Policies at the length an inspector expects, with the law rendered rather than written.
ELSA writes a complete policy, around six thousand words across twelve sections, in the shape a registered manager already recognises. The parts an inspector checks hardest are not written by the model at all: the CQC quality statements, the legislation and the further reading come from validated sources in code.
What this replaces today
- A policy bought from a template library three years ago that still names a regulation since replaced.
- A two-page document called a policy that will not survive being read next to one from an established compliance publisher.
- An incident log in a spreadsheet, a review record in an email, and no way to show a reviewer what changed between versions.
- Finding out at inspection that nobody has reviewed the safeguarding policy since it was written.
How it works
Say what the policy is for
Pick the topic and your service type. There is no length to choose: there is one depth, set at the measured median of thirty-eight published UK care policies, because asking a manager to guess how much policy a subject needs before seeing any of it is not a real question.
Four sections are written at once, and the rest assembled
The document is produced in parallel groups and reassembled in order. The regulatory sections are not drafted at all: they are rendered from validated catalogues of quality statements and legislation.
Review it, log it, share it
Policies carry a review schedule, reviews are recorded as structured entries, and the shareable copy is a link you can expire and revoke rather than an attachment you lose control of.
What you get
Twelve sections, in the order the sector reads them
Purpose, Scope, Objectives, Policy, Procedure, Definitions, Key Facts for professionals, Key Facts for people using the service, Notifications and statutory reporting, Training and competency, Monitoring and audit, Related policies and legislation, plus further reading and an appendix.
Every link comes from one place
The model is forbidden from writing URLs and any it writes are stripped. Every reference in a finished policy comes from a curated catalogue pointing at legislation.gov.uk, NICE and the ICO, so a citation cannot be invented and it stays right when a regulation is amended.
A review sheet on page one, built from the record
Version, dates, who reviewed it and what changed are assembled from the policy's own history rather than typed into the body, so it cannot go stale and nobody can write over it. Policies written before the sheet existed get a correct one with no backfill.
Over seventy templates, used as coverage maps
A template tells the writer what a topic must cover; it is never output as your policy. What you get is written for your service, not a find-and-replace of somebody else's document.
Appendix B and C as structured logs
Record-keeping entries and review amendments are real records with required fields, not free text at the bottom of a document. A submitted entry can still be amended and every amendment is audited, which is what makes writing up an incident over several days defensible.
Your brand, and readable in print
Section headings render as full-width bars in your own colour, with the text colour calculated for contrast rather than assumed, so a pale brand does not produce a heading nobody can read on paper. Exports to PDF, Word and Google Docs all carry your registered name and address on every page.
Staff see published policies, not drafts
The library belongs to the organisation. Anyone who can edit sees drafts and archives; everyone else sees what is published, because a carer following a draft as though it were approved is the risk worth designing against.
What stops it going wrong
The regulatory content is not the model to write
Quality statements and legislation come from validated whitelists, and anything the model offers that is not on them is dropped. This is the single most important design decision in the feature.
An AI review is never a sign-off
ELSA can read your library and record findings, but the entry is logged in the name of the person who ran it. Where the model fails on a policy it is skipped, never logged with a review nobody performed.
An edit cannot silently destroy structure
Clause numbers, tick boxes, table rows and headings are counted before and after every save. A change that would remove any of them is refused, with a sentence naming what would be lost.
A shared policy is frozen at the moment you shared it
The recipient sees the version you sent, not one that changed underneath them. Expired, revoked and never-existed all look the same from outside.
Who it is for
Unlimited on every paid plan
For any CQC-registered provider that has to hold, follow and evidence a policy set.
Policy generation is deliberately uncapped: no per-policy charge and no credit cost. Writing the policies you are required to hold should not be metered.