Most AI roadmaps stall at discovery. Executives have the mandate and the budget but they still can't say where AI should go first.
Enterprises usually start with workshops and employee surveys, and strategic guidance helps prioritize what comes out of them. Months in, the list of use cases turns out to be built on what people remember, not on how the work happens.
Why discovery needs more than consultation
Workshops and consultation are how most enterprises align AI initiatives with business goals. Both depend on an accurate picture of how work moves through the organization. That picture has never been available.
Ask someone how long monthly reporting takes and they'll estimate the time in Excel. The hours spent chasing data from other teams first never make it into the estimate. People underestimate repetitive work and forget the small inefficiencies they hit every day. In Fluency pilots, teams took 45 to 50 minutes on tasks they had self-reported at 20 to 30. Estimates like this put the use cases in the wrong order. The tasks that sound small in a workshop usually take the longest.
Workshops only turn up the pain people already know about. If a process has always been inefficient, the team has adapted to it and built workarounds. It never comes up in a discovery session because nobody thinks of it as a problem.
Discovery sessions also reach only a sample of the organization. You hear from a few people in each department while the workflows run through hundreds of employees across many functions. The biggest opportunities are in the handoffs between teams where no one person sees the whole sequence.
The strategy coming out of this process can be sound and still be built on a partial picture, because nobody has ever seen the work patterns underneath it.
What visibility into enterprise workflows reveals
When discovery is grounded in observed work, a process like monthly reporting shows up as a record of every step: the data requests, the follow-ups, the manual consolidation, the error correction, and the time each one took. The business case for automating any of them starts from that record.
A handoff between two teams is invisible in an interview because neither side sees the other half and most of the coordination around it never makes a calendar. When Microsoft analyzed its own Microsoft 365 usage signals in 2025, 57% of meetings turned out to be ad hoc calls with no invite. In observed work, a handoff is a step with a duration and a waiting period, whether anyone scheduled it or not.
The best automation candidates are the repetitive processes people have stopped noticing. In one online retail shared-services deployment, Fluency found that 85% of CRM usage was record-keeping. It was one of more than 250 opportunities in that deployment, worth $1.6 million to $3 million in annualized recovery. Nobody had raised it in a workshop.
Some people are consistently faster or more accurate at the same task, and a discovery interview won't tell you why. Observed work shows which steps their version of the workflow includes and which ones it skips. Once you can see the difference you can teach it.
The continuous intelligence advantage
Discovery run as a project produces a snapshot. By the time the findings are synthesized and the business cases built, the workflows have changed. When a team restructures or adopts a new tool, suddenly last quarter's recommendations stop matching the work.
Fluency's work ontology is a map of how work runs across the enterprise, updated as the work changes. When a team finds a faster way to do something, you can see it and copy it to the other teams while the method is still current.
Business cases stay relevant because they're built on current data, and discovery doesn't stop. As AI tools get deployed, the next candidate shows up in the same map.
From months to weeks
The two timelines below are illustrative rather than measured. The first assumes discovery runs on workshops and interviews, the second on observed work from the first week.
Traditional path:
- 4-8 weeks: Workshop planning, sponsor engagement
- 6-10 weeks: Interviews, surveys, data collection
- 4-6 weeks: Analysis, synthesis, prioritization
- Total: 4-6 months before pilot selection
With strategic guidance + work visibility:
- 1-2 weeks: Data collection across pilot teams
- 1-2 weeks: Strategic analysis with actual operational patterns
- Total: 3-5 weeks to validated targets
Discovery run on observed work turns up candidates as they appear, priced from current data.
Where does Fluency fit in AI use case discovery?
Fluency is a work intelligence platform that shows large enterprises where AI will have the biggest impact by finding the best work to automate first. It observes work at the point of execution, deploys AI agents, and measures results against an observed baseline. Discovery frameworks assume an enterprise already has that record of how work runs.
A desktop agent captures work across the applications a team uses, with no integrations and no survey round, and builds the work ontology, a living map of how the enterprise runs.
The hours spent chasing data before anyone opens Excel appear in Work Explorer as work, because they were work. Nobody has to remember them.
Every automation candidate in Opportunities shows projected hours and cost. The business case starts with a number. The same observation that found the opportunity measures whether the deployment improved it against how the work ran before. Most stalled AI pilots never had that before-and-after measurement.
One healthcare organization mapped 30 core workflows with Fluency and found more than 50 automation candidates, ranked by return and worth over 35,000 hours a year between them. The largest was in client onboarding, where staff were copying client data by hand from the CRM into the internal finance tools. It had never come up in a workshop.
The infrastructure that enables strategy
Consultants and the frameworks they use tell you to start where the impact is biggest and to prove value before you scale. Both instructions need an input the consultant doesn't provide. Ranking use cases by impact means knowing where the hours go today. Proving value means measuring an accurate baseline before anything changes.
Any change program should start from a record of the work at the level of steps and handoffs, kept current.
Interviews and workshops record what people remember, once. The record is out of date by the time anyone acts on it. Every ranking and business case built that way repeats the same error. When MIT measured the returns on enterprise generative AI in 2025, 95% of organizations were getting zero return. Half of GenAI budgets had gone to sales and marketing while back-office automation was returning more. The authors put it down to measurement. Sales results are easy to attribute and back-office ones are not. The money followed what was easiest to measure, and the back-office opportunities with the most hours in them were ignored.
Start with observation, then build the use-case list against a baseline you measured before anything changed.
