The last ten percent of a project takes so long because that is where every deferred decision comes due at the same time. Edge cases that real data exposes, environments and deployment, permissions, error states, the copy on the empty screens, the sign-off itself. None of them was hard on its own. Together, at the end, they are the reason "feature complete" and "live" sit six weeks apart. I stopped hoping the pattern would skip us and started planning for it.
Where does the time actually go?
The time goes into the tail: the tickets everyone agreed to do "at the end". On one project I counted it in week eight and found sixty-one items, almost all of them an hour or two each. Sixty-one of those is close to three weeks, and none had ever appeared on any plan.
The items themselves look tiny written down. A date format on an export. The behaviour when a user has no team assigned. What the invoice looks like at zero rupees. The timeout on a slow upload.
The tail also holds the work that only exists once real data shows up. Names with apostrophes, addresses without postcodes, phone numbers with spaces, a category nobody uses that breaks a report. Staging data is polite. Production data is not.
Why does everyone relax too early?
Because the demo path works. Around the middle of a project there is a moment where the main flow runs end to end and everyone exhales. That is a real milestone and also a trap, because the demo path is the part of the product designed to work. The rest was meant to be finished later.
| What you check | On the demo path | Everywhere else |
|---|---|---|
| Data | Three clean records, prepared in advance | Apostrophes, missing postcodes, odd categories |
| Roles | The one account we always log in with | Every permission level, including the empty one |
| Errors | Never fires during the demo | Needs a screen, a message and a way back |
| Deployment | Runs from somebody's laptop | Rehearsed twice, reversible, on their server |
I keep coming back to the idea that the demo working is the problem, because it names the moment precisely. The demo is where confidence peaks and where the remaining work is least visible. On AI features the gap is wider still, since a model that handles the ten examples in the demo says nothing about the four hundred shapes of input real users will send.
How do I plan for the tail now?
Two changes. First, the plan carries a phase after feature complete, called hardening, and it gets twenty to twenty-five percent of the timeline. It has written exit criteria of its own. When people see that phase on the plan from day one, nobody mistakes feature complete for launch.
The criteria are the same on every project:
- A real-data rehearsal, run against a copy of production.
- The deployment rehearsed twice, including the rollback.
- Every role tested, including the one with no permissions.
- Every error state given a screen, a message and a way back.
Second, every ticket that gets deferred is estimated at the moment it is deferred. "We'll do the empty states later" now comes with "that's about six hours" attached, and the six hours go into the tail's running total. When the total crosses what the hardening phase can hold, we know in week five instead of week eleven.
How do you explain this to a client?
Plainly, at kickoff, before there is anything to be defensive about. I show the plan with the hardening phase on it and say what it is for. Most clients have lived through a project where the last stretch dragged, and they are relieved to see somebody name it in advance.
The harder conversation is when a client wants to skip it. "The demo works, why can't we launch Monday?" My answer is usually to walk them through the tail list, item by item, and ask which ones they would like their users to find. It has never taken more than five items.
At Shanti Infosoft every software development plan we send carries that phase, because I would rather show a longer, honest schedule in week one than a shorter one that grows a mystery month at the end. The mystery month is the one that costs the relationship.
What is on your project's tail right now, and has anyone added it up?
Sonal Jain runs project delivery at Shanti Infosoft, a CMMI Level 5 team of 80+ engineers, and has yet to meet a project without a tail.
Top comments (0)