Build or buy, with a contract management system as the example
Buy when the job's the same in every business and someone already sells it. Build when the job is yours alone, or when the bought tool would have to be bent so far it breaks. For contract management, most businesses should buy the filing and the workflow and build only the reading.

The three questions
| Question | Points to buy | Points to build |
|---|---|---|
| Is the job the same everywhere? | Filing, approvals, reminders, signatures, reporting. | Your rules, your appetite for risk, your way of deciding. |
| How many businesses have it? | Hundreds or thousands. A market exists and the product is better than yours would be. | A handful. Nobody's built it well, and you'd be the vendor's only serious user. |
| How far would you have to bend it? | Configuration the vendor supports. | Changes the vendor doesn't allow, or that break at every update. |
If two of the three point the same way, that's your answer. If they split, it's usually a bought core with a thin built layer for the part that's yours.
Contract management, part by part
We can't say how the team that was asked this decided – we weren't there. This is how we'd work it through.
| Part of the job | Buy or build | Why |
|---|---|---|
| Storing contracts, versions, who signed what | Buy | Identical in every business, plenty of good products, and dangerous to get wrong. |
| Approval workflow and signatures | Buy | Same. |
| Renewal dates and reminders | Buy | Same, and the bought tools already send the reminders. |
| Reading a new contract for the clauses that matter to you | Build, as a layer | Which clauses matter, and what counts as acceptable, is particular to you. Getting AI to flag problem clauses covers how. |
| Summarising a contract for the account manager | Build, as a layer | Your fields, your language, your CRM. |
| Reporting across contracts | Buy first, build only if the bought reports can't answer your questions | Usually they can. |
The built layer reads from the bought system and writes a summary and a set of flags back into it, or into the CRM. It doesn't try to be the system of record. That's the shape that lasts – the vendor keeps the plumbing working and you keep the judgement.
The two traps
The build trap: a capable person builds the filing and the workflow because they can, and the business ends up owning a contract system nobody else understands, with no vendor to call when that person leaves. The buy trap: a product gets bought for the filing, then bent to do the reading with configuration it was never meant for, and every vendor update breaks it. The same rule avoids both. Buy the common part, build the particular part, and keep the join thin.
ERP integration
Yes, think about it, and then mostly don't do it yet. There are two places where a contract system and an ERP genuinely need to agree. The customer and supplier records, so both systems mean the same company. And the commitments, so a signed contract becomes the right thing in the ERP (a supplier record, a payment term, a purchase order limit). Integrate at those two points, through the ERP's own supported connection, with a person approving anything that creates a financial commitment.
What shouldn't be integrated is the reading layer. An AI summary of a contract has no business writing into a finance system. Keep it in the contract system and the CRM, where a person reads it, and let the person make the ERP change. And do none of this until the contract system has run on its own for a quarter – integrating a system you don't trust yet just spreads the mistakes further.