Answers built from your evidence, with the evidence named.
ELSA writes each answer against your company profile, your policies and the documents you have uploaded, and cites which document each claim came from. Where you have no evidence for something, it says so in the answer rather than producing a number that reads well and cannot be defended.
What this replaces today
- Rewriting the same safeguarding answer for the fourth time this year because nobody can find what you submitted in March.
- A bid written from memory that quietly overstates: the figure that felt about right, the certification that lapsed, the policy someone was sure existed.
- Reading a sixty-page specification to work out whether it is even worth bidding, and finding out on page fifty that you are ineligible.
- Discovering the week before submission that three required policies do not exist.
How it works
Score the opportunity before you spend time on it
Fit is scored across four weighted dimensions, each with a plain-English reason: service match, geography, public sector experience and financial capacity. A requirement you cannot meet is flagged as a blocker rather than quietly costing you points, because that is a go/no-go decision, not a lower score.
See what is missing before you write
The gap analysis is built by code from your policy library, your documents and your certifications, then narrated. The model can downgrade something to 'needs review', but it cannot invent a policy you hold. Anything genuinely missing gets a one-click route to generate it or raise it as a task.
Write, with the sources attached
Each answer is drafted against the retrieved evidence, respecting the word limit and the weighting. Citations point at real documents and are checked against what was actually retrieved. An invented reference is discarded before it reaches you.
What you get
Your own profile, rendered into every answer
CQC rating, staff numbers, coverage, turnover, insurance and accreditations are written into the prompt from your profile rather than left for the model to guess. Empty fields are left out entirely instead of appearing as 'not specified', which is what stops a half-finished profile making every answer read like a form.
The standard questionnaire, and seventeen topic libraries
The Standard Selection Questionnaire is carried in full against PPN 03/24 and the Public Contracts Regulations 2015, alongside libraries for health and safety, information security, GDPR, equality and diversity, modern slavery, insurance, financial standing, business continuity, safeguarding and more.
Thirty-seven questions ELSA already knows your answer to
Where a standard question is answerable from your company profile, the writer is told the answer is on file and where to find it: it is pointed at the record rather than handed the value, so there is never a second copy that can disagree with the first.
Social value, but only where it belongs
Your recorded social value reaches the writer on social value questions and nowhere else. Whether a question qualifies is decided before the prompt is built, because a model handed the material with 'use it if relevant' will work it in regardless. Delivered results and future commitments are kept apart: a commitment in a winning bid is something you are then held to.
Your house style, and a library that grows
Answers follow the writing style recorded for your organisation. Good answers can be saved to the content library and reused, so the fourth safeguarding answer starts from the one that won.
Quality, library and history beside the answer
The questions rail, the answer and the tools sit in one workspace, so you can move through a hundred-question questionnaire without losing your place. You can filter by status, jump to the unanswered ones, and see which evidence fed a draft weeks after it was written.
What stops it going wrong
A gap is named, never filled in
Where a figure is unavailable the answer makes the claim qualitatively and records what is missing. Placeholder text like '[Evidence Needed]' is rejected outright if the model tries it, so the only route for a gap is the one that surfaces it to you.
Citations are checked, not trusted
Every document reference is validated against what was actually retrieved for that answer. A reference the model invented is dropped rather than rendered. That matters, because a fabricated citation in a public sector bid is a far worse problem than a missing one.
No evidence is stated as no evidence
When nothing relevant is found, the prompt says so explicitly instead of leaving a silent gap for the model to fill. A blank space under an instruction to use the source documents is a reliable way to manufacture a confident invention.
The inventory cannot be flattered
The have/missing/review list is computed from your actual records first. The model is allowed to be more cautious than the code, never less.
Who it is for
Every paid plan, with the full tender suite included
For providers bidding for local authority and NHS contracts: domiciliary care, supported living, residential, complex care.
The quality of the output depends on your company profile being filled in: twelve fields drive the match score on their own, and the writer can only cite evidence you have actually uploaded.