AI engagements that work don't start with a model or a tool. They start with one conversation.
The build-side framework (the eight steps from purpose through evaluation) sits downstream of a question most agent projects skip: where, specifically, is this worth doing? Discovery is the upstream work. It is not strategy in the abstract. It is a structured conversation that surfaces a small number of high-leverage AI opportunities inside an organization and turns them into something concrete enough to build.
This is the framework we use.
Three layers where AI creates value
Every discovery conversation should cover three layers. They map to different stakeholders, different success criteria, and different mechanics for how AI actually lands.
Organization. Cross-team systems and workflows. Examples: portfolio monitoring across business units, executive briefings synthesized from many data sources, benchmarking one part of the business against another. The value at this layer is leverage at the firm or enterprise level.
Function. A specific function or business unit. Examples: customer support, deal sourcing, revenue cycle, financial reporting, document review. The value at this layer is operational density inside a workflow that already exists.
Individual. Executive personal effectiveness. Examples: preparation for important meetings, decision stress-testing, recovering hours from low-leverage prep work. The value at this layer is reclaiming the most expensive time in the organization.
Most teams default to one layer (usually Function) and miss the other two. The strongest opportunities often live at the seams: a single workflow that produces an organizational artifact while saving an individual several hours per week.
Three kinds of questions
Inside the conversation, three question modes do different work. A good discovery moves through all three, in order.
Dig questions are diagnostic. They get the executive describing how work actually happens today. "Walk me through the last deal you closed. Where did it feel smooth, where did it feel like grinding through paperwork?" The job is not to ask about AI. The job is to surface the friction in vivid, specific detail.
Spark questions are imagination. They help the executive picture an outcome without using technical language. "If you woke up Monday morning and there was a one-page briefing in your inbox with the three things that need your attention this week, what would change?" Spark questions reframe what's possible in terms of outcomes, never in terms of tools.
Close questions move the conversation toward a concrete next step. "Of everything we talked about today, what's the thing you keep coming back to?" Close questions are how discovery turns into something an organization will actually fund.
Lead with too many close questions and the executive feels rushed. Lead with too many spark questions and the conversation floats. The ratio matters: dig first, spark second, close last.
Watch for the moments they light up
The most important signal in a discovery conversation is not what the executive says they want. It is the moment they light up when describing something. Body language shifts. They lean in. They start telling a story instead of answering a question.
That moment is the highest-priority proof-of-concept candidate.
Stated priorities are often what executives think they should care about. Lit-up moments are what they actually wish they had. Engagements built around the latter tend to land. Engagements built around the former tend to stall in the second meeting.
The "brilliant chief of staff" frame
Most senior leaders do not think in terms of "agents" or "RAG" or "fine-tuning." They think in terms of people. The cleanest way to make AI concrete for a non-technical executive is to ask one question, and then listen carefully:
The answer is almost always the highest-value proof of concept. The question bypasses the vocabulary gap entirely and goes straight to outcomes.
Demonstrate, don't present
The principle that matters most: come back with a working demonstration on the executive's own data, not a deck about AI.
A deck explains what AI could do. A demonstration shows what it just did for them, in their environment, with their inputs. The first is interesting. The second is the difference between an exploratory conversation and a funded engagement.
The cost is modest: a few days of work to build the demonstration. The signal is everything.
What to leave the room with
The handoff that converts discovery into engagement is small and concrete: one real piece of the executive's data. A portfolio report. A board deck. A sample dataset. Anything the demonstration will be built on.
When the executive hands that over, the conversation has shifted from interesting to committed. When they do not, the engagement is one or two more conversations away from real, no matter how warm the room felt.
Connecting discovery to build
Discovery surfaces what to build. The eight-step build framework governs how. Two halves of the same engagement, in that order.
The order matters. Building before discovery produces well-engineered systems for problems that did not need solving. Discovering before building produces a clear thesis and a clear demonstration before anyone writes code. The second sequence is the one that compounds across an entire roadmap.
At FutureInSites, this is the framework we open every AI engagement with. Three layers, three kinds of questions, one lit-up moment, one demonstration. The eight steps come after.