Many companies ask which AI model to adopt to decide better. It’s the wrong question. The lever isn’t a more powerful model, but making the rules you decide by explicit (objectives, levers and constraints) so the AI can simulate the consequences of each option before the choice is made, and a person decides with the expected impact already in front of them. This is the discipline Gartner calls Decision Intelligence.

Why decisions don’t improve, even with a more powerful AI

It’s a paradox many companies know well. AI models get better at every benchmark, yet the decisions that matter, from what price to set to when to buy, from where to allocate budget to how many people to hire, don’t improve at the same pace. The bottleneck isn’t the model’s intelligence, but the way the company frames its decisions. A more powerful model reading implicit rules produces more fluent, more confident answers on top of the same ambiguities: not better decisions.

The numbers show the gap. Gartner estimates that over 40% of agentic AI projects will be cancelled by 2027, often not because the model is wrong but because decisions stay implicit. McKinsey finds that fewer than one company in ten takes agents beyond the pilot, and eight in ten point to data as the main obstacle to scaling. And while around 72% of companies have adopted AI, only 12% draw measurable value from it. The difference isn’t the tool, but the ability to turn data into governed decisions.

What it means to govern a decision with AI

Making better decisions with AI is, first of all, a change of definition. It’s not a faster answer to the same question, but a decision the company can govern, defend and improve over time. Three shifts set it apart.

From an answer to a governed decision. A useful recommendation doesn’t stand alone: it carries its source, its confidence interval and its expected impact, within a defined perimeter. The AI proposes, a person approves and stays accountable.

From understandable data to a modelable decision. Knowing the margin doesn’t yet tell you what to do to protect it. You need explicit objectives, levers and constraints: only then can a choice be simulated rather than guessed.

From AI that predicts to AI that compares. The value isn’t predicting what will happen, but comparing the possible alternatives against the same objectives and constraints, and proposing the one with the best expected impact.

A more powerful model doesn’t close a decision’s ambiguity: it makes it more confident. Closing it takes explicit, shared business rules.

For this to happen, an infrastructure capable of governing both definitions and decisions is required: data read once through a semantic layer, and connected to objectives, levers and constraints by a decision layer. It is the two-layer architecture we examined in a dedicated analysis, and the foundation for what follows: the application of this method to the recurring decisions of every business function.

The common thread: quantifying the cost of inaction

In almost every business decision there is a cost that appears on no invoice: the cost of a choice postponed or made on gut feel. Margin given away discount after discount, a purchase made at the wrong moment, a workforce sized on the average. It doesn’t show up in the year-end accounts because it is distributed and implicit. The method that makes decisions better with AI starts here: quantify that cost before the decision is made, separate the real levers from the merely apparent ones, and assign each action an owner, a timeline and an impact in euros. The examples below show the same method applied to seven recurring decisions, across retail and manufacturing.

Seven decisions AI makes better, area by area

Pricing and commercial margin

Price is the lever that acts fastest on margin, and also the one where the rules most often stay implicit. In retail, revenue can grow while margin erodes, discount after discount, with no one seeing the problem until the accounts close: in one case the erosion was worth 3.8 points of gross margin and a €7.1M gap, with €9.5M recoverable by making explicit how much margin is given away by category and promo mechanic.

Read more: Pricing for Retail

In make-to-order manufacturing, quotes are built on cost sheets frozen for months while real costs move: in one case this put €1.7M of margin a year at risk. A recommended price in real time, with an updated margin floor before the quote goes out, turns the discount into a governed choice.

Read more: Pricing for Manufacturing

Management control

Traditional management control photographs the margin once it is already lost. Turning it into a decision lever means reading the drivers week by week and acting before period close. In one retailer, revenue closed almost on budget but EBITDA fell by €3.6M, hidden until year-end: isolating the causes (markdown, stock, promotions) surfaced €3.8M of recoverable value.

Read more: Management control for Retail

In manufacturing the same principle applies to job-order margin: in one case €1.23M was missing versus plan, with 78% of the damage concentrated in four jobs out of sixty: recoverable only by acting while the job was still open, not after the fact.

Read more: Management control for Manufacturing

Production planning

Which orders to launch, in what sequence and on which lines is a margin decision, not just a scheduling one. In a multi-line plant, optimised plans went stale fast and gut-feel choices carried hidden costs: idle capacity, overtime, penalties. Breaking down the lost margin driver by driver turned a €9.5M-a-year gap into €6.9M of recovery, with dated actions assigned to an owner.

Read more:  Production planning

Procurement

In recurring purchases of raw materials and energy, the question isn’t only how much to pay but when to buy. Often “the price paid was right, the timing wasn’t”: a cost that appears on no invoice. By simulating a thousand scenarios over twelve-week horizons and comparing calendar orders with dynamic-threshold buying, the price gap becomes quantified savings and a weekly recommendation on quantity and timing.

Read more:  Procurement for Manufacturing

Marketing and customer value

In retail marketing the data is complete, but the real question is which decision moves the margin: where to allocate budget, which customers to retain, which promotions to stop. In one case a 25.5% churn against a reachable 20.5% was worth €16.5M of lost revenue, with promotions at 1.8x ROI versus a 3.0x benchmark; reallocating budget and retention onto the right behavioural clusters surfaced €10.1M of net value.

Read more:  Marketing for Retail

Workforce and HR planning

Sizing the workforce is as much an economic decision as pricing. In retail, a resource plan built on the average and frozen for twelve months, against demand swinging 20-30% a quarter, generated a €7.3M-a-year mismatch cost, of which €5M was correctable rigidity; acting on cross-training and rolling forecast lifted demand coverage from 75.8% to 91.3%.

Read more:  HR for Retail

In manufacturing the same logic protects production continuity: predictable retirements and critical roles left uncovered for 12-18 months become a multi-year plan that, under a budget constraint, raises critical-role coverage from 68% to 90%.

Read more →  HR for Manufacturing

Strategy and capital allocation

The biggest decisions, those where tens of millions are invested, are also the ones where the alternatives look “all defensible and none comparable.” Putting three options worth over €80M on the same scale, quantifying each one’s value and risk and the cost of every week of waiting, in one case was worth €24.7M of added decision value and a group EBITDA optimised from €62.6M to €87.2M.

Read more:  Strategy

How to start without rebuilding your systems

You don’t need to rebuild the data stack or launch a multi-year programme: the architecture sits on top of the systems the company already has (ERP, CRM, data warehouse) and the external data that matters. The starting point is a circumscribed decision whose value you can measure before extending the method. The path changes with the company’s maturity: those who want to understand where AI creates the most value start with an orientation on the processes and KPIs that matter, benchmarked against the sector; those with a precise problem seeking value in the short term adopt pre-configured agents on a recurring decision (pricing, management control, procurement, HR) with concrete results in weeks; those who want to change structurally how they decide build a model tailored to their own knowledge, rules and policies. In every case the principle holds: the AI proposes, people decide.

Key takeaways

• Making better decisions with AI doesn’t depend on a more powerful model, but on decisions made explicit: objectives, levers and constraints.
• Over 40% of agentic AI projects will be cancelled by 2027 (Gartner): the limit is often not the model, but implicit decisions.
• The value of AI isn’t prediction, but comparing alternatives against the same objectives and constraints and proposing the one with the best expected impact.
• The same method applies to seven areas: pricing, management control, production, procurement, marketing, HR and strategy, across retail and manufacturing.
• You start from a circumscribed decision, not from rebuilding systems. The AI proposes, people decide.

FAQ

How do you make better business decisions with AI?

By making the rules you decide by explicit (objectives, levers and constraints) so the AI can simulate the consequences of each alternative before the choice and propose the one with the best expected impact. A person approves and stays accountable. It takes modelable decisions, not a more powerful model.

Does a more powerful AI model lead to better decisions?

Not on its own. A more powerful model reading implicit rules produces more fluent, confident answers on top of the same ambiguities: reword a question and the recommendation can flip. Shared definitions, objectives and constraints are what make a decision genuinely better.

What is Decision Intelligence?

It is the discipline that models decisions as assets, connecting data, models and constraints to make every choice explicit, traceable and improvable over time. Gartner classifies it as a transformational technology. It is the approach WhAI, Vedrai’s Decision Intelligence platform, is built on.

In which business areas does AI help make better decisions?

In every area where a recurring choice has an economic impact: pricing and margin, management control, production planning, procurement, marketing and customer value, workforce and HR planning, strategy and capital allocation, across both retail and manufacturing.

Do you need to rebuild company systems to use AI in decisions?

No. The approach sits on top of existing systems (ERP, CRM, data warehouse and external data). You start from a circumscribed decision, measure its value and scale, without rebuilding the data infrastructure.

Does the AI decide, or do people?

People. The principle is “the AI proposes, the human responds”: the agent proposes scenarios within a defined perimeter, an owner approves, and the outcome is compared with what was expected. Trust comes from traceability, not from an after-the-fact explanation.