- Home
- AI development
- Expert systems
Expert systems
Expert systems: your rules written down and applied the same way every time
An expert system takes the rules your best person carries in their head, writes them down, and applies them to every case. A language model reads the messy input, such as a scanned form or a free-text request. A rules engine makes the decision, so the same case gets the same answer and every answer comes with its reasons.
- Prices published on the site
- First working version in two weeks
- Support month to month, cancel any time
Rules your team already follows, made explicit
An expert system is software that applies written rules to a case and shows which rule produced the answer. If an applicant's income is above a limit, the rule says so, names the limit, and the draft decision cites it. With the same facts and the same rule version, the rules engine gives the same answer every time.
A language model on its own is a poor judge here. Asked the same borderline question twice, it can answer differently and cannot reliably say why. So we split the work. The model reads the documents and fills in the facts. A rules engine, ordinary code over a table of rules, decides. When your policy changes, the owner of that rule edits the rule, and nobody has to rewrite a prompt.
From a pile of documents to a decision draft with its reasons
Comes in
The expert system
Reads, applies the rules, explains
- A language model reads the documents and fills in the facts
- A rules engine applies your written policy
- Each rule is versioned and owned by a named person
- The decision records which rule decided
- Anything outside the rules goes to a person
Lands in
The model never makes the decision. When it cannot read a field with confidence, the field is marked and the case goes to the exception queue.
Four decisions that suit written rules
Our own RightForms shows the rules half of this: guided questions produce a PDF packet ready for signature, and it refuses to continue when required data is missing. It is a document preparer, not a law firm, and gives no legal advice. See the RightForms case.
Eligibility checks
Applies your criteria to an application, such as income band, service area or a license on file, and returns eligible, not eligible or needs review, with the criteria that decided.
Quote and pricing rules
Turns a request into a quote from your own price list, discounts and exclusions, so two estimators price the same job the same way. An estimator approves every quote before it goes out.
Document completeness checks
Checks that a packet has every required page, signature and date before it moves on, and lists what is missing.
Compliance checklists
Walks a record against your checklist, such as a vendor onboarding list or an inspection form, and flags each item that fails with the rule behind it.
When written rules pay off, and when a checklist is enough
Worth building when
- You can write the rule on one page, and two of your people would apply it the same way
- The same decision is made dozens of times a week
- Someone has to explain afterward why a case went the way it did, to an auditor, a regulator or a customer
- The input is messy, such as scans and emails, but the decision itself follows rules
Start with a checklist when
- The rules change every week, because a shared checklist is cheaper to keep current than a rules table
- Nobody can write the rule down because it is pure judgment, so a checklist is the place to learn what the rule is
- The decision comes up a few times a month, and a person with a checklist keeps up
- A spreadsheet formula or a workflow rule in your CRM already does the job
Priced by the size of the rule set
One workflow running end to end, or one internal tool.
Several systems made to work as one: integrations, roles, reports.
Custom software built from scratch, over a longer schedule.
An expert system that reads documents, checks them against your rules and writes the result into your CRM or accounting system usually lands in connected systems. One decision with a short rule set can be one process. A rule set that spans departments, with several approval roles, is a full build.
Most of the work is writing the rules down with the person who owns them and testing them on 20 to 50 past cases. The first working version runs those rules on your data two weeks after access.
Care $300 to $500 a month. Active $750 to $1,500 a month. AI model usage, GPU time and hardware are paid to the provider, on accounts in your name.
Questions about expert systems
What is the difference between an expert system and a language model?
An expert system decides by written rules, and a language model produces text. We use the model to read the input and the rules to decide. The same case gets the same answer every time, and the answer cites the rule. A model alone can answer the same question differently on a different day.
Who writes the rules?
You and we do, together. We interview the person who makes the decision today, write the rules in plain language and test them on 20 to 50 past cases. Each rule gets a named owner and a version number. When policy changes, the owner edits the rule and the version goes up.
What happens to a case the rules do not cover?
It goes to a person with the reason it fell outside the rules. We count these cases. A kind of case that keeps showing up in the exception queue is a candidate for a new rule, which its owner approves.
Does the system make the final decision?
Every decision starts as a draft a person approves. For clear, low-risk steps, such as flagging a missing signature, you can later let the system act on its own. Decisions about a person's job, pay, credit, benefits or legal status stay with a person.
Where does our data go?
The rules engine and the audit trail can run on your own server or cloud account. We tell you which model it is and what data it receives before we build, and the account with that provider is opened in your company's name.
Related services
- AI agents Agents that finish a task, with your sign-off
- Language processing (NLP) Sort, extract and summarize text
- LLM evaluation Test which model does your job best
- AI workflow automation Documents, approvals and multi-step processes
- Custom software Internal tools, portals and integrations
- AI development Every AI service we build, on one page
Tell us what's slowing you down.
You get a free written review by email within two business days.