The team can show you what the tool produces. The harder question is what changes in the business when that output is used.
Artificial intelligence (AI) creates value when it improves an operating or financial result after the cost of checking, rework and running the workflow is counted. Adoption, licences and hours claimed as saved are useful signals. They do not establish that result on their own.
Start with the work, rather than the tool
A support team may draft responses faster and still spend the same time resolving each request. The difference could sit in verification, corrections or the cases the tool cannot handle. Measuring the draft alone misses that work.
I would start by defining an accepted outcome. In this example, it might be a request resolved correctly, with the agreed quality and no avoidable repeat contact. The unit of analysis becomes the complete job rather than the first output.
Then we can ask what changed in cost, volume, cycle time and quality. If speed improved while quality fell, the model needs to show both. If the team gained capacity, we need to say what happened to that capacity before treating it as a financial saving.
What the anonymised assessment found it necessary to examine
In an assessment for a technology enabled business, the leadership question was where to invest in AI, which controls needed strengthening and how to assess the return against cost. Several tools and initiatives were already in use. There was no single reliable view of spending, readiness, ownership and measurable value.
I reviewed the tool and vendor inventory against available financial records. That mattered because a platform with an AI feature is not the same thing as a cost incurred specifically for AI. Treating the whole platform bill as AI spend would distort the comparison. Ignoring the feature because it was embedded would miss part of the operating picture.
I also examined strategy, data, processes, technology, people and governance. Written policies were separated from evidence of everyday adoption. Candidate uses were compared on value, feasibility, dependencies and risk, with a distinction between a person using AI assistance and a system taking actions with greater autonomy.
The delivered roadmap organised decisions into proceed, prepare or defer, made owners and dependencies visible and identified the measurements needed for economics. The outcome was greater clarity for decision makers. The case does not claim realised savings, implemented pilots or an achieved return.
Make the next claim testable
Take one material workflow. Write down its current volume, cost, time and quality. Define the result that will be accepted, the person who owns it and the conditions under which a pilot would be stopped or changed.
Include the costs outside the model bill. Integration, training, oversight, exception handling and ongoing operation may matter as much as the licence or usage charge. Which costs belong in the calculation depends on the workflow; the model should show that choice.
Where you have no measured baseline, use an explicit scenario. State the assumptions and the range that follows from them. The next investment can then buy evidence about those assumptions. An estimate should not become a reported benefit simply because it is presented precisely.
The question for the next leadership discussion
Instead of asking every team for its AI success story, ask one team to show an accepted outcome before and after the change, including the work required to reach it. That is a smaller request with a more useful answer.
If that evidence does not yet exist, there is still a sensible next step: agree the baseline, owner and test before scaling the spend. AI Value Analysis is the JRDN engagement for screening use cases and examining the strongest economics. AI Readiness Assessment answers whether the business has the conditions to scale them.
Book a free 30 minute discovery conversation if you need to decide which question comes first in your business.
Written by Nirmalya Mukherjee, founder of JRDN Advisory. The case details come from the approved JRDN anonymised AI readiness and value assessment case study. The support workflow is an illustration, not a reported client result.