Why AI-built prototypes are not the same as expert built planning solutions
Vibe coding is having a moment. For anyone who has missed the phrase, vibe coding is the use of AI tools to rapidly build applications, dashboards or workflows by describing what you want in natural language. Instead of writing every line of code yourself, you prompt, refine and keep going until something starts to look like a working solution.
To be fair, it can be very impressive. You ask for a dashboard, and a few prompts later, there it is. You ask for a user interface, and suddenly you have one. For early ideas, quick prototypes and simple visualisations, AI can get you surprisingly far, surprisingly quickly.
So when a retail planning team says, “We need a better solution,” it is understandable that someone might suggest: “Couldn’t we just build this ourselves in Claude?” It sounds faster, it sounds cheaper and it sounds flexible. It also sounds like a way to avoid another big software project.
But in retail planning, especially when we are talking about MFP, WSSI, Range Planning, Assortment Planning or Allocation and Replenishment, there is a big difference between something that works in a demo and something that can run the business.
That difference is where the vibe coding dream starts to wobble.
The first 10% is exciting. The remaining 90% is the problem.
This is where vibe coding can be a little dangerous.
The first 10% feels brilliant. You can get screens, dashboards and early concepts up and running quickly. Everyone gets excited. Someone says, “This only took three days.” Someone else says, “Imagine where we could be in three months.”
Then the reality of retail planning arrives, because planning is not just a nice-looking screen. It is business logic, workflow, data integration, security, governance, version control, scenario modelling and the ability to support teams making decisions that affect stock, margin, cash and availability.
The first 10% is often visual. The remaining 90% is where the complexity lives. That is when you start asking the tool to handle minimums and maximums, weighted average cost price, returns logic, intake phasing, markdowns, open-to-buy controls, stock provisioning, size curves, store grading, multiple hierarchies and rolling forecasts.
Suddenly, the “quick AI build” becomes a long internal development project, and by the end, you may have spent more time getting less functionality than you would have had with an expert-built planning solution from the start.
Dashboards? Maybe. Planning systems? Be careful.
There are absolutely areas where AI-assisted development can be useful. Dashboards are a good example. If you already have a robust EPM or planning platform underneath, AI-built interfaces or rapid dashboard prototypes can help teams explore ideas, test layouts and move faster. That is very different from using an LLM to replace the planning engine itself.
LLMs are probabilistic engines. In simple terms, they predict what is most likely to come next. That makes them brilliant at creating text, visuals, summaries and early concepts that are broadly right. Retail planning is not about being broadly right, it is about being specific, detailed and repeatable.
A planning solution cannot give one answer on Monday and a slightly different one on Tuesday because the same prompt was interpreted differently. It cannot calculate stock cover one way for one version and another way for the next. It cannot “almost” apply the right logic to returns, intake, markdowns or margin.
People sometimes confuse LLM output with reasoning and logic. That is where the risk comes in. AI can support planning, but it should not be the system of record for planning logic. You need calculations, workflows and controls that behave consistently every time.
A dashboard shows information, a proper planning solution runs processes, and those processes need to be trusted.
Retail planning expertise matters
One of the biggest risks with vibe coding is not the technology itself. It is the assumption that building the tool is the same as understanding the planning process. Retail planning is highly specific.
MFP is not just a spreadsheet with nicer buttons. WSSI is not just a weekly stock report. Assortment planning is not just picking products and assigning them to stores. These processes involve detailed commercial logic, cross-functional decision-making and years of embedded retail knowledge. They need to connect finance, merchandising, supply chain and operations in a way that supports better decisions, not just prettier outputs.
Some IT teams understand planning very well. Many do not, and that is not a criticism. Their role is usually to support systems, infrastructure, data, security and delivery across the whole business.
Planning specialists bring a different lens. They understand the awkward middle ground between what the business wants, what the data allows and what the system needs to do. They know which requirements are genuinely important, which ones add unnecessary complexity and which ones might cause problems six months later. That experience de-risks the project.
When you implement an EPM or retail planning solution with an expert team, you are not starting from a blank page. You benefit from proven approaches, accelerators, best practice models and experience from other retail projects.
That does not mean every solution should be identical. Retailers are different, and planning solutions should reflect that, but it does mean you are not asking your internal team to discover every planning pitfall the hard way.
Security, stability and support cannot be an afterthought
A planning solution is not something you build once and walk away from. It needs to be hosted, secured, maintained, upgraded, monitored, supported, documented and tested. It needs clear ownership and governance.
Most importantly, it needs to survive Monday morning. If the system is down at the start of the trading week, who is responsible? If a data load fails, who fixes it? If a logic change breaks a key calculation, who spots it? If the business changes its planning model, who updates it safely?
These questions are not as exciting as an AI-generated prototype, but they matter far more. Retailers rely on planning systems to make decisions about stock, margin, sales, cash and availability. Downtime and errors do not just create inconvenience, they can affect commercial performance.
With a proper EPM or retail planning platform, the architecture, governance and support model are part of the solution. With a vibe-coded build, those responsibilities still exist, but they are much easier to underestimate.
The cost saving is not always real
The argument for vibe coding is often cost. Why pay for software and implementation when we can build something ourselves?
The problem is that the cost comparison is rarely complete. It may look cheaper if you only compare licence fees or initial build time. But the real cost includes internal resource, testing, rework, documentation, hosting, security, support, maintenance and future enhancements.
Then there is the cost of tokens, which is not always easy to predict upfront. Before you hit the query, you do not necessarily know how much it will use, how many rounds of prompting it will take, or how much rework will be needed to get the output into a usable state. That makes budgeting difficult. In what other scenario would you start building a business-critical system without a clear idea of what it will cost to finish?
You could be halfway through prompting an AI tool into building something and hit a limit. Then again, and again. You keep going because you have already invested time, but the costs continue to creep up. At that point, the business faces an uncomfortable question: spend more to try and finish it, or abandon it and accept the time and money already lost.
Token pricing is also not something retailers can control. Costs, limits and usage models can change. What looks affordable at the start may not look quite so attractive once you are deep into the build, relying on it, and still trying to get the basics working properly.
So the saving is not always as clear as it first appears. If you hire a dedicated team to build and maintain the solution, the cost saving starts to disappear. If you rely on people who already have full-time roles, delivery slows down and business-as-usual work gets in the way.
The outcome is often the same cost with more risk, lower cost with less capability, or slower delivery with more compromise. None of those are ideal when the planning team needed a robust solution in the first place.
AI is a tool, not a planning strategy
None of this means AI should be ignored. AI has a valuable role to play in modern planning, it can support forecasting, summarisation, anomaly detection, data exploration, scenario generation, workflow guidance and faster analysis.
But AI works best when it is applied within a strong planning framework. Without the right data model, business logic, governance and user adoption, AI can become a very confident way to create more confusion.
There is a reason AI-generated solutions can start to look the same. If the tool is doing most of the thinking, and there is limited planning expertise guiding it, you often get a generic answer.
Retail is not generic. A fashion retailer, a grocery business, a homeware brand, a wholesale-heavy retailer and a multi-channel international business all have different planning challenges. The solution needs to reflect the way the business actually works.
AI can accelerate parts of the journey, but it should not replace the thinking.
The better question is: what are you trying to solve?
The most useful conversation is not, “Can AI build this?” it is: “What planning problem are we trying to solve, and what is the safest, fastest and most scalable way to solve it?”
Sometimes, a quick AI-assisted prototype may be useful. It can help visualise an idea, test a concept or bring stakeholders into the conversation, but when the requirement is retail planning, the bar is much higher. You need a solution that can support the business today and adapt tomorrow. You need proper integration, robust calculations, clear workflows, secure access, consistent reporting and confidence in the numbers.
That is where experienced EPM and retail planning teams add value.
At bdg, we work with retailers to design and implement planning solutions that connect the commercial process with the technology behind it. From MFP and WSSI to range and assortment planning, our focus is on building solutions that are practical, scalable and grounded in how retail teams actually work.
The goal is not to have the most impressive demo, it is to make better decisions, faster, with a solution the business can trust.
So, is vibe coding bad for retail planning?
No, but it is often used for the wrong purpose.
Vibe coding can be brilliant for exploration, prototypes and speeding up certain development tasks. It can help teams think differently and move quickly in the early stages. But as a replacement for an expert-built retail planning solution? That is where we would be cautious.
The risk is not that AI cannot build something. It can. The risk is believing that “something” is the same as a secure, supported, scalable planning solution that can run core retail processes.
In retail planning, the first 10% gets attention, and the remaining 90% is where the real work begins. That is also exactly where experience matters.
Frequently asked questions (FAQ)
What is vibe coding?
Vibe coding is the use of AI tools to help build applications, dashboards or workflows through natural language prompts. It can be useful for rapid prototyping, but it still needs strong technical and business oversight.
Can vibe coding be used for retail planning?
It can support early concepts, dashboards and prototypes, but it is risky to rely on vibe coding alone for core retail planning processes such as MFP, WSSI, range planning or allocation and replenishment.
Why is retail planning difficult to build from scratch?
Retail planning involves complex calculations, multiple hierarchies, data integration, approvals, scenario modelling, stock logic, forecasting, reporting and ongoing business change. A working screen is only a small part of the full solution.
Why are LLMs risky for planning logic?
LLMs are designed to predict likely outputs, not to act as deterministic planning engines. They can be useful for ideas, summaries and visual concepts, but planning logic needs to be consistent, repeatable and governed.
Is an EPM platform better than a custom AI-built planning tool?
For core planning processes, an EPM platform usually provides a more robust foundation because it includes governance, scalability, security, workflows, integration and planning-specific functionality. AI can still enhance the process when used in the right way.
How can AI support retail planning?
AI can support forecasting, anomaly detection, scenario analysis, reporting, data exploration and workflow assistance. The best results come when AI is applied to a well-designed planning model with reliable data and clear business rules.
