Yearly running cost Assuming 12,000 conversations a year
By hand $30K / year 1,200 hours of work
Hosted service $10.8K – 12K+ / year
This workflow $576 / year Machine usage only; setup, hosting and review are extra. Where should it answer product questions? Start with an internal advice tool or connect it to your storefront. Both use your catalog and approved selling rules.
One conversation includes up to 5 assistant replies and 10 retrieved product records per reply.
An internal product advice tool Evidence-backed replies for your team to review $1,008 – 2,760 setup Connect to my storefront Product replies with current catalog checks and handover $1,656 – 4,560 setup
Which store should it connect to?Shopify WooCommerce Medusa Other No purchases, discounts or order changes. Hand over when the product evidence is missing or the request needs a person.
What is included Use a supplied catalog snapshot and approved product notes to prepare answers and alternatives for staff. No direct shopper messages or live availability claims.
The guide covers product questions and recommendations in your chosen language. It does not include checkout, discounts, customer-order lookup, behavioural tracking, voice or image analysis.
How reliable should the product advice be? Choose targets for supported answers, suitable products and the questions that must go to a person.
Standard Grounded advice with suitable products and clear handovers. Supported product claims: ≥98% Suitable recommendations: ≥95% Included Strict Extra checks for compatibility, missing facts and unsafe suggestions. Supported product claims: ≥99.5% Suitable recommendations: ≥98% Setup +$288 – 840 My own targets Agree your own acceptance requirements with the provider. Quote separately
How these standards are measured Hold out product families and shopper intents. Merchandisers label correct facts, hard constraints and when no suitable item exists. Unsupported safety advice, cross-customer leakage or a prohibited purchase action fail release.
Targets for your selected standard What is checked Target Supported product claimsEvery factual claim must be supported by the correct product variant and current approved source. Unverifiable claims count as incorrect. ≥98% Suitable recommendationsEvaluate each recommended item against the shopper's stated compatibility, size, budget and availability constraints; ignoring a hard constraint fails the item. ≥95% Answerable questions handledCorrect non-empty answers divided by all questions labelled answerable from approved sources. Always handing over cannot pass this measure. ≥90% Required handovers caughtIdentify missing evidence, contradictory sources, regulated advice and prohibited actions using a separately labelled challenge set. ≥98%
Test with staff before enabling shopper replies. Keep human handover and an immediate off switch. No promised conversion lift, sales volume or autonomous buying.
How fast should it process a conversation? Choose total AI processing time across the included replies. Time spent waiting for the shopper is not counted.
Within 60 seconds Included Within 40 seconds Setup +$144 – 480 Within 25 seconds Setup +$288 – 840
Timing details Sum retrieval, queueing, model calls, checks and retries across the bounded conversation. Measure on a fixed replay, excluding human typing and idle gaps. Report individual reply latency as well; a total-time pass must not hide a stalled reply.
The target applies to at least 95% of agreed test runs, with 2 in progress at a time.
How much do you want to spend per conversation? Choose the AI budget for answering questions, checking product fit and suggesting alternatives.
Up to $0.12 Included Up to $0.09 Setup +$72 – 360 Up to $0.07 Setup +$216 – 600
Cost details Includes the bounded conversation, all AI steps, retries and allocated hosting. Longer chats are additional units. Store, helpdesk and messaging subscriptions, human handovers and paid product feeds are separate.
Reference machine cost: $0.07 – 0.11 per conversation at 1000 conversations a month. The selected cap is a target to test, not a replacement for this estimate.
Where do you want it to run? Choose where catalog retrieval and shopper questions are processed, and whether an approved external model can receive them.
Run it onYour cloud Local environment AI model accessApproved model API No external calls
Data and access details Runs in a cloud account you control, with access controls and logs.
Only approved source fields go to the selected external model. Agree access and retention first.
Minimise retained shopper text, separate sessions and avoid order or payment data. Local-only inference can run the internal catalog tool. A public storefront connection cannot meet the no-external-calls boundary.
How do you want to use it? Use a test-and-review web page, connect your own AI agent or use a dedicated shopping assistant. A storefront widget is included only in the connected scope.
A review web page Test product questions, inspect evidence and review uncertain advice. Included My existing AI agent Let your existing AI agent ask for source-backed product suggestions. Setup +$72 – 240 A dedicated agent Use a dedicated product advice agent with the same catalog and handover rules. Setup +$144 – 480