AI will not fix weak planning foundations. It will expose them.

Gartner has now put a number on the enterprise AI correction: more than 40% of agentic AI projects are expected to be cancelled by the end of 2027, driven by escalating costs, unclear business value and inadequate risk controls. Forrester’s 2026 outlook points in the same direction. AI is moving from hype to hard hat work. The demo phase is giving way to delivery, scrutiny and return on investment, CFOs are being pulled into the value conversation, and governance is no longer a side issue. It is becoming the condition of adoption.

A weak planning environment can survive manual workarounds for longer than it should, because a person usually knows where the model is fragile. A finance lead remembers why an assumption was added. A planner knows which forecast output needs a second look. A project owner can explain why a process is different in practice from the way it appears in documentation. AI does not have that context unless the organisation has created it. It inherits the logic it is given and works with the data it can access, at speed and with confidence, without knowing whether the foundation underneath it is still fit for purpose.

AI does not introduce weakness into planning. It industrialises the weakness already there.

What the market is getting wrong

The market is still treating AI as a capability question. What can the tool do? How quickly can it do it? How much time could it save? Which process could it automate? Those are reasonable questions, but they are not the ones that decide whether AI creates value.

Gartner’s warning matters because the failure modes it identifies are not model capability problems. Escalating costs, unclear business value and inadequate risk controls are organisational problems. They sit on the customer’s side of the line. They are about ownership, use case selection, governance, process design and the ability to prove that AI activity is changing something that matters.

Forrester’s language is just as telling. AI is no longer being judged on possibility. It is being judged on practical value. That changes the standard: the next phase belongs to organisations that can connect AI activity to business consequence and then explain how the decision was made.

Why planning is different

Most AI failures are irritating before they are dangerous. A pilot stalls. A chatbot goes unused. A summary tool saves less time than expected. A budget gets challenged and the project quietly disappears.

Planning is not like that. A forecast does not stay inside a planning model, it becomes a hiring decision, an inventory position, a revenue commitment, a supplier conversation, a cash decision, a board number or a target someone is held accountable for. That means AI in planning cannot be treated as a productivity layer alone. It is moving closer to the numbers people act on, and that changes the risk profile completely.

If AI helps generate a forecast explanation, the business needs to know what changed. If it supports scenario modelling, the business needs to know which assumptions were used. If it highlights a variance, the business needs to know whether the comparison is valid. If it recommends an action, the business needs to know who owns the decision.

The output might look polished, sound credible, and even be directionally useful, but planning decisions require more than fluency. That is particularly true for large language models. LLMs are built to generate fluent, convincing responses, but fluency is not the same as accuracy, context or accountability. An LLM might help a user query the model in natural language, asking why forecast demand has moved or where inventory risk is emerging. But the answer is only useful if it is querying a planning model with clear logic, trusted source data, reviewed assumptions and visible lineage behind the number. The real risk is not that someone asks the wrong question. It is that the answer sounds right because the language is confident, while the underlying query is pulling from a structure the business can no longer fully explain. The risk is inherited, not invented

Every mature Anaplan model contains history: logic that made sense at the time, workarounds from a migration, assumptions created under pressure, mappings that were meant to be temporary, manual steps that never made it into documentation, ownership that was clear during implementation and less clear two years later.

Today, those issues are often manageable because people compensate for them. Someone knows the exception. Someone remembers the decision. Someone double-checks the output before it reaches the meeting. That is not governance. It is institutional memory and it is fragile control.

The problem with AI is not that it creates these weaknesses. It cannot conjure a stale assumption out of nowhere, invent poor data lineage where the data is clear,or make decision rights ambiguous if the organisation has already defined them. The problem is that AI can make existing weaknesses operate faster.

The four questions that test your planning environment

The next phase of AI adoption in planning should not start with a list of use cases. It should start with a test of the planning environment itself.

“Can you explain how the number is produced?”

Not whether the model can produce it. Whether a person can trace it from source data, through calculation logic, through assumptions, adjustments and approvals, to the final output. If the answer depends on one individual who has been in the business long enough to remember, that is not documentation. It is a risk.

“Do you know which assumptions are deliberate?”

Every planning process contains assumptions. The question is whether they are owned, reviewed and understood. Some are intentional: a commercial decision, a finance policy, a supply chain constraint, a leadership position. Others are inherited, added once under pressure and never revisited. AI cannot tell the difference. It will treat both as design intent.

“Is your data lineage clear enough to defend?”

AI is only as useful as the data environment it can see. Planning data is rarely neutral: it has timing, ownership, mapping, hierarchy and version issues. If the route from source to output is unclear, AI can still produce an answer, but the organisation may not understand what that answer depends on, and the confidence of the response becomes misleading.

“Have you decided where the human stays?”

This is the question most organisations have not answered properly. The useful distinction is not simply automated or manual. It is three-way: what can be automated, what should be augmented, and what must remain human. AI should take on low-value mechanical work where the rules are clear, the risk is contained and the output can be checked. It should support higher-value decisions by surfacing options, explaining movement, and pulling together context. And some decisions must stay human, because accountability cannot be automated away.

Organisations that answer this before AI is embedded will build trust. Those that answer it afterwards will discover their governance model during an incident. The governance dividend

The useful thing about this work is that it is not only valuable because of AI.

Documented logic, clear ownership, reviewed assumptions, visible data lineage, defined decision rights and a proper audit trail are not AI-specific requirements. They are what a strong planning environment should have anyway. They make handovers easier, audits less painful, month-end faster and change requests safer. They reduce dependency on individual knowledge and stop models drifting quietly away from the business they were built to serve.

• AI simply removes the option of ignoring the gaps.

This is where the conversation needs to shift. Ticket resolution, small enhancements and reactive support is too narrow a definition for the environment planning teams are now moving into. The real value is active governance and optimisation. Is the model still aligned to the business, are assumptions still current, are users still following the intended process, are calculations still explainable, is the model ready for the next capability being added to the platform?

• That is not maintenance, it is readiness.

What to do with the runway

The market is moving quickly, but planning teams still have a window. AI agents are arriving in stages, and most organisations will not move from manual to fully agentic planning overnight. That runway should be used properly.

In the next 90 days:

• Document your model logic. Identify the areas where only one or two people can explain how the number is produced. That list is your risk register.
• Audit your assumptions. Separate the deliberate from the inherited. Retire what is stale, document what remains and assign clear ownership.
• Review your data lineage. Make sure the route from source to output is visible, understood and defensible.
• Define decision rights. Decide which activities could be automated, which should be augmented and which must stay human. Set thresholds before you need them.
• Check your governance cadence. If assumptions, logic, ownership and usage are not being reviewed regularly, the model will drift.
• Ask the people who challenge the numbers. Finance leaders, commercial owners, supply chain teams, auditors and executives will all have different concerns. Those concerns should shape your AI readiness more than any vendor roadmap.

None of this requires an AI agent to be live tomorrow. All of it makes AI more valuable when it is.

Our position

We have spent more than a decade helping organisations build planning environments they can trust. As AI becomes more embedded in enterprise planning, that work does not become less important. It becomes the foundation.

The technology will keep moving. Agents will become more capable, platforms will automate more tasks, planning teams will be given new ways to analyse, explain, forecast and act. But the organisations that benefit most will not be the ones that simply adopt the newest capability first.They will be the ones whose planning environments are clean enough to trust, governed enough to defend and actively managed enough to keep pace with change.

That is what Bedford Advantage is designed to support. Not passive support after go-live, but regular review, documented improvement, model optimisation and governance discipline, so the Anaplan environment continues to reflect the way the business works.

AI readiness is not a separate project. It is the condition your planning environment is either in, or it is not.If the foundation is weak, AI will not hide it. It will expose it.

What to read and do next

Where to start. Our Anaplan Value Check takes two minutes and ten questions. It assesses the strength of your planning foundations, shows you where the gaps are, and gives you a practical view of what to address first.

If this article felt recognisable, we would value the conversation: reach us at info@bedfordconsulting.com or follow Bedford Consulting on LinkedIn.