Teach QuoteKernel how you make pricing decisions.
You know your business. QuoteKernel turns that knowledge into firm rules that set the boundaries and domain knowledge it reasons with to estimate and recommend, all testable, for any connected agent to use.
What you're teaching
QuoteKernel is hybrid AI reasoning for dependable quoting: probabilistic reasoning inside deterministic boundaries. Teaching it means giving each of its four parts what it needs.
Your firm rules: rates, minimums, constraints, eligibility and approvals. Applied exactly, every time.
How your industry and your business work: the questions you ask and the patterns in your past quotes.
How you read a vague request, how much work a job takes and which package you would recommend, learned from your examples and corrections.
Where it must stop: what needs a person, how confident it must be to answer, and what gets checked.
How you teach it
- 01
Start with an industry model.
Choose the work you perform. The kernel begins with relevant inputs, pricing factors and common exceptions, so you are not starting from a blank page.
- 02
Show us how you quote.
Enter your services, packages, rates and rules, and the questions you ask before you quote, or pull them from price sheets, example estimates and past invoices. You review what was picked up before anything is used.
- 03
Practice with scenarios.
"Six-hour wedding, 175 guests, 40 miles away. What would you charge?" Review the result. Say "correct", "too high", "too low" or "needs review", then explain why. You can also run jobs you already know the price of and check the result.
- 04
Validate the edge cases.
Test missing information, unusual jobs, discount limits, travel charges and jobs that should go to a person. A kernel is ready only when it behaves within the boundaries you set.
What you give it
The services
Everything the business offers that a customer might ask a price for. Wedding coverage. Deep clean. Two-bedroom move. Tap replacement.
The prices and the rules
What each service costs, and what changes the price:
- the base price, and the unit it's priced in: per hour, per room, per square foot, per item,
- price bands by size or quantity,
- extras that add a flat amount or a percentage,
- the minimum charge,
- tax and how prices are rounded,
- discount codes, with how much they take off, any cap, when they expire and which services they apply to.
The questions
The questions the business asks a customer before it quotes, in the business's own words. For each one: what kind of answer it expects, the allowed values or range, whether it's required, and when to ask it. For example, "ask about a second photographer only above a set guest count".
What it won't price
Where pricing should stop and a person should take over:
- areas the business doesn't serve,
- services it doesn't offer,
- dates it's closed or fully booked,
- jobs that must wait for a person's approval.
QuoteKernel checks everything you enter and reports every problem at once: a negative price, price bands that overlap or leave a gap, a question that refers to something that doesn't exist, an extra attached to a service the business doesn't offer.
Not a model that memorized your invoices. Reasoning you can rely on, inside the rules you set.
When your kernel is ready, you receive:
A kernel any connected agent can call
over MCP or API. Connect it yourself or through your agency or developer.
A check report
Every problem in the pricing, listed with where it is.
-
Rules you can read
The exact amounts come from rules you can read, with a line-by-line breakdown. Where a job is ambiguous, you see the estimated range or recommendation and the confidence behind it.
-
A version you can roll back.
Change a price and you get a new version. Check what would change before it goes live. If something's wrong, go back. Quotes already given keep the version they were priced on.
Keeping it current
Prices change. When they do, update the kernel. QuoteKernel checks the new version the same way before it goes live, so a typo can't turn a $200 job into a $2,000 one without someone noticing.