ContractIQ reads your contracts, purchase orders, statements of work, trackers and meeting transcripts — then tells you what it found, how confident it is, and exactly where it read it.
Real contract estates are messy — scanned PDFs, spreadsheets nobody maintains, transcripts of the meeting where it was really decided. All of it is readable.
PDFs, Word documents, Excel and CSV trackers, purchase orders, statements of work and plain text — read in the browser, page positions preserved. Text is taken from the first 80 pages of a PDF and the first 8 sheets of a workbook, which covers an ordinary agreement and its schedules.
Photocopied and image-only PDFs used to be invisible to contract tools. ContractIQ detects them and reads them with on-device OCR — the scanned image never leaves your machine; only the text recovered from it is used. OCR covers the first 30 pages of a scan.
Ingest the meeting, not just the paperwork. Speaker attribution is preserved so a commitment can be traced to the person who made it.
Drop in a CSV of your estate and every record is created at once, with renewal exposure visible before a single analysis has run.
Each analysis produces a complete commercial picture, not a summary — costed, graded and evidenced.
Effective unit cost, uplift history, benchmark position and the specific lines where you are paying above the market.
ContractIQ extracts and grades every financial term: annual value, uplift schedules, payment terms, minimum commitments, currency and the date range. Each extraction shows confidence score and the exact line from the contract it came from.
Contracted volume against actual consumption. Dormant licences and unused modules are where the fastest savings usually sit.
An instant grade and a 0–100 score, so low-risk agreements can be waved through and attention goes where it earns its keep.
Each with an estimated value, the leverage behind it and the argument to use — ranked by what is realistically winnable.
Liability, indemnity, termination, data protection, audit rights, benchmarking, exit assistance and more — each marked present, missing or unclear with a suggested position.
The position in a paragraph a director can read, with the negotiation plan underneath it.
AI that cannot be checked cannot be trusted with a negotiation. Every value ContractIQ produces carries a confidence score, the reasoning behind it and a quote from the source document.
Not one score for the contract — a score for each extracted value, clause and risk, judged independently.
50% and below goes to mandatory review. 51–80% gets a one-click accept or correct. Above 80% passes automatically and is logged. Nothing uncertain reaches a decision unchecked.
Each finding cites the document and the exact phrase it came from — or states plainly that it was inferred, or absent. A reviewer can verify it in seconds.
A low-confidence extraction never silently becomes part of the contract record. It waits for a human.
Who accepted what, when, and what the value was before and after. Exported with the report.
Where a document contradicts itself, ContractIQ reports the conflict rather than picking a number. Where it truncated a long document, it says so.
Every contract tool reads contracts. Almost nothing captures what your people know that the paperwork never said — and that knowledge leaves when they do.
Concessions granted, precedents set, history that explains why a clause reads the way it does — extracted from transcripts and held against the contract.
Verbal commitments checked against the signed terms and flagged where they do not appear. Unenforceable as they stand — paper them or lose them.
Who holds undocumented context, across how many agreements. The top of that list is your succession risk, and it is usually a surprise.
Transcripts carry other people's personal data. Before the first upload, ContractIQ requires an explicit acknowledgement that participants were informed and a lawful basis existed — recorded against the user and the date.
Looking at agreements one at a time hides the expensive patterns. These views work across the whole estate.
Twelve months plotted by notice deadline — the date you must act by, not the date the contract ends. Windows already closed are flagged in red, with how long ago they passed.
Groups contracts by supplier and merges name variants automatically, so the same company bought twice under two legal names becomes one line. Near-miss names are flagged for review, and two suppliers serving the same category surface as a consolidation question.
Every analysed contract against every clause in one grid. Gaps down a column are your exposure on that agreement; gaps across a row are your negotiating pattern.
Every commitment from every contract in one list, with an owner, a due date and a completion state. Overdue and unassigned counts at the top, because an obligation nobody owns is an obligation nobody does.
One box searches record details, the full text of every ingested document, and every analysis finding — returning the surrounding passage, not just a file name.
Analysing forty contracts one at a time is the reason most tools get abandoned.
Select as many records as you like and run them in one pass, at up to 20 analyses an hour; anything beyond that queues for the next hour. The credit cost is shown before you start, and the run stops cleanly rather than failing halfway.
A conversational analyst on every contract. Renewal dates, missing clauses, obligations due this quarter, the strongest lever — answered only from the documents you have ingested, with the source shown and a confidence score on every extracted value. Up to 120 questions an hour.
Usage is measured in credits: ten for a full analysis, two for a question, nothing for storage, ingestion, alerts or exports. Re-running an analysis inside your revision window is free on paid plans.
ContractIQ banks your unused credits automatically, but caps them at one period's allowance. Growth plans cap at 500, Scale at 1,800, Enterprise at 5,000. This protects you from accidentally hoarding six months of credits and then running your entire estate through the system in one weekend — the cap exists to control cost exposure.
Notice deadlines exported straight to your calendar as an .ics file, so the date lives where you will actually see it. The reminder then fires from your own calendar; ContractIQ does not send emails or push notifications.
One click turns any analysis into a formatted Word report, a presentation outline, or a print to PDF through your browser's print dialogue — cost position, compliance checklist, obligations and negotiation plan included.
A signed agreement tells you what was agreed. It does not tell you how badly it would hurt if that supplier stopped tomorrow, or what they have actually committed to in writing. That is a different question, and it is the one procurement gets asked in the room.
Every supplier scored out of 100 from what is missing in the paperwork, how concentrated your spend is with them, the worst risk grade across their contracts, and how much you have not yet analysed.
Liability cap, insurance, data protection, security commitments, business continuity, exit assistance, audit rights and subcontracting — each marked present, missing or not yet assessed.
Every component that made the number is shown with its weight and its reasoning. An opaque risk score is not something you can take to a supplier meeting and defend.
A dash means the contract has not been analysed for that protection. A cross means it was looked for and is not there. Conflating those two is how false confidence gets built.
A blank chat box teaches nobody what a product can do, and free-form questions give different answers depending on how they are phrased. Ask Cedric actions are ready-made questions, grouped by what you are trying to find out.
Thirty questions across renewal, cost, risk, compliance, obligations, suppliers, institutional knowledge and cross-document checks. One click runs it against the contract you have open.
Every question is deliberately generic — no company names, no figures, nothing specific to you. Copy one and it works in any tool, and it leaks nothing about your estate.
Running a fixed question across forty contracts gives you forty comparable answers. Phrasing it slightly differently each time does not.
Sign-in is not a shared password. Everyone has their own verified account, and what they can do depends on what they are.
A code rather than a link, because corporate mail scanners routinely pre-fetch links and consume them before the recipient ever clicks. Codes do not have that problem.
Most people in a procurement team already have a Microsoft or Google identity. Let them use it rather than inventing another password.
A viewer can read the portfolio but cannot change a record or delete one. Enforced in the database itself, not just hidden in the interface.
Invite a colleague by email with the role they should have. The invitation is valid for seven days and only for the address it was sent to.
Analysis does not run in your browser tab waiting for you. Work is queued and handled in the background, so a large run finishes whether you are watching or not.
Select forty contracts, start the run, and shut the lid. The account runs up to 20 analyses an hour, so forty come back over two hours — the work carries on and the results are waiting when you return.
A model that is briefly busy is not a lost analysis. Failed work backs off and tries again, with an increasing gap so a struggling service gets room to recover.
Work that fails repeatedly is parked in a dead-letter queue with the reason attached, so somebody can look at why rather than wondering where it went.
Metering happens on the server, after the work completes. A failed analysis costs you nothing, and the balance cannot be edited from a browser.
The controls a procurement director, a DPO and an IT reviewer each need before this gets near a real contract.
Compliance rules are configuration, not code. Toggle UK statutory, EU GDPR, US state privacy or your own commercial guardrails, and the next analysis grades against them — no release, no redeployment. Packs are versioned, so a historic analysis can be explained against the rules in force at the time.
Calibrate the checking rules to your own risk appetite. Every clause is examined against your standard positions, so nobody accidentally signs unlimited liability or 90-day payment terms.
An Enterprise option for compliance-focused environments. Normally the text read from your documents is saved to your workspace so contracts can be searched and re-analysed. With Zero-Retention on, it is never saved at all: it stays in your browser tab for the analysis, and only the extracted detail and the analysis are kept. The database itself refuses to store it.
Isolation is enforced at the database level, not in application code — one customer's data cannot be reached from another's session.
Everything ContractIQ does is built on infrastructure designed for multi-tenant isolation, performance, and compliance.
React 18 with progressive enhancement. The app loads and displays content with JavaScript disabled. Motion and interactivity are progressive overlays, not prerequisites.
Supabase PostgreSQL (AWS eu-west-2). Row Level Security isolates tenants at the database layer. No shared access, no back-channel queries. Data is yours alone.
Edge Functions (Deno) process documents in parallel. Large PDFs stream in chunks. OCR runs on-demand for scanned contracts. Rate limits (20 analyses/hour) protect you and the system.
Stripe integration with Supabase webhooks. Credits are held in reserve, settled on completion. Unused credits roll over monthly, capped at one period's allowance. Bolt-on credits never expire.
It is an automated screening aid, not legal advice, and we are not a law firm. Its output is generated by artificial intelligence: it is probabilistic, it can be wrong, and it must be verified before you rely on it. The confidence scores above are the model's own estimate that a value is correct — they route how much human attention each finding gets. They are not a measured accuracy rate, and we do not publish one until it has been established against a labelled benchmark. We would rather tell you that than imply a precision we have not proven.