low, if data can be exported | High | Medium; set data ownership in the contract |
| Best for | Small, hands-on teams with few inquiries | Larger firms needing ERP or private deployment | Factories with no AI operator and many channels |
None of these include WhatsApp message fees. Meta has charged per message since 2025-07-01 (Meta developer docs), and several BSPs including respond.io and Gupshup disclosed in September 2026 that service messages inside the window are charged from 2026-10-01, with the first 1,000 per number per month free—check Meta’s official page. Budget this separately.
Why do most small exporters choose subscriptions?
Low upfront cost and easy exit. The hidden cost is that the subscription buys only the tool; someone on your team must prepare data and review conversations every week. Before subscribing, decide who owns it and how many hours a week they can give.
When does a project make more sense?
When the need is system integration—real-time inventory and cost in quotes, on-premise data rules, an IT team to own it. For most small exporters, it is wiser to start by catching inquiries first.
How should you set a reasonable budget?
Five factors decide the tier you need: inquiry volume, number of languages, SKU count, number of channels, and whether you need rule-based quote drafts.
- Tier 1, SME SaaS, ~RMB 5k–30k/year: few inquiries and languages; the AI only clarifies needs before handing off.
- Tier 2, advanced SaaS, ~RMB 100k–300k/year: many languages and channels, complex specs, handoff rules. Done-for-you is quoted by scope and not in public ranges; compare it here, adding the cost of your own staff time for SaaS.
- Tier 3, private or custom, ~RMB 100k–500k+ plus 10%–20%/yr: ERP integration or on-premise deployment.
Ranges are from public vendor guides (2025), not industry statistics. For a line-by-line cost breakdown, see what it costs to build an AI Sales Agent.
Compare it with one sales rep
An experienced export rep in China costs about RMB 120,000–200,000 a year plus commission and social insurance, covering one time zone for eight hours a day (common industry level). The question is not whether AI replaces that person, but who answers after-hours inquiries now, how much rep time goes to repeat questions on specs, MOQ and lead time, and how many more serious buyers reps could follow up if those were handled first. Speed has value: HBR (2011) found firms contacting leads within an hour were nearly 7 times as likely to qualify them; the average first response was 42 hours and 23% never replied.
What should the contract spell out?
- Deliverables: channels, SKUs, FAQ entries, handoff rules.
- Acceptance criteria: which test questions, what pass rate, who signs off.
- Data updates: who changes products, prices and scripts, how fast, at what cost.
- Data export: chats, customer data and knowledge base, format and ownership.
- Renewal pricing: year-two price, caps on increases, volume surcharges.
- Refunds: what happens if acceptance fails or service stops.
Also include handoff requirements in line with China’s recommended standard GB/T 47746—2026 (history passed to humans, no repeating), and human approval for quotes, since Moffatt v. Air Canada (2024) held a company responsible for its chatbot’s statements.
Which model is DeepSales?
DeepSales is done-for-you: an annual service priced by scope (SKUs, languages, channels, quote drafts), covering build, delivery and operation, not per seat; prices are not published. See how DeepSales is priced. Our team builds it from your materials and best reps’ conversations and keeps calibrating it. It runs on your own website chat, WhatsApp, Messenger, Instagram and LINE accounts; email inquiries can be viewed in the platform but are handled by people. Quote drafts need human approval, and negotiation stays with people. If you get few inquiries and want to experiment, a monthly SaaS fits better; if you need private deployment with ERP, choose a project. Otherwise, book a demo and we will run a sample conversation on your catalog and quote by scope.