Fix, configure, connect, replace: choosing the right path for a broken process
When a process breaks, the instinct is to buy new software, but there is usually a cheaper fix closer to home. Here is the ladder I use, from fixing the process to building something new, and where AI fits on it.
By Gary Kellett7 min read
For directors and operations leads with a slow, error-prone or person-dependent process they want sorted.
When a process is slow, error-prone or dependent on one person, the instinct is to look for new software. Sometimes that is right. More often there is a cheaper, quicker fix much closer to home. I think of the options as a ladder, from the cheapest and least risky to the most expensive and most disruptive: fix the process, configure what you have, connect your systems, replace a system, or build something new.
The rule I follow is simple. Start at the bottom of the ladder and only climb when the rung below genuinely cannot solve the problem. Here is what each rung involves, when it is the right answer, and the warning signs that you are standing on the wrong one.
The ladder at a glance
- Fix the process: change the steps, remove duplication, agree one way of working. Cheapest, fastest, lowest risk.
- Configure: switch on and set up features in systems you already pay for. Low cost, low risk.
- Connect: link systems so data flows between them instead of being re-keyed. Moderate cost and risk.
- Replace: move to a different product. Higher cost, higher risk, and usually months of disruption.
- Build: create something bespoke. The highest cost and an ongoing duty to own it, but sometimes the only route to a real advantage.
A rough order, not a price list
These are relative positions, not fixed prices. A large integration can cost more than a small replacement. But as a starting assumption, each rung up the ladder adds cost, time and risk, so the burden of proof should rise as you climb.
Fix the process
Many broken processes are broken in the steps, not the tools. Work gets passed back and forth, approvals happen in the wrong order, two people do the same check, or nobody owns the outcome. Mapping how the work really flows, on one page, usually shows the problem within an afternoon.
This is the right answer when the pain comes from hand-offs, unclear ownership or everyone doing the job differently. The warning sign that you have stopped too early is a tidy new process that still needs someone to copy the same information from one screen to another.
Configure what you already have
Most business systems can do far more than they have been set up to do. Workflow rules, approval routes, templates, automatic reminders, extra fields and reports are often included in the licence and never switched on. Good configuration is frequently the best value change a business can make.
It is right when the system can already do what you need and simply has not been told to. The warning signs that configuration has run out are heavy customisation the vendor will not support through upgrades, or a growing list of features that nearly work and each need a workaround.
Connect your systems
When two good systems do not talk to each other, people become the integration. They type orders from one into the other, export and re-import lists, and reconcile by hand at month-end. Connecting the systems through their standard interfaces, or a light integration service, removes that work and the errors that come with it.
It is right when each system does its own job well and the pain sits in the gap between them. The warning signs are a connection that needs constant fixing, or one propping up a system that cannot really do its job, in which case the next rung may be cheaper in the long run.
Replace a system
Replacement is right when a system genuinely cannot do what the business needs: the vendor confirms the gap, support is ending, or it cannot cope with your growth. Done well, it removes a whole class of problems. Done badly, it moves the old ones into a more expensive home.
Before you replace, write a requirements list from how the work should flow, brief suppliers so their quotes are comparable, and plan the data migration as a project in its own right. The warning sign is choosing a product before you have fixed the process it is meant to support.
Build something new
Building is right when nothing on the market fits, when the process is a genuine source of advantage, or when the gap is narrow enough for a small, well-made tool to close. It is also the rung with the longest tail: someone has to host, support and change what you build for as long as you use it. Price that in from the start, not after go-live.
Where AI fits
AI is not a rung of its own. It is one ingredient that can appear on several of them. A connection might use AI to read supplier invoices in any layout. A configured system might use the vendor's built-in AI to draft replies. A bespoke tool might use it to summarise documents for a person to check. In each case the decision is still fix, configure, connect, replace or build. AI changes what is possible on a rung, and it brings its own running costs and governance questions that need an owner.
What it will not do is rescue a broken process. Automating a bad process just makes the mess faster.
Worked example
A services firm's quotes take days, because the price is worked out in a spreadsheet, copied into a document template and then entered into the CRM. Fixing the process removes a second pricing check that duplicated the first. Configuring the CRM's built-in quote templates removes the separate document. Connecting the CRM to the accounts package removes the re-keying when a quote is won. No replacement or build is needed, and AI only enters later, to draft the covering email for a person to edit.
Choosing your rung
Step 1: Describe the problem in one sentence
Name the process and the pain, for example 'supplier invoices take two days to approve'.
Step 2: Ask whether the steps are sound
If not, fix the process first, then see what problem remains.
Step 3: Ask whether your systems can already do it
Check your licence and ask the vendor in writing. If they can, configure.
Step 4: Ask whether the gap is between systems
If the data exists but is re-keyed or reconciled by hand, connect.
Step 5: Only then weigh replace against build
Compare both with real figures, including running costs and who will own the result.
If you are not sure which rung your problem sits on, or which kind of help would suit it, the Where should I start? tool asks nine short questions and points you to the right next step.
Questions directors ask
Should I fix my process or buy new software?
Fix the process first. Many problems come from hand-offs, duplicated checks or unclear ownership, and new software will not remove them. Once the steps are sound, check whether your existing systems can be configured or connected before you consider buying anything new.
What is the cheapest way to fix a broken business process?
Usually changing the steps themselves: removing duplication, agreeing one way of working and naming an owner. The next cheapest options are configuring features in systems you already pay for, then connecting systems so data is not re-keyed. Replacing a system or building something new comes last.
Where does AI fit when fixing a business process?
AI is an ingredient, not a separate option. It can help on several paths, such as reading documents inside an integration or drafting text inside a configured system. It will not fix a broken process, and it adds running costs and governance questions that need a named owner.