Most organisations I speak with are not short of AI ideas. They have pilots, vendor demos, a steering committee and often a mandate from the board. What they are short of is AI that has changed how work actually gets done.
The pattern is familiar. A team picks a promising tool, runs a proof of concept, gets an encouraging result on a small sample and presents it. Then it stalls. The workflow around it was never redesigned, the data it needs lives in three systems, nobody owns the decision it was supposed to improve, and the people expected to use it were never part of the design. Six months later the pilot is still a pilot.
The problem is rarely the technology. It is the order of operations.
Start with the work, not the tool
Every AI initiative should begin with a piece of work that matters: a decision made too slowly, a hand-off that loses information, a review that consumes scarce expert time, a customer interaction that frustrates both sides. Start there, and the technology becomes a means of changing that work, and you can tell whether it did.
Start with the tool, and you end up searching for problems it can solve, then measuring the tool instead of the work.
Four questions before any AI investment
Before approving an AI initiative, I ask leadership teams to answer four questions. They are simple to state and surprisingly hard to answer well.
1. Value: where exactly will AI improve the work?
Name the measure that will move: cycle time, quality, cost to serve, insight or experience. “Productivity” is not an answer. “Cut the turnaround on a specific approval from days to hours” is.
2. Workflow: what work, decisions and hand-offs must change?
This is the question most often skipped, and the one that decides success. If the output of a model lands in the same inbox, gets reviewed by the same committee and waits for the same approval, nothing changes. Map the current flow, decide who does what differently, and redesign decision rights around the new capability.
If you can’t answer the workflow question, you don’t have an AI project. You have a demo.
3. Data: what information, controls and human review are required?
Identify the data the work depends on, where it lives, how reliable it is and what must be governed. Decide where a human stays in the loop, what gets checked and what gets logged. Responsible use is not a compliance appendix. It is part of the design.
4. Adoption: how will leaders, teams and measures reinforce the new way?
People adopt what their leaders ask about and what their measures reward. Build training into the live work, change the review cadence so leaders see the new metrics, and retire the old path once the new one proves itself.
What this looks like in practice
Take an illustrative case: a shared-services team that checks supplier invoices against contract terms.
- Value: time spent on exceptions, and leakage recovered.
- Workflow: AI pre-screens every invoice and flags exceptions with reasons; reviewers work only the exceptions; approval limits are reset to match.
- Data: contract terms are extracted and validated once; a sample of auto-cleared invoices is audited every week.
- Adoption: team leads review exception trends weekly, reviewers learn the new queue on live work, and the old manual checklist is retired once audit results hold.
Nothing in that list is exotic. What makes it work is that the workflow, controls and measures are designed together with the technology.
A 30-day starter plan
| Week | Focus | Output |
|---|---|---|
| 1 | Pick one outcome | One business-critical workflow, a visible measure and a leader who owns it |
| 2 | Map the real work | Decisions, hand-offs, data sources and delays, walked with the people who do the work |
| 3 | Design the future flow | Where AI assists, where humans decide, what gets checked and how change is measured |
| 4 | Pilot in the live workflow | Results against the baseline, and a clear decision to scale, adjust or stop |
Thirty days will not transform an enterprise. It will tell you, with evidence, whether a use case deserves to scale, and it builds the muscle your organisation needs for the next one.
The leadership question
The organisations getting real value from AI are not the ones with the most pilots. They are the ones where leaders insist on changed work and measured value, and where AI is treated as part of the operating model rather than a side project.
Start with the work. Then apply AI.
