Ask Priya, she knows
Your fintech has grown from a founding ops person to a team of several. The first few people know how everything works: how to handle a returned payment from a particular provider, what to do when a card is reported stolen abroad, which partner contact to email about a stuck account. There is a Notion space with procedures, some accurate, some from last year, some half written.
New hires shadow for a few weeks and then ask questions in Slack. The same questions come up with every new hire. When one of the original team is on holiday, certain tasks wait. When one leaves, the team finds out what they did by what stops happening.
Why the documents never keep up
Procedures are written once, usually after something goes wrong, and then the process changes and the page does not.
- Documents sit apart from the admin and review tools, so nobody reads them while doing the task.
- Changes to a process are agreed in Slack and never reach the page.
- Pages describe the ideal process, not the workarounds people actually use.
- There is no owner for each procedure and no review date.
- Search in the wiki returns five pages that might be relevant.
The cost of knowledge held by a few
New hires take longer to become useful and make mistakes that the experienced people then fix. Senior ops people answer the same questions repeatedly. Outcomes vary depending on who handles a case, which is hard to explain to a partner who asks how a process works. And the business is exposed every time a key person is away.
Procedures inside the work
What we build puts the procedure where the task happens, and keeps a single written source behind it.
- We sit with the people who do each common task and write down what actually happens, including the workarounds, and agree a version with the ops lead.
- Each procedure gets an owner and a review date, stored with it.
- Inside your ops tools, each task type shows its steps as a checklist on the case, with links to the right screens and templates.
- Steps that can be automated, such as sending a standard request or filling a field, are automated, so the checklist gets shorter over time.
- A search assistant, built on a model such as Anthropic Claude, answers questions from your procedures only and links the source page. If the procedures do not cover the question, it says so.
- When someone finds a step is wrong, they flag it from the case, and the owner is asked to update the procedure.
| Task | Before | After |
|---|---|---|
| Returned payment from a specific provider | Ask a senior colleague | Checklist on the case, owner named |
| Card reported stolen abroad | Wiki page, maybe out of date | Guided steps with the card action built in |
| Partner escalation | Knowing who to email | Escalation template with contacts kept current |
| New hire question | Slack message | Assistant answer from your procedures, with source |
A team that does not depend on memory
New hires follow the checklist on each case and get answers from the assistant, with a link to the source. Senior people get fewer interruptions and spend the time on the unusual cases. When a process changes, the owner updates it once and the change appears on the next case.
When a partner asks how a process works, you can show the procedure, its owner, its last review and the cases that followed it.
Is your ops knowledge like this?
- New hires learn mostly by asking in Slack.
- Wiki procedures are out of date or contradict each other.
- Certain tasks wait when one person is away.
- Outcomes depend on who handled the case.
- Nobody owns each procedure or reviews it.