Why Monthly Business Reviews Break Down — and Why AI Alone Won't Fix Them

---

A Monthly Business Review exists for one purpose: to help leaders evaluate risk, allocate resources, and act. When the process starts consuming more time than it leaves for those activities, the instinct is to reach for automation — generate the reports automatically, use AI to summarise the commentary, remove the manual steps. That instinct is understandable. In most cases, it is solving the wrong problem.

The example below is drawn from direct operational experience and has been anonymised to protect organisational confidentiality.

The real constraint is upstream

In one large-scale MBR process covering infrastructure cost performance across multiple business units — 25+ stakeholders, a Director and VP audience, a fixed 18–19 day window from financial close to executive review — the data extraction side was not the problem. SQL queries ran on schedule, outputs landed in a master workbook, and the analyst refreshed the file to update all tables. That part worked.

What did not work was everything downstream. Commentary arrived via email, Slack, and Excel with no enforced format and no consistent structure. Non-responders were only identifiable at the deadline, by which point the time to chase them had already passed. Analysts spent the four to five day development window collating multi-format submissions, chasing missing drivers, reformatting explanations for executive consumption, and checking whether the reasoning submitted actually accounted for the variance it was supposed to explain.

By the time the document reached the lead for review, the team had run out of time for the thing the review was designed to produce: analysis, risk evaluation, and decision support.

Why AI makes it worse before you fix the workflow

The temptation at this point is to introduce AI to handle the messy middle: extract the key points from emails and Slack messages, summarise the commentary, flag what is missing. Modern AI can increasingly extract, summarise, and classify this kind of unstructured information.

But there is a structural problem with applying AI to an unstructured intake process. AI applied to freeform submissions can detect missing information after the fact. What it cannot do is solve the structural problem that allowed incomplete information to enter the workflow in the first place. Structured intake prevents that at source. Those are not the same intervention, and the distinction matters more than it might appear.

When a business unit submits a freeform email explaining a variance, an AI model can read it and identify what is present. What it cannot do is compel what is absent — the quantified driver, the forward outlook, the supporting approval for a budget increase that was not anticipated. If the explanation does not contain those things, the AI flags a gap. The analyst still has to go back to the team. The clarification cycle that was consuming the development window is still there. It has just been detected slightly faster.

Commentary quality is a different problem still. Whether an explanation is analytically coherent — whether the reasoning actually makes sense given the business driver trajectory, the prior month, and what leadership already knows about this team's cost profile — is a judgment that requires context AI does not have. That judgment stayed with analysts in the current process and should stay with analysts in any future-state model. The question is not whether to automate it. It is whether analysts have the time and space to do it well.

The right sequence

When I applied a structured transformation framework to this process, the assessment produced a clear dependency logic. Each intervention had to create the conditions for the next one to be reliable.

The transformation does not compress the fixed monthly calendar. What changes is how analyst effort is distributed within the same 18–19 days — away from coordination and collation, toward the analysis and decision support the review was designed to produce.

What remains human-led

The future-state model does not attempt to automate every activity. Human ownership remains essential for assessing whether commentary is analytically coherent, interpreting business-driver context, resolving exceptions that fall outside defined rules, evaluating risk and trade-offs, approving forecasts and recommendations, and making resource allocation and investment decisions.

AI supports the process by identifying patterns, drafting summaries, and flagging potential gaps. It does not replace accountability for the decision.

Expected outcomes

The proposed model is designed to produce earlier visibility of missing submissions, fewer clarification and reformatting cycles, more consistent treatment of standard variances, a clearer audit trail, more analyst capacity for analysis and decision support, and more leadership time focused on risks, priorities, and actions.

These are expected outcomes. The AI components described have not yet been implemented; benefits are directional rather than measured results.

What this means more broadly

The lesson is not that every reporting process needs AI. It is that every transformation effort should determine the right intervention before selecting the technology.

In any recurring business process where information quality, stakeholder coordination, and decision support are the primary constraints, the same dependency logic applies. Fix the structure of how information enters the system before trying to automate what happens to it. Automate the rule-based work before asking AI to interpret the ambiguous cases. Keep the judgment-intensive activities — the ones that require business context, pattern recognition, and accountability — explicitly under human ownership.

That sequencing is not a technology question. It is a workflow design question. And it is often the one that gets skipped.

---