Table of Contents
A fixed-fee AI discovery sprint rarely breaks because the interview guide is weak. It breaks because six senior stakeholders fill the calendar, frontline reality arrives too late, and Friday becomes a choice between rushed synthesis and unpaid work. Week-one capacity determines the credibility of the entire sprint.
Most teams schedule whoever is easiest to reach: the sponsor, several department heads, and one helpful technical lead. That produces a polished account of the organization’s AI ambitions, not evidence of how work, data, and informal AI use actually operate.
The second failure is arithmetic. One researcher can usually run five or six useful stakeholder interviews in a week after scheduling, preparation, note cleanup, and internal check-ins. Manual synthesis then adds several days—or weeks if recordings and notes sit untouched while delivery work continues.
This is the common failure mode: a small research team books eight interviews, but rescheduling and a security review whittle that down to five completed sessions and a Friday transcript backlog. The resulting read overstates leadership’s automation priority, because only one frontline operator ever got heard.
That is the broader stakeholder interview overflow problem in a sold AI readiness audit: scope is fixed before the real number of necessary voices becomes visible. A full calendar is not representative coverage.
I divide stakeholders by the evidence they hold. Leadership explains strategic intent and investment constraints; frontline staff reveal actual workflows and workarounds; IT and data owners establish what is technically available, permitted, reliable, and governable.
Six interviews can support a narrow diagnostic: two leaders, three frontline staff, and one IT or data owner. It cannot credibly represent a multi-function organization. For a normal fixed-fee sprint, I target 12–18 stakeholders and expand further when roles, regions, or business units differ materially.
Do not sample frontline staff solely through their managers. Ask leaders for nominations, then add people identified through workflow dependencies and interview referrals; otherwise, you get articulate champions while quiet skeptics and workaround-heavy teams remain invisible.
The wrong sequence leaves all technical constraints until Thursday or lets executive language shape every later question. Start with intent, test feasibility immediately, then investigate work.
At hand-run capacity, this schedule compresses quickly: five or six sessions consume most of the week before synthesis starts. I use AI-moderated interviews through Usercall when broader coverage is necessary, while retaining researcher control over the guide, probes, audience, and analysis framework.
Asking “Where could AI help?” invites speculative wish lists. I ask stakeholders to reconstruct a recent piece of work, show where information entered and changed hands, and explain the last exception that forced human judgment.
Behavior beats sentiment. “Our team is enthusiastic about copilots” is weak evidence; “seven analysts paste public filings into an approved model, then spend 20 minutes checking every citation” is usable discovery data.
A theme count alone is not research-grade synthesis. Ten mentions of “data access” may describe ten different problems: missing permissions, inaccessible formats, uncertain ownership, poor quality, or a policy nobody understands.
I then score opportunities on business value, workflow frequency, data readiness, adoption friction, and risk. The output should distinguish high-confidence actions from hypotheses requiring a prototype, data audit, or additional interviews.
Usercall once ran 100 interviews with internal stakeholders at a financial consulting firm to assess how AI was actually being used across the organization. The practical constraint was breadth: usage varied across roles, and investment decisions had to cover both tools and education; once interview volume was no longer the bottleneck, the firm could base priorities on organization-wide evidence rather than a handful of confident voices.
The cleanest sprint contract defines the decisions, deliverables, research window, and included synthesis—not an artificially tiny interview count. Five conversations may surface hypotheses, but dozens of voices create legitimacy when leaders must fund tools, change workflows, or enforce new controls.
My week-one standard is simple: hear all three decision layers, challenge the brief by Wednesday, synthesize continuously, and expand capacity before representation becomes the hidden compromise. That works whether you lead a boutique consultancy or are driving AI readiness inside your own company.
Usercall runs AI-moderated user interviews that collect qualitative insights at scale, with deep researcher controls and the depth of a real conversation—without the overhead of a research agency. Use it to expand stakeholder coverage, analyze qualitative evidence at research-grade depth, and intercept people at key analytic moments to explain the “why” behind operational metrics.
