The expert system
It scores the tender like a panel - then builds your bid with you
The hardest question buyers ask us is a fair one: why trust an AI with a bid? Here is the honest answer. BidSquirrel is built to assess like a UK procurement panel, and it does not stop at a score - it drafts your response from your own evidence, marks its own work against the scoresheet, strips out the fluff, and hands it back with the exact facts for you to confirm. Never invented, always yours.
What you actually get
A scored verdict with its reasoning on every tender - and, when you choose one, a full draft built answer by answer from your evidence, a costed price band with the working shown, and the facts you still need to confirm called out. The finished words, and the number, stay yours.
Daily commercial cleaning is your core self-delivered service, and the CPV codes map precisely to your proven £540k public win for a Greater Manchester multi-academy trust...
We will mobilise across the head office and both depots within four weeks, led by your named contract manager, with a TUPE plan that preserves the incumbent team's terms…
CopyWe will recruit one local apprentice and support them to a completed qualification within the term - additional to your usual intake - and report the outcome quarterly against the buyer's social value criteria, drawing on your banked commitments…
CopyWe never invent facts - fill these in before you submit.
- Exact mobilisation lead time for all three sites
- Named contract manager for this bid
- Public liability insurance
- ISO 9001 certificate
- Tameside MAT case study
Quality carries 70% of the score here. Price at market from your own cost build-up and spend the effort on quality and social value - a big price cut buys few marks on this split.
Under this split (price worth 30% of the score), a 10% price cut moves at most ~30 of 100 marks - and a relative formula usually recovers far fewer.
Commit to 2 local apprenticeships and 40 supervised training hours in the contract area within year one.
Commit to paying all sub-contractors within 30 days, evidenced by your published payment record.
Grounded in PA23 s.19 and GOV.UK MAT guidance - we advise where to sit and what to emphasise, never a target price or undercut. You own the number. Source: legislation.gov.uk - checked Jul 2026.
We never invent numbers - replace each estimate with your own figure.
Example. Real output is scored to your own service line and drafted against the tender's ITT.
How the bid build works
Seven steps, every time you build a bid
This is the part a general AI tool skips. We do not draft and hope - the system runs your answer through the same loop a good bid team would, before you ever see it.
- 01
It reads the actual tender
First it classifies the regime - Procurement Act 2023 above or below threshold, light-touch, NHS Provider Selection Regime, or the Scottish and Welsh rules - because what scores changes with the rulebook. Upload the ITT and it works from that tender’s real questions, weightings and word limits, not generic best-practice.
- 02
It recalls your evidence
It drafts from your Evidence Vault - your banked standard answers, certifications, insurances, past projects and wins - and your business profile. Uploaded documents are text-extracted in the background so claims can show support status against real passages when ready (images stay filename-only). It writes from what is genuinely yours, so the answer carries your track record, not a stranger’s.
- 03
It drafts every answer
Section by section, grounded in your own evidence, criterion-responsive and front-loaded the way a panel rewards. Where a fact would strengthen an answer but is not in your evidence, it does NOT invent it - it leaves a clearly marked [you confirm: ...] flag and adds it to your to-do list.
- 04
It scores its own draft like the panel
It runs the whole draft back through our adversarial Bid Checker - the same engine that scores live tenders - predicting a band for each criterion and pinpointing exactly where the answer would lose marks.
- 05
It refines the weak sections
It rewrites the answers that fall short against those gaps - still only from your evidence, never by inventing - so the draft gets stronger where it matters most before you ever see it.
- 06
It pressure-tests the whole draft
A sceptical bid-director pass hunts down generic, AI-sounding prose, challenges any claim your evidence does not support, and pulls answers back on-question. Machine-smooth filler is exactly what evaluators mark down - this strips it out.
- 07
It hands it back to you - with a price
You get the finished draft, the exact list of facts to confirm, and the supporting documents the ITT asks for - cross-checked against what is already in your vault. Alongside it: a costed price breakdown from the scope, a recommended price band, a what-if against the tender's scoring formula and the value-for-money case, read against recent comparable awards - never a single oracle number. You confirm, you finish, you price, you submit. The words, the number and the final call are yours.
The part that compounds
Bank it once. Win faster every time after.
The insurances, accreditations, case studies and standard SQ / PSQ answers you otherwise rewrite from scratch on every bid - bank them once in your private Evidence Vault. After that, every build draws on them, so each draft starts further ahead than the last and comes together faster. Your banked evidence also feeds the matching, so the more you put in, the sharper your scores get. It is the part that compounds: the more you bid, the more the system has to win with.
- Public liability insurance.pdfInsurance0.4 MB
- ISO 9001 certificate.pdfAccreditation0.2 MB
- Tameside MAT cleaning - case study.docxCase study0.3 MB
- Financial standing statementFinancial
- Modern slavery statementPolicy
- Social value commitmentsSocial value
Banked once - reused on every future bid, so each one is faster than the last.
Why a general AI tool can't do this
- 01
It knows which rulebook applies
A general chatbot has no idea which regime governs a tender, so it cannot know what a strong answer looks like. Ours classifies it first and judges accordingly.
- 02
It reads the tender through a sector-experienced lens
This is separate from the rulebook - the procurement regime is the law; the sector is the trade. For the five sectors we cover deepest - Construction, Cleaning & Facilities Management, Healthcare & Care, IT & Digital, and Security Services - it applies the standards, accreditations and evaluation patterns a specialist bid lead in that field would recognise, and raises the questions an experienced practitioner would ask. It works from named, current standards - CDM Regulations 2015 and the Common Assessment Standard in construction, SFG20 and BICSc in cleaning and FM, CQC Regulation 12 and the DSPT in healthcare, Cyber Essentials and WCAG 2.2 AA in IT, SIA licensing and BS 7858 vetting in security - each verified against its primary source and re-checked monthly, so it never leans on a lapsed one. A general chatbot has read about these but cannot tell which apply here, and fills the gap with plausible-sounding standards that do not exist. Outside those five the same discipline applies - the same regime classification, the same evaluation doctrine, the same refusal to name a standard it cannot verify. What differs is the sector itself: construction, cleaning and FM, healthcare, IT and security are unusually CODIFIED, with named, current, checkable standards to hold a bid against. Where a sector carries fewer of those, there are fewer to cite - which is a fact about the sector, not about the care we take with it.
- 03
It marks its own work
It does not just write and hope. It scores its draft against the criteria like the panel will, then fixes the weak spots - and every score it gives you carries its reasoning. Each scoring change is regression-tested against a private library of expert-built bid scenarios before it ships; if it makes the judgement worse on any, it does not ship.
- 04
It refuses to invent
The hard rule: it drafts only from the evidence you give it. Anything it cannot evidence becomes a [you confirm] flag, never a confident guess. A general tool fills gaps with plausible fiction - which is exactly what gets a bid discredited.
- 05
It helps you price to win
Price and value-for-money can be 20 to 70% of a public score, and a general chatbot will not build a defensible costed breakdown from the scope, model it against the tender's own scoring formula, or read it against recent comparable award values. Ours does - a line-item cost build-up on your rates or labelled estimates, a recommended band trading win-probability against margin, and the value-for-money case. Never a single oracle number, never a claimed market rate: you own every figure.
- 06
You stay in control
It is decision support, not a decision. You see the reasoning behind every score, you confirm every flagged fact, you own the final words, and you always verify against the official notice. You make the call.
Where we draw the line on transparency
We'll always show you what the system concluded and why - every score carries its reasoning, and every drafted claim traces to your evidence or a [you confirm] flag. We don't publish the internal scoring rubric, the prompts, or the library of expert test scenarios. That calibration is months of procurement expertise, and it's precisely what a copycat would need. Keeping it under the hood is what lets us keep the judgement sharp and honest for you.
Questions, answered
Is this just ChatGPT with a logo?
No. A general AI tool has read the internet but has never sat on an evaluation panel, and it fills gaps with plausible-sounding fiction. BidSquirrel applies the correct regime to each tender, drafts only from your own evidence, scores its own draft against the scoresheet, strips out the generic prose a panel would mark down, and flags every fact it cannot evidence rather than inventing it. It is pointed at a different job - building a specific, evidenced bid with you, not churning out confident filler.
Will it write my bid for me?
It builds the full draft with you, from your evidence - answer by answer, scored against the criteria and stress-tested - then hands it back with the exact facts for you to confirm and the final words for you to own. What it will not do is invent a claim, a figure or a referee, because unevidenced, machine-smooth prose is precisely what evaluators mark down and what buries the track record, named people and local knowledge that actually win.
How does it build the draft without making things up?
The hard rule built into the writer: it drafts only from the evidence you give it - your vault answers, your profile, and the tender’s own ITT. If a fact would help but is not there, it leaves a [you confirm: ...] placeholder and lists it for you, rather than filling it in. It then scores and stress-tests its own draft to strip anything unsupported. Every output shows its reasoning, and you should always verify an opportunity against the official notice before bidding.
Does it help me price the bid?
Yes, and it is an axis few bid-help tools cover - price and value-for-money can be 20 to 70% of a public score. It reads the scope and builds a costed price breakdown line by line, on your own rates or clearly-labelled estimates you confirm; recommends a price band that trades win-probability against margin rather than a single number; lets you what-if the price against the tender's own scoring formula; and writes the value-for-money case for the most-advantageous bid, not just the cheapest. It reads recent comparable award values to sense-check the band, but presents them as comparables, never a precise market rate. You price to win, with the working shown - and you own every figure.
What is the Evidence Vault?
A private, secure store for the things you reuse on every bid - your insurances, accreditations, case studies, and the standard SQ / PSQ answers you otherwise rewrite from scratch each time. You bank them once; every future build draws on them, so each bid starts further ahead than the last. Your documents are used only to produce your own results.
Do you use my bids or documents to train your AI?
No. Your documents are used only to produce your own results. We do not sell your data and we do not use your bids or uploads to train models. See our Privacy Policy for the detail.
Why won’t you publish exactly how the scoring works?
We are open about how it behaves and how we keep it honest - regime-aware, evaluator-calibrated, regression-tested, drafting from your evidence and stress-testing its own work. We do not publish the internal scoring rubric, the prompts or the test scenarios, because that calibration is the work that makes the judgement trustworthy. What matters to you is the result and the reasoning behind it, and both are shown on every output.
See it score - and build - your next bid.
The expert system, on every tender that matches your business. £49.99/mo. How it compares to a bid writer →
Get started free