How to pick the one thing worth a week of engineering
Every business I talk to has a list. It is usually long, usually in someone's head, and usually ordered by how recently each item caused pain. Ask which one is worth a full week of senior engineering and the room goes quiet — not because nobody has an opinion, but because three people have three different ones and none of them has a number.
That is the actual problem. Not a shortage of things to build. A shortage of agreement about which one is bleeding.
A bottleneck has a receipt
Here is the distinction that does all the work: a bottleneck costs you something you can write down. An improvement is something you would like to be true. Both are legitimate. Only one of them survives contact with a week-long deadline.
So before anything else, put a receipt on each candidate. Hours a week, dollars a month, deals lost, customers churned, people quit. If you cannot attach any number at all, that item is not ready to be built — it is ready to be investigated, which is a different job.
| Bottleneck | Hours / week | Annual cost |
|---|---|---|
| Copying orders between two systems | 6 | $18.7K |
| Answering the same five support questions | 10 | $19.5K |
| Hand-built weekly reporting | 4 | $14.6K |
| A SaaS seat nobody can cancel | 0 | $4,800 |
Two things usually fall out of this table the first time a team fills it in. The item everyone complains about loudest is rarely the most expensive one. And the most expensive one is usually something nobody complains about at all, because it has been running that way for two years and has stopped registering as a problem.
Then check whether it fits in a week
An expensive problem is not automatically a week-sized problem. What makes something shippable in five days is not size — it is boundedness. You want work where the finish line can be written in one sentence, and where nothing outside your control has to happen first.
| “Fix our reporting” | “One emailed Monday sheet” | |
|---|---|---|
| Can you write “done” in one sentence? | No — which reports? | Yes |
| Does it need a decision from someone else? | Yes — three stakeholders | No |
| Does it touch data you already have? | Partly | Yes |
| Is there a human who will use it on Monday? | Unclear | Yes — the ops lead |
| Can it ship broken-but-useful? | No | Yes |
Those are the same project. The left column is a quarter of work and a committee. The right column is a week, and it is the one that teaches you whether the rest of the quarter is worth spending. Ship the right column first — every time — and let what people actually do with it tell you what to build second.
The forty-minute version
- 01Dump the list. All of it.
Every annoyance, every workaround, every tool nobody likes. Do not filter while you capture; filtering is a separate pass and mixing them loses the good items.
- 02Attach a receipt to each line.
Hours a week, dollars a month, or a named deal that died. No number, no vote. Items without a receipt go to a research list, not the build list.
- 03Cross off anything that needs someone else to decide first.
Waiting on a vendor, a contract, a hire, or a boardroom is not a week of engineering. It is a week of waiting with extra steps.
- 04For the top three, write the one-sentence finish line.
“Done means the ops lead stops opening the second system,” or “done means the Monday sheet arrives without anyone touching it.” If you cannot write the sentence, you do not understand the item yet.
- 05Pick the one with the biggest receipt and the shortest sentence.
Not the most interesting. Not the most strategic. The one that pays for itself fastest and can be judged by Friday without a meeting.
The patterns that keep coming up
Across a few dozen of these conversations, the week-one winner is almost always one of four shapes:
- <strong>Two systems that do not talk.</strong> Someone is the integration. They copy between tabs, three times a day, and they are the single point of failure for a process the whole company depends on.
- <strong>The same five questions.</strong> Support, sales, or ops answers them constantly. They are the cheapest thing in the building to handle automatically and the most expensive thing to keep handling by hand.
- <strong>One feature of an expensive tool.</strong> You renew four figures a year for a product where your team uses exactly one screen. That screen is usually a week of work and then it is yours.
- <strong>A report that exists only in someone's head.</strong> It gets rebuilt by hand every Monday, it is subtly different each time, and no decision made from it is reproducible.
None of those are glamorous. All of them free real hours for a named person in the first week, which is the only proof that matters early: someone's Tuesday got better and they can tell you how.
The best first project is the one whose success your ops lead can confirm without looking at a dashboard.
Do this exercise before you hire anyone, brief anyone, or open a prompt box. It takes under an hour, and it is the difference between a week that produces working software and a week that produces a plan for a plan.
One agreed deliverable, five days of senior engineering, code in your repo by Friday. The first call is fifteen minutes and its only job is picking the target.
See how the free week works →