How to estimate product search API costs for a prototype
Estimating product-search API costs for a prototype
A simple way to estimate product-search costs before you build: count calls per conversation, multiply by expected volume, then cut waste with caching and better first queries. Includes a worked example using Simlir's live credit prices.
Before building a shopping assistant prototype, most teams want a rough answer to one question: will product search cost tens of pounds a month or hundreds? You can get a decent estimate in ten minutes by counting the calls in a typical conversation and multiplying by expected volume. This guide walks through the method with a worked example, then covers the habits that keep the number down.
Step 1: list the calls in a typical conversation
Write out a realistic conversation and note every product-data call it triggers. Most product-search APIs price per call, and the type of call matters. On Simlir, the costs are:
Finds products from a photo, with an optional text query
2
Exact lookup
Fetches a product by identifier such as a GTIN
1
Product by UUID
Fetches one product you have already found
1
Take a headphone conversation. The shopper asks for noise-cancelling headphones under £200 (one search). They then ask for something lighter (a second search). They pick one and ask for its full details (one product fetch). That conversation uses 3 credits.
Step 2: average across conversation types
Not every conversation looks the same. Group them into two or three types and estimate the mix. For our example assistant:
Average per conversation: (0.8 × 3) + (0.2 × 4) = 2.4 + 0.8 = 3.2 credits.
Step 3: multiply by volume
Now estimate conversations for each stage of the prototype.
Stage
Conversations
Credits (× 3.2)
Plan that covers it
Build and test
500 test conversations
1,600
Free: 10,000 welcome credits
Friends-and-family trial
1,500 conversations
4,800
Still within the free welcome credits if testing used 1,600
Pilot month
100 a day × 30 days = 3,000
9,600
Developer: £39/month with 10,000 credits per billing period
Testing plus the trial uses 1,600 + 4,800 = 6,400 credits, leaving 3,600 of the free 10,000. The pilot month at 9,600 credits fits inside the Developer plan's 10,000.
If usage is uneven, credit packs are the alternative: from £12 per 1,000, £80 for 10,000, £500 for 100,000 and £1,500 for 500,000. Credits roll over and do not expire, so a pack bought for a pilot is not wasted if the pilot runs quieter than expected. For a steady pilot of about 10,000 credits a month, the Developer plan at £39 costs less than a £80 pack. Check the pricing page for current details before you commit.
Step 4: cut wasteful calls
Estimates often come in higher than they need to be because prototypes call the API more than necessary. These habits help without making the assistant worse.
Cache within a conversationIf the shopper asks a follow-up about products already shown, answer from the records you have instead of searching again.
Fetch details only on demandSearch results already include specs, images and prices. Fetch a single product only when you need something fresh or missing.
Ask before searching vague requests"A present for my sister" will likely need a second search. One clarifying question first often saves a call.
Avoid speculative pagesEach extra page of results is another search. Fetch more only when the shopper asks for more.
Use exact lookup for known identifiersIf the shopper gives a model number or barcode, a lookup is more direct than a text search.
Prefer text when it is enoughAn image search costs 2 credits. If the shopper has described the item well in words, start with a 1-credit text search.
Stop runaway loopsCap the number of tool calls per turn so a confused assistant cannot search ten times for one question.
Log credits per conversation from the first day of testing. Real numbers will replace your guesses quickly, and a sudden rise usually points to a loop or a missing cache.
Other limits worth knowing
Cost is not the only constraint. The free plan has a rate limit of 60 requests per minute, which is plenty for testing but worth knowing if you run load tests. Model, hosting and other infrastructure costs are separate, and for many prototypes they are larger than product-data costs. For how the calls fit into an assistant, see building product search for a shopping assistant, and for choosing an access method, MCP or REST for product data. Endpoint details are in the docs.
Frequently asked questions
How much does a product search API cost for a prototype?+−
It depends on how many calls each conversation makes and how many conversations you expect. In the worked example, testing and a small trial fit within 10,000 free welcome credits, and a pilot month of 3,000 conversations fits the £39 Developer plan.
How many conversations do 10,000 free credits cover?+−
It depends on your flow. At 3.2 credits per conversation, roughly 3,100. At 3 credits, about 3,300. Count your own calls to get a closer figure.
Do unused credits expire?+−
Credits roll over and do not expire, according to the current pricing page.
Is an image search always worth the extra credit?+−
When the shopper only has a photo, yes. When they can describe the product in words, a text search is cheaper and often enough.
