How to check a software estimate when you can't read code
"Two sprints, assuming the API is stable."
When a developer gives you an estimate like this, you are often the one who has to turn it into a promise to a client, a launch date or a budget. You cannot read the code. So how do you tell whether the number is realistic?
The good news: judging an estimate is less about technology and more about assumptions, unknowns and dependencies, all of which you can ask about in plain language.
1. Treat every estimate as a range
An estimate is a best guess, not a commitment. Ask: "What is the best case and the worst case?" If the answer is "two weeks, maybe four", plan around three or more. A single number with no range usually means nobody has thought about what could go wrong.
2. Find the word "assuming"
Every estimate rests on assumptions. Developers often say them out loud ("assuming the API is stable") and then everyone forgets them. Write each assumption down and ask: "What happens to the estimate if this is not true?"
In our example, "the API is stable" means the outside service the team connects to will not change or misbehave. If that service belongs to another company or another team, you have just found a risk you do not control.
3. Look for work that is not code
Estimates often cover building the feature and nothing else. Ask whether these are included:
- Testing by QA or by users,
- Code review time,
- Deploying and checking the release,
- Data migration, if existing customer data has to change,
- Documentation and handover.
Each of these can add days.
4. Count the dependencies
A dependency is anything the work waits on: another team, a client sending access details, a design that is not finished, a third-party service. Ask: "What are you waiting on from anyone outside the team?" Every external dependency is a place where the schedule can slip without anyone on your team doing anything wrong.
5. Ask what is new to the team
Work the team has done before is predictable. Work that is new, like a new technology, a new integration or a part of the product nobody has touched in years, is not. Ask: "Is there any part of this you haven't done before?" If yes, consider a spike: a short, time-boxed research task to reduce the unknowns before you commit.
6. Check the size of the pieces
Large tasks hide surprises. If an estimate says "three weeks" for one item, ask the team to break it into pieces of a few days each. The total often grows when they do, and that is a good thing, because it is now more honest.
7. Add a buffer, out loud
Experienced teams add a safety margin and say so. A common rule is to add 20 to 50 percent depending on how many unknowns you found. Agree the buffer openly with the team, so it is planning rather than padding.
A quick checklist
| Check | Question to ask |
|---|---|
| Range | What are the best and worst cases? |
| Assumptions | What if this assumption is wrong? |
| Non-code work | Does this include testing, review and release? |
| Dependencies | What are we waiting on from others? |
| Unknowns | What haven't we done before? |
| Size | Can we break this into smaller pieces? |
| Buffer | What margin should we plan for? |
Let Plainlee do the first pass
Plainlee's estimate check reads an estimate or plan and gives you a verdict (realistic, tight, unrealistic or unclear), the things that could stretch it, a suggested buffer, and the questions to ask your engineers. It is available on the Pro plan, and you can try Plainlee free today.