Here is a question worth asking in every organisation: what percentage of your team's time is spent actually making decisions, the judgements, approvals, and recommendations that ultimately drive your business forward?
Not chasing documents, sending emails, cross-referencing data, validating inputs, or copy-pasting between spreadsheets. The time spent doing the specialist task that leads to the outcome.
In almost every organisation we have worked with, the answer is somewhere around 30%. The other 70% is the effort just getting to the part that is of true value. Essential effort, but effort that does not produce the thing the business is here to do.
We call this the Decision Pyramid. And understanding it is the first step to fixing it.
What is the Decision Pyramid?
Every complex decision, whether it is validating a claim, approving a funding request, or assessing an application, follows the same pattern.
The bottom 70% is effort. Collating and synthesising data from scattered sources. Chasing documents, extracting facts from PDFs, emails, spreadsheets, and forms. Cross-referencing, validating, organising. This is where organisations spend the vast majority of their time and money. It is necessary work, but it is not where the value is.
The top 30% is value. The actual decision. The judgement call. The approval, the recommendation. This is what the business exists to do and the primary thing that specialist skills are employed to do, and it is the part that gets squeezed because everyone is drowning in the work below it.
The ratio is inverted: maximum effort at the bottom, but the biggest value is at the top.
Why does the Decision Pyramid persist?
This is not a technology problem in the traditional sense. Most organisations have invested heavily in systems, platforms, and tools. The problem is that those investments often digitise the effort without reducing it. You have moved from paper files to digital files, but someone still has to read them, extract the relevant facts, often synthesise or aggregate them, and feed them into the next step.
The data preparation layer, the bottom of the pyramid, is stubbornly manual because it is messy. Real-world data does not arrive ready for action or decision. It arrives as scanned PDFs with handwriting, emails with multiple attachments, spreadsheets with inconsistent formatting, and forms designed by someone who left three years ago. Data is spread across multiple sources and often has to be compared or aggregated. Handling this mess requires flexibility, interpretation, and context, exactly the things that traditional automation struggles with.
"Smart, expensive people spend their days as highly skilled data janitors."
And the decisions they are paid to make get whatever time is left over. Even worse, we shape careers around this inefficiency, build operating procedures, and work hard to retain experience so the processes do not fall over when people move on.
How does AI change the Decision Pyramid?
AI changes what is possible at the bottom of the pyramid. Modern document processing, intelligent extraction, semantic search, and synthesis capabilities mean machines can do in seconds what used to take hours or days. The messy, unstructured, context-dependent data preparation work that resisted traditional automation is exactly where AI excels.
But, and this is the critical point, AI alone is not enough to complete the pyramid. Many will claim that it does, but experience surfaces the truth.
In regulated industries, decisions must be deterministic, explainable, and auditable. A regulator does not accept "the model thought so." A board does not sign off on "probably correct." An underwriter cannot defend a decision with "the AI seemed confident."
AI infers. It approximates. It estimates. For the bottom of the pyramid, extracting data, synthesising information, identifying patterns, that is exactly what you need. For the top of the pyramid, the actual decision, you need certainty.
Where AI meets deterministic rules
The answer is not AI alone. And it is not traditional business rules engines alone either. It is where these two meet, and also work with the human-in-the-loop.
AI clears the bottom of the pyramid, extracting, synthesising, and preparing data at speed. Deterministic rules govern the top, ensuring every decision is accurate, explainable, and auditable. The human gets the best of both: speed and trust.
We call this the Decision Pyramid approach, AI that has been engineered to be production-ready. Not AI with guardrails bolted on after the fact. An architecture designed from the ground up to deliver decisions you can stand behind. AI for synthesis. Rules for governance.
What does this look like in practice?
Consider an insurance broker assessing a commercial risk. Today, they spend most of their time in the bottom of the pyramid: gathering proposal data from multiple sources, rekeying information from PDFs into quoting systems, cross-referencing risk details across emails and documents. The actual value they provide, judgement, advice, relationship, gets whatever time is left. In any margin-sensitive industry, this imbalance can mean the difference between survival and outright success.
With the Decision Pyramid model applied, the system knows where to get the data, then extracts, synthesises, and structures it. A deterministic rules engine handles the things that cannot be incorrect, calculations, validations, decisions, and leaves an audit trail.
The broker's time shifts from data preparation to nurturing client relationships and providing advice, which is exactly what they were hired for, because AI accelerated, rules validated, and together they surfaced the facts the broker needs to do what they do best - service and recommend.
The same pattern applies across almost all verticals. A government compliance team processing regulatory submissions. A service provider assessing claims against complex eligibility criteria. The pyramid is the same. The opportunity is the same.
The bottom line
Every complex operation has a Decision Pyramid. Most of the time, effort, and cost sits at the bottom, in data preparation that is necessary but not valuable. The decisions at the top, the things organisations actually exist to do, get squeezed.
Flip that ratio and the impact is immediate. Teams do more with less, not by working harder, but by eliminating the work that should not require a human in the first place. Operating costs compress because the most expensive line item in any knowledge business is skilled people doing unskilled work. Growth becomes profitable because you scale the decisions, not the headcount.
And there is a benefit that does not show up on a spreadsheet: people start doing the work they were actually hired to do. Underwriters underwrite. Analysts analyse. Brokers advise. The noise that consumed their days, the chasing, the rekeying, the cross-referencing, is handled. That is not just efficiency. That is retention, engagement, and a team that can see the point of what they do every day.
AI clears the effort. Deterministic rules govern the decision. Together, they do not just save time, they change what your organisation is capable of.