The project is late, the supplier has gone, and the board wants a date. This article sets out the 30-day sequence: stop the loss, take control of access, inventory what runs, audit the code without a trial, then write the repair-or-rebuild decision and a new scope with acceptance you can run.
A project is late, the previous supplier has gone, the invoices are paid, and the board wants a date. The instinct is to demand a rewrite or to replace the team by Friday. Both feel decisive, and both usually make the loss bigger, because neither is based on facts about the system you actually have.
Rescue work is a sequence, not a gesture. In the first 30 days you stop the loss, take control of access, establish what runs and what does not, and write a decision about repair or rebuild with evidence behind it. Only then does a new scope make sense. This article sets out that sequence, the evidence to collect at each stage, and the contract terms that keep a rescue from becoming a second failure.
Week one: stop the bleeding
The first week has one job: stop the loss and take control. Freeze new work and non-urgent releases, but keep a fast path for hotfixes, because a frozen system that cannot be repaired is worse than a busy one. Collect production access: the cloud console, the database, the domain registrar, the repository, the payment dashboard, the carrier accounts, email and analytics. Without that access you are advising, not rescuing.
Take a backup and restore it once into a clean environment. A backup nobody has restored is a hope, and the week you need it is the worst week to find out it was incomplete. Turn on error tracking and uptime monitoring, with alerts that reach a named person rather than a shared inbox. Then pick the two failures that cost money today, usually checkout, login or data loss, and fix only those.
What you do not do in week one matters as much. Do not refactor, redesign, renegotiate the whole contract, or dismiss the team. Each of those is a decision that needs the audit, and doing them first destroys the evidence you need to make it. A rescue that starts with a purge starts with a second loss of knowledge.
The inventory: what you actually own
The inventory is the first document that changes the conversation, because it turns opinion into a list. It covers nine areas, and each one produces evidence rather than a verdict.
| Area | The question that decides it | The evidence you collect | The decision it unlocks |
|---|---|---|---|
| Access and ownership | Who holds production keys, the cloud accounts, the domain and the repository? | A written account list, with an owner for each and the last key rotation | Whether you can operate the system at all |
| Environments | Can you deploy to staging and production without the previous team? | One full deploy performed by your team, start to finish | Whether the pipeline is an asset or a dependency |
| Data | Where is the data, is it backed up, and has a restore been tested? | A restore into a clean environment with the elapsed time recorded | Whether a bad release is recoverable |
| Deployment | What runs in production, and what sits in the repository? | A diff between the running build and the main branch | Whether the code is the source of truth |
| Dependencies | Which libraries are abandoned, unsupported or licence-encumbered? | A list with release dates and licences | Which upgrade is urgent and which can wait |
| Tests | Which critical paths have automated tests? | A test run against checkout, login and payment | How fast you can change anything safely |
| Monitoring | Would you know today if checkout failed? | An alert delivered to a named person | Whether you detect incidents or customers do |
| Contracts | What was delivered against what was signed? | A line-by-line mapping of contract to running system | What you can ask for and what you must rebuild |
| Users | What do users experience now, and what do they complain about? | Three months of support tickets grouped by failure type | Which failure costs money this month |
If rows come back empty, that is the finding. An empty access row means you do not control the system; an empty deployment row means the repository is not the source of truth; an empty monitoring row means your customers are the alerting.
The audit without a trial
The audit answers three questions: what the system does, what it cannot do, and what it will cost to change. It is not a review of taste, and it is not a case against the previous team. Use the product yourself with a test account and complete one real transaction end to end, then trace the same transaction through logs and database rows.
Then read the system the way an engineer reads it: the last few hundred commits, the deployment configuration, whether migrations run on an empty database, whether secrets sit in the repository history, how old the dependencies are and under which licences, what the test suite actually covers, and whether permissions are enforced on the server or only hidden in the interface. Measure performance at the 75th percentile against the published thresholds: largest contentful paint at 2.5 seconds or less, interaction to next paint at 200 milliseconds or less, and cumulative layout shift at 0.1 or less. Check accessibility with a keyboard and a screen reader.
Write findings as facts, with file names, timestamps and reproduction steps. “The payment handler has no idempotency key, so a retried notification can charge twice” is a fact you can act on. “The previous team was careless” is not. The difference is not politeness; the first report is usable by a new team, by a committee and in a contract dispute.
Keep the previous supplier cooperative for as long as you need explanations and credentials. Put every request in writing, give each answer a date, and escalate through the contract rather than through accusation. Access is the constraint that sets the pace of a rescue, and a supplier who feels attacked has little reason to hurry. You can hold a firm commercial position and still get the handover.
Repair or rebuild
This decision is not a matter of taste either. Repair is usually the better path when the data model can express the business, the failures sit in a small number of modules, the deployment pipeline can be fixed in days, and the domain knowledge is still reachable. Rebuild is usually the better path when the data model cannot represent the business without constant workarounds, when authentication and permissions are structurally broken, when nobody can deploy without the previous supplier, or when a licence or platform decision has no future.
There is a third option that is often the right one: replace the system module by module behind a stable interface, keeping what works and starving what does not. It is slower to describe and usually cheaper to survive, because it does not require running two versions of everything at once. Whatever you choose, the cost of a rebuild is not the build. It is the migration, the parallel run, the retraining and the features you will not ship while it happens. Write that number down before you choose, because the rebuild estimate that leaves it out is the one that fails.
From reading to shipping
Want this built, not just explained?
Bring us the problem behind the article. We will tell you what it takes, what it costs, and whether we are the right team.
Put the decision on one page: the evidence, the option chosen, what is deferred, the review date, and who signed it. A rescue without a written decision restarts the same argument every month.
Reset the scope
Rescuing a project means shrinking it before it grows. Define the minimum operable system: the shortest version users can rely on and the business can run. Put every request into one of three lists — must work at launch, can wait a quarter, will not be built — and publish the third list. A deferral that is written down is a decision; a deferral that is implied is a rumour that reaches the board.
Acceptance changes shape too. Stop accepting requirements as sentences and start accepting them as scenarios a person can run in front of a committee: a new customer registers, adds two items, pays, receives a correct invoice, sees the order in admin, and stock moves once. A scenario that cannot be run cannot be accepted, and what cannot be accepted will be argued about at the end.
Communication while it is broken
Users do not need the full story, but they need the truth: what works, what does not, what to do instead, and when the next update comes. One page with those four lines, updated on a fixed day, removes most of the support load. Inside the company, send one weekly note with the facts, the decisions, the risks and the next date. Resist the optimistic date; give a range and the condition that would move it.
Brief the support team before the announcement, with written answers to the five questions customers actually ask. A support team improvising during an outage turns a technical incident into a reputational one, and the repair costs more because the same failure is explained five different ways.
The 30-day checklist
Use this list as the gate between one week and the next. Every item produces something written that another person can inspect, not a promise made in a meeting. Also review technical consulting, and the earlier post on choosing a software company in Saudi Arabia, which explains how to read proposals before you sign.
Access inventory: every account and key, its owner, its last rotation, and who can revoke it.
One real restore: a backup restored into a clean environment, with the result and elapsed time recorded.
One deploy by your team: your people ship a small change from repository to production without outside help.
One traced transaction: a real order from purchase to invoice, followed through logs and data rows.
Written findings: an audit report with facts, file names and reproduction steps, not general verdicts.
A signed decision: one page choosing repair or rebuild, with the evidence, the deferrals and the signature.
The minimum scope: a list of what must work at launch, and a published list of what will not be built.
A communication rhythm: one decision owner, a weekly note, and a customer status page updated on a fixed day.
The contract and acceptance
If you bring in a rescue team, buy the audit as a separate, bounded phase with a written report you own, then decide. Do not let the audit be free and the scope open; a free audit is a sales document. Rasad's published band for technical consulting is SAR 20,000 to 30,000 over six to ten weeks, and a rescue is priced from the audit rather than assumed to fit that shape. Review the published bands before you ask for a proposal.
The contract needs four things beyond price: access and intellectual property from day one, acceptance scenarios that run, a named team that stays, and an exit that leaves you able to deploy without the supplier. Add a warranty period after handover and a change-control rule, so a new request carries a cost and a date instead of becoming an argument. Then move the stabilised system onto a written care plan, and read what a monthly retainer should cover in a care plan after launch.
Frequently asked questions
Do we have to replace the team? Not as a first move. Replacing people before the audit removes the only source of explanation about the system, and often the only people who can hand over access. Decide after the findings, and separate the commercial dispute from the technical recovery.
How long before we know whether to repair or rebuild? The audit is a bounded phase with a written report at the end, and its length depends on how much of the system is undocumented. The decision does not wait for a full rebuild estimate; it needs enough evidence to compare two paths.
Should we stop new features? For a fixed window, yes, with the deferral list published. A new feature on an unstable base adds cost to every later change, and it makes the recovery date less believable.
What if the previous supplier refuses to hand over access? Start with a written request and a date, then read the contract with whoever advises you on it. Most access problems are contractual rather than technical, and a public accusation usually slows the handover.
How do we know the rescue is working? Measure the same four things before and after: failed transactions, deploys per week, time to restore service, and support tickets by failure type. A rescue that improves the demo but not those four has not started.
Sources
- web.dev: Web Vitals — largest contentful paint at 2.5 seconds, interaction to next paint at 200 milliseconds, and cumulative layout shift at 0.1 at the 75th percentile.
- Rasad technical consulting — how we review a system, its scope and its risk before a build decision.
- Choosing a software company in Saudi Arabia — reading proposals and checking the team before you sign.
- Rasad care after launch — what remains after stabilisation: monitoring, fixes and updates.
- Rasad's published price bands — the figures this article works from.
Cover image: Stevie Benton (WMUK), CC BY-SA 4.0, via Wikimedia Commons.

