Sopable AI
AI that reads your policies and checks your team understood them
Where our AI actually stands, in plain language
Plenty of software is described as AI-powered without much behind it. So here is a straight accounting.
Live and working today: AI Quiz generation. You can turn it on for any text-based document in your handbook right now, and this page shows you exactly what it produces.
In development: three features — AI SOP writing, AI Quick Onboarding, and an AI assistant that answers questions from your own documents. They are not available yet. We've described them below so you can see where the product is going, and we've labelled each one honestly.
If a feature isn't in the “Available now” section of this page, you can't use it yet. We'd rather tell you that here than let you find out on the call.
Part one
Available now
Everything in this half works today.
Available now
AI Quiz — comprehension you can actually measure
The problem it solves
An acknowledgment tells you a document was opened and scrolled through. That's worth something — it's a real record — but it isn't the same as knowing the technician can tell you the minimum clearance, the correct shut-off sequence, or what to do in the first ten minutes of an incident.
The traditional answer is a quiz, and the traditional quiz has two failure modes. Writing questions for forty policies is work nobody has time for, so it doesn't happen. And when it does happen, a fixed set of questions gets solved once and circulated — after which the quiz measures who has the answer key, not who knows the material.
Refrigerant Handling · Quiz results
8 / 10 · PassedQuestion 3 of 10 · Medium
According to our refrigerant recovery procedure, what is the maximum fill level for a recovery cylinder?
- 60% of rated capacity
- 80% of rated capacityCorrect
- 95% of rated capacity
- There is no stated limit
From the policy
“Never fill a recovery cylinder beyond 80% of its rated capacity. Confirm the fill level by weight, not by sight.”
Safety & PPE / Refrigerant Handling · v3 · § 4.2
Retake generates entirely new questions from this policy
How Sopable does it
- Sopable writes the quiz from your policy
- Turn the quiz on for a document and Sopable generates multiple-choice questions from that document's own text. Not a generic safety question bank — questions about your rules, in your wording, at the level of detail your policy actually specifies.
- New questions on every attempt
- There is no fixed answer sheet, because there is no fixed question set. A retake is a real second test. There is nothing to memorise and nothing to pass around the crew chat.
- Every answer shows the sentence that proves it
- When results appear, each correct answer is displayed alongside the verbatim line from your policy that establishes it. So when someone gets a question wrong, they don't just learn they were wrong — they read the exact rule, from your own document, at the moment they're paying most attention.
- You set the difficulty
- Choose the question count, the difficulty mix — Easy, Medium or Hard — and the pass mark per document. A short customer-greeting policy might be three easy questions. Your lockout/tagout procedure might be ten hard ones at a 90% pass mark.
What the employee experiences
Under an assigned policy they see a Check your understanding card showing the question count and pass mark. They start it, wait a few seconds while questions are generated, and work through them — jumping between questions using the numbered bar, or hitting Finish later to save progress and come back.
Results appear immediately, with the policy quote behind every answer. If they don't pass, they retake it with entirely new questions.
What you see as a manager
The Results view gives you attempts, average score, pass rate, and how many people haven't finished — plus a per-employee table where each person's full attempt history is one click away.
That table is where the value sits. A whole crew averaging 40% on the refrigerant policy isn't a training-completion problem, it's a knowledge gap with a location and a name attached — and you found it before an incident did.
At the library level, each folder card shows an “N of M quizzes on” counter, and you can filter every document by quiz status. Coverage is visible at a glance.
It stays current on its own
Because questions come from the current text of the document, editing a policy updates its quiz with no separate maintenance step. Employees who passed an older version can be asked to retake. You'll never have a quiz testing a rule you changed eight months ago — which is the normal failure state of hand-written quiz banks.
What AI Quiz can’t do
- It needs text
- A scanned form, an image-only document or a diagram has nothing to generate from. Sopable marks these
Not quizablerather than producing nonsense. - Short documents yield fewer questions
- If a policy doesn't contain enough material for the question count you set, Sopable tells you explicitly and asks you to lower it — rather than padding the quiz with filler questions that make attempts non-comparable.
- It tests what your policy says
- If your document is vague, the questions will be too. The quiz is a check on comprehension of your text, not a substitute for well-written procedures.
- It's multiple-choice
- It measures recall and understanding of stated rules. It does not assess hands-on skill. Nothing replaces a ride-along for that.
Send us one document. We’ll turn on a quiz and show you the questions it writes.
Three deliberate choices
- Generated from your source, not from a general model's assumptions
- Questions come from the text of your document. If your company requires 14-character passwords and the industry norm is 12, the quiz asks about 14 — because that's what your policy says. A general-purpose AI answering from its training data would get that wrong in a way that quietly undermines your own rules.
- Every answer cited
- Requiring each correct answer to be backed by a verbatim quote from the source document is what keeps the feature trustworthy. If the answer can't be traced to a line in your policy, it doesn't belong in the quiz. This is also the practical reason the feature works as training — the citation is the teaching moment.
- Fresh questions instead of a stored bank
- A stored question set is a static asset that decays: it drifts out of alignment with edited policies and it gets solved and shared. Generating on each attempt costs a few seconds of waiting and removes both problems permanently.
Part two
On the roadmap — not yet available
None of the three features below is available today.
What we’re building next
These three features are in development. None of them is available today. We’re describing them because our direction is a reasonable thing to evaluate us on, and because customers shape what we build and in what order.
Each item below shows its current status. When something ships, it moves into the Available Now section of this page and we email everyone who asked to be told.
In development
AI SOP writing — turn what you know into a documented procedure
The problem
Most home service companies don't have an SOP problem because they don't know their procedures — they have one because writing them down is a task that never reaches the top of anyone's list. The knowledge is in a senior tech's head, and it stays there until they leave.
What we’re building
Describe the procedure in plain language — or answer a short set of questions about how your company does the job — and Sopable drafts a structured SOP document: purpose, scope, step-by-step procedure, safety notes, and what to do when it goes wrong. You edit it, approve it, and it becomes a document in your handbook like any other, ready to assign, acknowledge, sign and quiz.
The point is not that AI knows your business better than you do. It doesn't. The point is that reviewing and correcting a solid draft takes twenty minutes, and starting from an empty page takes an afternoon nobody has.
Tell us what you'd generate first — Early input shapes the order we build in. Leave your email below
In development
AI Quick Onboarding — a new hire set up in one step
The problem
Sopable already makes onboarding setup fast: Pathways turn a dozen assignment decisions into one click, and Workflows trigger the task side of it. But somebody still has to build the pathway, build the checklists, and decide which combination a given role needs. For a company setting Sopable up for the first time, that's the real work.
What we’re building
Tell Sopable the role you're hiring — “install technician”, “dispatcher”, “lead plumber” — and it proposes the complete onboarding setup: which of your existing documents that role should read, a suggested pathway, and the checklists their first week needs, drawn from how your company has onboarded similar roles before.
You review it, adjust it, and apply it. What currently takes an ops manager an afternoon of setup becomes a review-and-approve step.
Get notified when this ships Leave your email below
In development
Ask your handbook — an AI assistant grounded in your own documents
The problem
A technician is in a customer's attic and needs to know the clearance requirement. The answer is in your install standards document — somewhere. They're not going to scroll a 40-page PDF on a phone in an attic. They'll call the office, or they'll guess.
What we’re building
A search-and-answer assistant that answers questions using only your company's documents. A tech asks “what's the minimum clearance for a horizontal unit?” and gets the answer, with a link to the exact policy and section it came from.
The grounding is the whole design. It doesn't answer from general knowledge about HVAC or plumbing — it answers from what your company has documented, and it shows you the source so the tech can verify it. When your documents don't cover the question, it says so rather than inventing an answer. That's the same principle already behind AI Quiz: the citation isn't decoration, it's the safety mechanism.
Get notified when this ships Leave your email below
Want to know when these ship?
Leave your email and we’ll tell you when each of these becomes available — and nothing else. No newsletter, no drip sequence.
How we think about AI in this product
Grounded in your documents, not in general knowledge
Every AI feature we ship or plan works from your company's own content. A model's general knowledge of the trades is a liability in this context, not an asset — it produces confident answers that contradict your own procedures.
Cited, so it's checkable
AI Quiz shows the policy line behind every correct answer. The assistant will cite its source. If you can't check where an answer came from, you can't rely on it for anything that matters.
A human approves anything consequential
Generated SOPs are drafts for review. Proposed onboarding setups are suggestions for approval. In work where a mistake means an injury or a failed inspection, automation that removes the human is the wrong design.
We say what's shipped
This page is split in half for that reason. It'll stay split as long as the roadmap has anything on it.
See the AI that works today
No obligation · We'll tell you if you're not a fit