Part 1. Workflow Discovery
Most organizations have individuals who’ve figured out how to use AI to do their own work faster and better. Translating that individual productivity into organizational value is challenging.
The first step is to map how work gets done in your organization today. This includes: how time is used, where decisions are made, when information is shared, and how clients/customers/users access value.
By the end of phase 1 you will have:
- A clear inventory of your organization’s core workflows
- An honest picture of where time is being spent
- A map of your technology stack — what’s paid for, what’s actually used, and what people have built on the side to work around the gaps
- A prioritized list of AI opportunities, scored by effort and impact
- The clarity to make real build/buy/partner decisions in Phase 2
How To Start:
1. Map core functions and pick the 2-3 that are most central to how value gets created or discovered. If you don’t know this, interviews with key decision makers across the organization is sufficient for this qualitative analysis. Larger organizations may choose to include a high-level ROI analysis at this stage.
2. Appoint discovery lead — one person who owns this initiative. They schedule interviews, coordinate recordings, synthesize findings, and produce deliverables. They need real authority and visible backing from senior leadership.
3. Set a timeline. You can use the table below to calibrate time and cost expectations, depending on the size of your organization.
| Org Size | Approach | Timeline | Rough Cost |
|---|---|---|---|
| 1–5 people | Self-led with AI-assisted synthesis | 2–4 weeks | $0–5K |
| 5–50 people | Self-led, optional fractional support | 4–6 weeks | $5K–25K |
| 51–500 people | Working group + external facilitator | 6–12 weeks | $25K–100K |
| 500+ people | Dedicated internal lead or external engagement | 3–6 months | $100K–1M+ |
Phase A: People
Here’s what you’re trying to find out:
- Where does time actually go on a daily, weekly, and monthly basis
- How do teammates store and use proprietary knowledge and information
- Where are decision bottlenecks and how are they distributed in the org
- What manual work is invisible on the org chart but real in practice
Step 1 — Conduct Interviews
The purpose of these meetings is to understand the high-level opportunities for augmentation and automation with AI. You should ask each person:
- What tasks consume the most of your time each week?
- What are your top three manual pain points?
- What tools do you use daily?
- What tools do you wish you had?
- Where do you most often feel stuck waiting on someone or something?
Use a call recording and annotation software. Analyze responses with an LLM and share a 1-2 page takeaway document with your key stakeholders.
Tools you can utilize:
- Otter.ai / Fireflies — record and transcribe. Pick one and use it for everything.
- Claude / GPT — after each interview, feed the transcript in with this
Step 2 — Build A Time Map
From your interviews and transcript data, build a simple table:
| Task | Role(s) | Hours/Week | Manual? | Pain Level (1–5) |
|---|---|---|---|---|
| Task #1 | Ops Manager | 6 hrs | Yes | 4 |
| Task #2 | Account Manager | 4 hrs | Yes | 3 |
| Task #3 | Analyst | 8 hrs | Yes | 5 |
This doesn’t need to be perfect. You’re looking for the tasks with high hours, high manual effort, and high pain — those are your initial intervention candidates. Keep this to 3-5 tasks maximum for now. Once harvesting the “low-hanging fruit”, the above process can be repeated in an iterative fashion as a continuous transformation engine within your organization.
Phase B: Process
Now you map how the work actually happens — not how it’s documented, but what people actually do step by step. Here’s what you’re trying to find:
- Where handoffs (data, context, know-how) breaks down between people
- Where information has to be manually moved between systems
- Where decisions are rules-based vs. where they need human judgment
- Where errors enter the system and where rework happens
Step 3 — Screen record the core workflows
For each high-priority workflow from your time map, have the person (or people) who do it share their screen and actually complete a live task.
A 20-30 minute screen recording of someone actually doing their job is worth more than an hour of them describing it. You will see things such as — tab-switches, parallel work, and copy-pasting that flies under the radar.
Aim for 5–10 recordings covering your highest-volume and highest-pain workflows. In smaller organizations, 2-3 is sufficient.
Tools:
- Tango / Scribe — auto-generate step-by-step documentation with screenshots as someone works. Best for straightforward linear workflows.
- Loom — narrated screen recordings. Better for complex workflows where context matters.
- Celonis / UiPath Process Mining — for organizations with mature ERP or CRM systems. Reads actual event logs and maps real workflow patterns across thousands of transactions. Overkill below ~200 people; powerful above it.
Step 4 — Map The Process
Bring the key people from each function together for 2–3 hours. Build the process map together on a shared board. For each workflow:
- what triggers it
- who does what
- in what order
- where does it touch other people or systems
- where does it most often break or slow down
Tools: Miro or Lucidchart.
Step 5 — Prioritize Opportunities
As you map, mark every instance of:
- Manual data movement — anything being copied, reformatted, or re-entered between systems
- Repeated content creation — the same report, summary, or document being drafted from scratch on a recurring basis
- Classification and routing — inbound requests, tickets, or documents being sorted manually
- Research and lookup — people spending time finding information that exists somewhere in your systems
- Human bridges — a person whose job is largely to move information from one system or team to another
These five categories account for the majority of high-ROI AI opportunities.
Phase C: Technology
The first step of solving your tooling stack is:
- What you’re currently paying for
- What people actually use daily
- What people have built or adopted on their own
- Where systems could connect but don’t
Step 7 — Build Tool Inventory
Layer 1 = What you pay for. Pull every active subscription and license. Include everything, even tools that haven’t been opened in months. Most organizations are surprised by how much they’re paying for and how much of it overlaps.
Layer 2 = What people actually use. Cross-reference Layer 1 against what surfaced in your interviews. Add anything that came up in the shadow layer questions. The gap between these two lists is usually significant and always instructive.
Layer 3 = What’s connecting to what. For each tool in daily use, note what it pulls from or pushes to. Mark every gap where two systems that share data aren’t connected — and where someone is manually bridging them.
Below is a simple, four-column spreadsheet to use as a framework:
| Tool | Who Uses It | What For | Connects To |
|---|---|---|---|
| Salesforce | Sales, Ops | CRM, pipeline | Nothing — data exported manually |
| Notion | All | Docs, wikis | Slack (loosely) |
| Excel | Ops, Finance | Reporting, reconciliation | Everything, manually |
You now have a list of intervention candidates. Score those qualitatively and quantitatively across five dimensions (rows below).
| Dimension | Candidate 1 | Candidate 2 |
|---|---|---|
| Time volume | Under 2 hrs/week total | Over 40 hrs/week total |
| Repetitiveness | Highly variable, judgment-heavy | Same steps every time |
| Error / rework rate | Near-zero errors | Frequent errors and rework |
| Ease of implementation | Major integration + change management | Off-the-shelf, live in days |
| Strategic importance | Administrative / support | Core to revenue or customer |
Start with two or three highest-impact items that are also easy to implement.
The Output
At the end of this phase you should have four things:
1. Time Map — where hours are actually going across your key function
2. Process Map(s) — as-is workflows for your most important processes
3. Tool Inventory — what you have, what you use, and where manual work lives
4. Opportunity List — your top AI opportunities scored by impact and effort
Roadblocks to Avoid:
- Starting with tools. Picking platform(s) before the workflow analysis is done turns discovery into a justification exercise.
- Starting without AI. Integrating AI into your transformation journey itself (e.g., screen recording) will begin to build comfort throughout the org.
- Mapping the whole org. Narrow scope produces a usable output. Broad scope produces a document that sits in a folder.
- No mandate. The Discovery Lead needs visible backing from someone with real authority. Without it, participation will inconsistent and findings will not deliver outcomes.
The Senseai Institute 