Most AI projects fail on unready data and unexamined processes, not the technology. zedders focuses on what actually needs fixing, and where to start.
Should you get your data and your processes in order before you invest in AI?
Mostly yes, but it’s not quite that simple. You don’t need to spend months cleaning your data before starting anything. You just need the specific data and processes behind the real work you’re trying to automate, and it doesn’t even have to be particularly clean. AI these days is very good at navigating missing entries and inconsistent formatting. Once that’s set up whatever automation you build will be based on the work you really do. This is why zedders’ Discovery phase is so important: to understand your business before trying to implement and to know what is actually worth automating.
Why do most AI projects fail?
Most fail on the data and the processes underneath, not the technology on top. RAND puts the AI project failure rate at over 80%, about twice the rate of IT projects that don’t involve AI. Gartner predicts that through 2026, 60% of AI projects will be abandoned for a lack of AI-ready data. The quality of the input: prompts, context, data, is usually the weak link, not the AI itself.
Any AI, and for that matter, any employee, is and will remain confined to their own knowledge and understanding of your business’ operations. If you skimp on training staff, they won’t know what to do when things go downhill. AI is the same: if you don’t provide it with quality data, representative of the outputs you’re actually seeking, you can’t expect incredible results.
Informatica’s 2025 survey of data leaders ranked data quality and readiness among the largest obstacles to AI success, ahead of any shortage of technical skill. So really, most AI disappointment isn’t a new AI problem at all. It’s an old problem: scattered records and undocumented processes, wearing a new expensive coat.
What does “AI-ready” data actually mean?
It’s not just tidy spreadsheets. AI-ready data is connected, current, and represents the real situation. For most businesses the first obstacle isn’t dirty data, it’s scattered data.
The technical name is a data silo: information held in one system or one department that the rest of the business can’t reach. Try to imagine it in your own business for a second. A retailer’s stock, sales, and supplier records sit in three tools that don’t connect. A trades business keeps quotes in one place, job history in another, and the real knowledge in someone’s head. When AI can only see one part, it reasons from that and gives you a confident, incomplete answer.
So how much of your business would AI actually be able to see today?
In the Cloudera and Harvard Business Review study, siloed and hard-to-integrate data was the most common barrier of all, named by 56% of respondents. For most businesses, ‘getting your data in order’ doesn’t need to be a massive governance project. It can simply be connecting the systems you already have and agreeing on a single source of truth.
Do you need perfect data before starting AI?
No. Waiting for perfect data is a sure way to never start. Data clean-up has no natural finish line, operations adapt and minds get changed. Anyone who sets out to perfect everything first should be ready to spend a year in planning while their competitors start tinkering.
Modern AI also tolerates a good deal of mess. It can read inconsistent records, rough notes, and badly formatted documents that would have broken older software.
So the goal isn’t to have all your data ready. It’s to have the right data for job one ready.
Should you automate a broken process?
No. Just no.
Automating a broken, undocumented, or pointless process doesn’t fix it. It just makes the mess run faster, and at greater volume.
Years ago Bill Gates said: “The first rule of any technology used in a business is that automation applied to an efficient operation will magnify the efficiency. The second is that automation applied to an inefficient operation will magnify the inefficiency.”
The best move is almost always to redesign a step, or even the whole process, before any AI or automation is set up on top. If you couldn’t explain the process in full to your staff, how could you expect to explain it to an AI? And this isn’t just the best-case version, it’s the exceptions, the edge cases and what judgement to apply when something new pops up. That’s why zedders always suggests starting with a full Discovery phase.
Where should you actually start with AI?
Ask me where to begin, and I’ll always give you the same answer. Start from the job. Not the data and not the tool. Pick one piece of work that costs real time every week, talk to the people who do that work, and get to know not only the process you think they’re doing but the one they’re actually doing. Then figure out where the data is that every decision relies on. Measure your starting point and experiment with new tools to see what really works.
If you know the problem but don’t know what to do about it, or you know what to do but don’t know how to do it, Contact us and we can help you find out what the next step is.
The discovery-first habit protects you from spending on the wrong thing, because looking closely at one job often reveals that there are ways to improve the process, making automation even more effective. This is the core of zedders’ Discovery phase. Learn how the business actually runs, find where AI or automation would save real time and money, ready the data that’s needed, and point out where a process doesn’t suit or isn’t ready for automation.
The AI-first rush
Starts from the tool or the trend
Tries to make all the data ready at once, or none at all
Assumes the process is worth keeping
Counts success as “we deployed AI”
Adds AI on top of the mess
The discovery-first start
Starts from a job that costs real time
Readies only the data needed for the job
Asks whether the process should be fixed or dropped first
Measures the before, so the improvement is quantifiable
Clears the mess, then decides if AI helps
Starts from the tool or the trend
Starts from a job that costs real time
Tries to make all the data ready at once, or none at all
Readies only the data needed for the job
Assumes the process is worth keeping
Asks whether the process should be fixed or dropped first
Counts success as “we deployed AI”
Measures the before, so the improvement is quantifiable
Adds AI on top of the mess
Clears the mess, then decides if AI helps
The technology is rarely the hard part now. It’s knowing which job is worth it, what to get ready, and what to leave alone, and it all happens before you even think about AI.
So before you buy anything, sit with a few honest questions about your own business first.
- Which job costs you the most time every week?
- Do you actually know how long it takes at the moment?
- Is the process behind it worth keeping, and do you really even know what it is?
If you can’t answer those yet, that’s fine. That’s what the first conversation is for. A thirty-minute chat is the fastest way to find out whether zedders can help your business. Contact us to get started.
Common questions
Do we need to fix our data before using AI? Not all of it. You need the specific data behind the one job you want to improve to be connected and reliable enough to trust. Trying to fix everything first is one of the most common ways to stall.
How much data do you need to start with AI? Less than you might think, but more focused. The right data for one clear job beats a large, messy pile that covers everything and settles nothing.
What is the first step for a small business? Pick one task that costs real time each week. Look at the steps and decisions behind it, along with what informs each one. Measure what it’s costing you right now. Otherwise how will you know whether a change is worth it. Then, if you know what you need, Contact us and we can jump straight into Implementation.
Can AI work with messy or unstructured data? Mostly, yes. Current tools handle rough documents and inconsistent records better than older software did. You don’t need to spend hours cleaning data manually.