QDOAA: The Framework That Prevents AI Waste in Finance

Why Most AI Projects in Finance Fail

McKinsey's research puts the AI project failure rate above 70%. Gartner's figures are similar. Ask most consultants why, and they'll tell you the problem is change management, or data quality, or executive buy-in.

Those are symptoms. The root cause is almost always simpler: companies automate broken processes.

When a finance team runs a 47-step month-end close that takes eleven days, the instinct is to ask "which parts can AI speed up?" That is the wrong question. The right question is: why are there 47 steps?

If you automate a process with unnecessary steps, duplicate approvals, and inherited workarounds from a system you replaced three years ago, you get a faster broken process. The waste is now automated. The cost is now harder to see. And you've just spent £80,000 on a problem that required a spreadsheet audit and a whiteboard session to fix.

This is not a technology problem. It is a sequencing problem. AI is step five — not step one.

What QDOAA Is

QDOAA is a five-step framework applied before any automation decision is made. The acronym stands for:

• Question — Challenge why each step exists at all
• Delete — Remove steps that cannot justify their existence
• Optimise — Improve what remains, manually, before adding technology
• Accelerate — Make the optimised process faster without adding headcount
• Automate — Apply AI to what is left

The sequence is deliberate. Each step depends on the one before it. Skipping to Automate without completing the first four is how you end up with sophisticated AI running on a process that should have been redesigned in 2019.

"AI is just one tool in the system. The system has to be worth automating first."

Step 1: Question

Take every step in a given process and ask one question: why does this exist?

Not "how does it work" or "who owns it." Why does it exist? What would break if it stopped happening tomorrow?

This sounds obvious. It is not practised. Most finance teams inherit processes from previous systems, previous controllers, previous crises. The institutional memory that explains why something was added has often left the building. What remains is the process and the assumption that it is necessary.

Common examples of what the Question step uncovers:

• A daily bank reconciliation run at 8am because a previous CFO wanted the figure before the morning meeting — a meeting that was cancelled eighteen months ago
• A manual VAT cross-check step that duplicates a validation the ERP has been performing automatically since the last upgrade
• A sign-off requirement from a department head who delegated that authority two years ago and no longer reviews the requests
• A weekly accruals report emailed to four people, two of whom have left the company

Question everything. Accept no answer that begins with "we've always done it this way." That is not a reason — it is the absence of one.

Step 2: Delete

Remove every step that failed the Question test.

When we apply QDOAA rigorously across a mid-market finance function, the Delete step typically eliminates 30 to 40 percent of process steps. Not streamlines — removes entirely. Steps that add no value, serve no current requirement, and exist purely because no one asked the first question.

The resistance here is real. Finance teams are risk-averse by training, and deletion feels like exposure. The instinct is to keep the step "just in case." Push back on this. If a step cannot be justified by a current control requirement, a regulatory obligation, or a demonstrable downstream dependency, it should not survive the Delete phase.

A practical example: a client's purchase invoice approval workflow required three sequential sign-offs for invoices under £500 — a threshold set when the company had twelve employees and the CFO signed everything personally. At 340 employees, those three approvers were adding an average of four working days to payment cycles on invoices that represented less than 2% of total AP spend. Deleting two of the three sign-offs removed four days from the average payment cycle and eliminated a recurring creditor relations issue that the AP team had been managing manually for two years.

No AI required. A decision required.

Step 3: Optimise

Optimise means improving the remaining steps manually — without adding technology.

This is the step most transformation projects skip. They move from identifying inefficiency directly to selecting a tool. The problem is that a tool applied to an unoptimised process embeds the inefficiency at a lower cost per transaction, which makes it invisible and harder to challenge later.

Optimise asks: given that this step must exist, is it being done in the right sequence, by the right person, with the right inputs, at the right time?

Common optimisations at this stage:

• Reordering steps so that validation happens earlier, reducing rework downstream
• Consolidating inputs so that a step that required data from three separate sources can be completed with one standardised template
• Reassigning steps to the role with the right access and context, rather than the role that historically owned the task
• Establishing clear handoff criteria so that the next step does not begin until the preceding one is genuinely complete

Optimise does not require new software. It requires clarity about what each step is for and discipline about how it is executed.

Step 4: Accelerate

Acceleration is about throughput without headcount. The process is now leaner and better designed. Accelerate asks: can we make this faster using tools that are already available or low-cost to introduce?

This includes things like:

• Standard templates that eliminate the reformatting step before a file can be uploaded
• Batch processing schedules that group similar transactions rather than processing each individually
• Keyboard shortcuts, macros, or basic workflow rules within your existing ERP that no one configured
• Parallel rather than sequential processing where steps have no genuine dependency on each other

Acceleration is not AI. It is removing friction from a process that has already been cleaned and optimised. The distinction matters because Accelerate changes do not require a business case, a procurement cycle, or an implementation project. They require someone with the right access and a few hours.

Step 5: Automate

Now — and only now — consider AI.

At this point the process is justified (Question), trimmed (Delete), correctly designed (Optimise), and running at pace (Accelerate). What remains is a process worth automating. The AI you introduce will operate on clean inputs, in a logical sequence, with meaningful outputs.

The automation yield at this stage is also materially higher. Automating a 12-step process that has been through QDOAA delivers greater time savings than automating the original 47-step version — because the steps that remain are the ones that genuinely require processing, not the accumulated overhead of years of workarounds.

Appropriate automation at this stage might include:

• Intelligent document processing for invoice capture and GL coding
• Anomaly detection across transaction data to flag exceptions before month-end
• Automated reconciliation matching with exception-only human review
• AI-assisted variance commentary generation from structured financial data

Each of these has a clear input, a defined output, and a measurable error rate. That measurability is possible because the process was cleaned first.

A Practical Example: Month-End Close

Consider a mid-market company running a month-end close that takes eleven working days. The process has 47 documented steps across four teams. The Finance Director has asked whether AI can compress the close to five days.

Before any automation discussion, we apply QDOAA.

Question: We work through all 47 steps. Eight cannot be justified by current requirements. Four duplicate controls that the ERP now performs natively. Three relate to a reporting format that was retired when the company changed reporting lines two years ago.

Delete: Fifteen steps are removed. The process is now 32 steps. No technology has changed. Two days of elapsed time disappear because the steps being deleted were sitting in queues waiting for approvers who were not reviewing them anyway.

Optimise: Of the 32 remaining steps, nine are resequenced so that data validation happens at the point of entry rather than at consolidation. Four steps are consolidated using a standardised input template. The revised process has 23 steps with significantly fewer rework loops.

Accelerate: Batch processing is configured for intercompany eliminations. Three ERP workflow rules are activated that were never switched on post-implementation. The process is now 18 steps, running with fewer handoffs and cleaner data at each stage.

Automate: AI is applied to three specific steps: accruals calculation from contract data, GL variance commentary, and the reconciliation matching queue. The process now has 12 steps where human judgement is genuinely required.

The close now runs in four and a half days. The AI contribution was real and material — but it accounts for approximately 30% of the total time saving. The other 70% came from the first four steps of QDOAA, which required no AI budget at all.

Why This Matters Commercially

AI budgets are finite. The average mid-market finance function spends between £40,000 and £150,000 on AI and automation initiatives in a given year. How that budget is sequenced determines whether it produces measurable return or becomes a line item that finance leaders struggle to justify at the next board meeting.

Applying AI to an unoptimised process is the equivalent of fitting a turbocharger to a car with flat tyres. The engine is more powerful. The car goes nowhere faster. And you have spent money that could have fixed the tyres several times over.

The commercial case for QDOAA is straightforward: the first four steps cost time and expertise, not software licences. The optimisation work is typically recoverable within the first quarter of the revised process running. The automation step, applied to a clean process, delivers ROI that is both higher and more predictable than automation applied as the opening move.

Companies that come to FinanceFlo.ai asking "which AI tools should we buy" are asked a different question first: walk me through your current process. Where does it stall?

The answer to that question determines the roadmap. Sometimes the answer is AI. More often than not, the first three steps of QDOAA return more value than any automation project would have done — and they create the foundation that makes automation viable when the time comes.

Where to Start

Pick one process. Not the whole finance function — one process. Month-end close, AP invoice processing, expense management, intercompany reconciliation. Something with a defined start, a defined end, and a team that can tell you how many steps it involves.

Work through Question and Delete in a single workshop session. You will find more to remove than you expect. That alone will tell you whether the rest of the framework is worth the time.

If you want a structured starting point, the AI Readiness Assessment diagnoses the constraint pattern across your finance function and identifies where QDOAA will generate the fastest return. It takes twelve minutes and produces a report you can act on the same week.

AI is a powerful tool. It works best when it is the last decision in the sequence, not the first.