Contract lifecycle management
Contract lifecycle management is usually sold as a repository, and a repository solves the easy half. The expensive half is that a clause you agreed to two years ago is now binding you to something, and nobody in the building knows which document it is in.
The two moments that cost money
Two moments cost real money. The first is a renewal that rolled over because the notice window closed on a Friday and the person who watched it left in March. The second is a question — can we do this, are we allowed, what did we promise — that takes somebody three days of reading to answer, so mostly it does not get asked.
Neither is a storage problem. The documents are usually all there. What is missing is anything that reads them and knows what it read.
What gets built, in the order that pays
- Dates out of the documents. Renewal, notice period, term, price review — extracted with the sentence they came from attached, so a date can be checked rather than trusted. This alone usually pays for the rest.
- A watch on those dates. The alert fires with enough time to act, to a named person, and it keeps firing until somebody says what was decided. An unanswered alert is not a decision.
- Answers with the clause attached. Ask a question in plain words, get the answer and the paragraph it rests on. Without the paragraph it is a guess with good grammar.
- A comparison against your own standard. Where this contract differs from the version your lawyers wrote, listed, so a review starts from the differences instead of from page one.
- An obligations list. What you actually committed to do — reporting, notice, service levels — pulled out of the prose and put somewhere a person can see it without reading the contract again.
What it must never decide
- Whether to sign. It reads, it flags, it compares. A signature is a decision with a name on it and it stays that way.
- That a clause is acceptable. It can tell you this differs from your standard wording. Whether the difference matters is a judgement about your business and your risk, and that is exactly the part you are paying a lawyer for.
- What a document does not contain. Proving an absence is the one thing this is worst at: a missing clause looks identical to a clause phrased unusually. It says what it found, and it does not claim the rest is not there.
- To act on a renewal by itself. It can raise the date, draft the notice and put it in front of somebody. Sending it is a commercial decision, and an automated notice sent in error is expensive in a way an ignored alert is not.
Under a hundred contracts, do nothing
- You have fewer than a hundred live contracts and one person knows them. A person with a spreadsheet is genuinely better at that size.
- The documents are photographs of signed paper, unindexed, in somebody's drive. That is fixable and it is the first job, but it is a different job and it should be priced as one.
- Nobody will own the alerts. A watch that fires into an empty inbox is worse than no watch, because it creates the feeling of coverage.
The reading half, already in production
- Returns & defects handler — documents read against written rules, at volume, with the decision and its reason recorded — the same shape of work on a different kind of paper.
- Data platform — the discipline of never handing over a number without showing where it came from. A contract answer without its clause attached fails in exactly the same way, on prose instead of on figures.
- Promotion guard — a rule held against a date with nobody watching a calendar, acting before the deadline rather than reporting after it. That is the renewal problem, already running on a different trade.
What a legal or commercial lead asks first
Is this the same as a CLM product?
It overlaps with one and it is not a replacement for one. If you need workflow, e-signature and an approval chain, buy a contract lifecycle management product — that category is mature. What gets built here is the reading layer: getting dates, obligations and differences out of documents you already have, including the ones that live outside whatever you bought.
How accurate is the extraction?
Good on structured, repeated documents and worse on bespoke ones, which is the honest answer for any extraction. The thing that makes it usable is not the accuracy number, it is that every extracted value carries the sentence it came from, so a person can check a doubtful one in seconds rather than re-reading the contract.
Can it answer questions about our contracts?
It can answer from them and show the clause. It should refuse when the documents do not settle the question, and that refusal is the feature — a system that always produces an answer is producing some answers it should not.
What about confidentiality?
The contracts stay in your systems and the extraction runs against them under your rules about what may leave the building. This is a decision to make explicitly before anything is built, not a setting to discover afterwards.
Where should we start?
The renewal dates. It is the narrowest slice, it is the one with a number attached — take the last renewal that rolled over by accident and price it — and it proves the reading works before anybody trusts it with a harder question.
Price the last one that rolled over
There is a faster test than counting anything. Ask your commercial lead one question — what did we promise about notice periods in the deal we signed last spring — and time how long the answer takes. If it comes back in a minute, you do not need this. If it takes three days, that is the whole argument, and you just measured it. The first conversation is an hour and it is free.
Book that hour →