What the weeks are actually spent on
| Phase | Duration | What happens |
|---|---|---|
| Discovery and specification | 1–3 weeks | Watching the process, writing down rules and exceptions |
| Build | 3–8 weeks | The part everyone imagines is the whole project |
| Parallel running and testing | 2–4 weeks | Running alongside the manual process, comparing |
| Launch and settling | 1–2 weeks | Cutover, training, the first wave of small fixes |
Note that building is roughly half the calendar. Projects planned as if it were 90% run late by exactly that error.
Typical totals by scope
- One process, one system, no exceptions worth speaking of: 4–6 weeks
- One process, two or three systems, real exception handling: 8–12 weeks
- Several processes or a system with an awkward interface: 3–5 months
- Anything requiring a third party to build something for you: add whatever they say, then some
What actually causes delay
- Unanswered questions. The single biggest factor. A question that takes five days to answer costs five days, every time.
- Third-party access. API credentials, sandbox accounts and permissions from a vendor can take weeks and are outside everyone's control.
- Data that is not ready — the export nobody has produced before, the field that turns out to be free text.
- Process still being argued about. If two departments disagree about the rule, the build waits.
- Availability of the person who knows how it works, who is invariably the busiest person in the business.
Ask for third-party API credentials in week one, before they are needed. It is the item most likely to sit on someone else's desk, and it is free to start early.
How to compress a schedule safely
Cut scope, not stages. Removing the second intake channel from phase one saves weeks; removing testing saves days and costs months.
- Automate one channel first, add the others later
- Handle the top three exceptions automatically and route the rest to a person initially
- Use a manual step where an integration would take four weeks to arrange
- Launch to one team before everyone
What you can do to make it faster
Answer questions within a day. Nominate one person who can decide. Get credentials early. Provide real data early, including the awkward records. Be available for the demonstrations.
Clients who do these things routinely finish two to three weeks earlier than clients who do not, on identical scopes. It is the largest variable within your control.