There is a single question that separates serious AI engagements from expensive disappointments. It is not about technology. It is not about architecture or tooling. It is a question about outcomes — and most AI proposals never answer it.
I was recently in a proposal meeting with a major global media company. We had spent weeks scoping the engagement — navigating compliance requirements, technology considerations, data anonymization protocols, and more. The proposal was detailed. The scope was defined. And right before we got to numbers, the client asked me one question I wasn’t expecting:
“How will we know when this actually worked?”
It is a consulting proposal. I love those questions because they keep everybody disciplined.
That question is the most important question in any AI project. And most proposals never properly answer it.
Most AI proposals are full of deliverables:
- The number of agents built.
- Workshops delivered.
- Platforms configured.
They are thorough on implementation and thin on outcomes.
But when a serious decision-maker asks, “How will we know when this works?” — they are not asking about deliverables. They are asking about business outcomes.
This is where AI projects break down before they even start.
A vendor designs a scope around what they can build. The client approves based on what sounds plausible. Six months later, nobody agrees on whether it worked — because nobody had defined what working actually means.
The data reinforces this gap. McKinsey found that among 12 generative AI adoption and scaling practices, tracking well-defined KPIs for generative AI solutions is the practice most strongly associated with bottom-line (EBIT) impact. Yet fewer than 1 in 5 respondents say their organizations actually track KPIs for their generative AI solutions (McKinsey, 2025).
When that question came across the table, my answer was specific.
We measure two things: the time that disappears from a defined workflow, and the business output that increases because of it.
Everything else — the technology, the architecture, the tooling — is noise. It does not bring numbers. It brings implementation.
The reason that answer landed is because it was already in the proposal. Before I walked into the room, I knew which workflow we were targeting. I knew what it cost the business in human hours per week, and I knew, based on comparable implementations and research done by our AI agents, what a reasonable outcome looks like in 60 days.
That preparation is what made the answer land.
I have seen the same dynamic play out with my client, Krishome. They came in thinking about broad AI implementation across the organization: tools, training, strategy. In our first meeting, I asked one question:
“What is your single most expensive manual workflow today?”
And then the story became the same. That one question narrowed the engagement from a broad vision to a specific, measurable target. The result was projected $2M+ in annual savings and a 35% planned increase in process efficiency — because we defined what working meant before we started.
Here’s the practical takeaway: ask “How will we know when this works?” before anything else. Before the architecture, tooling, or timeline.
A good answer names three things: a workflow, a number, and a timeline. If the vendor cannot answer that question specifically and confidently, the proposal is not ready. If the client cannot answer it either, the engagement is not ready.
The best AI projects are not defined by what gets built. They are defined by what changes — in human hours saved, in output increased, in decisions made faster. That is what working actually means.