Lesson 1: Mapping Your Workflows — Finding Where AI Adds Real Value

Course: AI Implementation & Workflows Pathway: Business (Paid) Level: Intermediate Estimated Reading Time: 8 minutes

---

Introduction

Before you touch a single AI tool, you need to understand your work. Not in the abstract — the actual, specific sequences of tasks that keep your operation running. This lesson is about building that map, then finding the spots where AI can genuinely help.

Most failed AI implementations share the same root cause: someone bought a tool and went looking for a problem to solve. We're going the other direction. Start with the work. Then see where AI fits.

What Is a Workflow, Really?

A workflow is any repeatable sequence of steps that moves something from input to output. It might be:

• A customer enquiry arriving by email, getting triaged, assigned, responded to, and closed
• A monthly report being compiled from three data sources, reviewed, formatted, and sent to stakeholders
• A new employee being onboarded — accounts created, documents signed, training scheduled

Every organisation runs on dozens of these. Some are documented. Most aren't. They live in people's heads, in habits, in "that's just how we do it."

Your first job is to make the invisible visible.

Step 1: Identify Your Core Workflows

Start by listing every repeatable process in your area of responsibility. Don't filter yet — just capture. Think about:

• Daily routines: What happens every morning? What tasks repeat throughout the day?
• Weekly cycles: Reports, meetings, check-ins, reviews
• Triggered processes: What kicks off when a customer contacts you? When a new order comes in? When someone resigns?
• Monthly/quarterly: Reporting, planning, invoicing, compliance

Talk to your team. Ask them: "Walk me through what you actually do when X happens." You'll be surprised how many steps exist that nobody's written down.

A simple table works well for this:

Step 2: Break Each Workflow Into Individual Steps

Pick your top five workflows by frequency or business impact. For each one, document every step in sequence. Be specific. Not "process the order" but:

1. Open the order email
2. Copy the order details into the tracking spreadsheet
3. Check stock levels in the inventory system
4. Confirm pricing against the current price list
5. Generate the order confirmation
6. Send confirmation to the customer
7. Flag the order for the warehouse team

This level of detail matters. AI doesn't help with vague processes — it helps with specific tasks within processes.

Step 3: Categorise Each Step

For every step, ask yourself four questions:

1. Is this step primarily about information processing? Reading, summarising, extracting data, categorising, translating, reformatting. AI is strong here.

2. Is this step repetitive and rule-based? If the same logic applies every time with minor variations, that's a good candidate. If every instance requires novel judgment, less so.

3. How much does this step cost in time? A task that takes 2 minutes but happens 200 times a month is 400 minutes of potential savings. A task that takes 3 hours but happens once a quarter is 12 hours a year. Both matter, but differently.

4. What's the consequence of getting it wrong? This is your risk assessment. Sorting emails into folders? Low risk. Approving medical claims? High risk. The higher the consequence, the more you'll want human oversight even if AI does the heavy lifting.

Mark each step with a simple rating:

• 🟢 Strong candidate — Information-heavy, repetitive, lower risk
• 🟡 Possible candidate — Some AI potential, needs human oversight
• 🔴 Keep human — High judgment, high stakes, or relationship-dependent

Step 4: Find the Patterns

Once you've mapped and categorised several workflows, patterns emerge. Common AI-ready tasks tend to cluster into a few categories:

• Data entry and extraction: Pulling information from documents, emails, or forms into structured systems
• Drafting and summarisation: First drafts of reports, meeting summaries, response templates
• Classification and routing: Sorting incoming items by type, urgency, or department
• Research and synthesis: Gathering information from multiple sources and consolidating it
• Quality checks: Reviewing data for inconsistencies, formatting errors, or missing fields

You'll likely find that 20% of your steps account for 80% of the time savings potential. Focus there.

Agentic AI Workflows — The 2026 Reality

A significant development in 2026 is the rise of agentic automation — AI systems that don't just assist with individual tasks but autonomously orchestrate multi-step workflows. Blue Prism's analysis of 2026 trends highlights how agentic AI marks "the true democratisation of AI where every company can wield intelligence at scale." However, only organisations with proper governance foundations will transform this availability into genuine advantage.

Agentic workflows differ from traditional automation in key ways:

Traditional RPA: Pre-defined rules execute fixed sequences. If the input deviates, it breaks.

AI-assisted steps: Humans remain in control; AI augments specific tasks.

Agentic workflows: AI systems make decisions about which tools to use, when to escalate to humans, and how to handle exceptions — all within guardrails you define upfront.

Examples of agentic workflows:

• Customer support triage: An AI agent reads incoming tickets, researches the issue across knowledge bases, drafts responses, checks against policy rules, escalates complex cases to human agents with full context summaries, and closes routine enquiries autonomously.
• Procurement automation: An AI agent monitors inventory levels, identifies reorder needs, sources suppliers within budget constraints, negotiates terms via predefined parameters, creates purchase orders, and tracks delivery — escalating only when prices exceed thresholds or suppliers are unavailable.
• Compliance monitoring: An AI agent continuously scans regulatory updates relevant to your industry, cross-references them against your current policies, flags gaps, drafts policy revisions for legal review, and tracks implementation progress.

Governance as a competitive differentiator: The organisations that win with agentic AI aren't those with the most advanced models — they're those with clear governance frameworks defining what agents can do, how they should behave, and who oversees them. This connects directly to the AI governance course: your risk assessments, accountability structures, and transparency requirements become the foundation upon which agentic workflows are safely deployed.

Step 5: Prioritise Your Opportunities

Not every green-rated step should be tackled first. Prioritise based on:

• Impact: How much time or cost does this save?
• Feasibility: How straightforward is it to implement AI here? (Does the data exist? Is it accessible?)
• Risk: What's the downside if it goes wrong during the transition?
• Dependencies: Does this step need to change before others can?

A simple 2x2 matrix of Impact vs. Feasibility gives you a clear starting point. High impact, high feasibility? Start there.

A Practical Example

Consider a property management company handling maintenance requests:

1. Tenant emails about a leaking tap 🟢
2. Admin reads the email and categorises the issue 🟢
3. Admin checks the property file for relevant details 🟢
4. Admin contacts the preferred plumber 🟡
5. Plumber confirms availability 🔴
6. Admin schedules the job and notifies the tenant 🟡
7. After completion, admin logs the work and cost 🟢

Steps 1–3 and 7 are strong AI candidates. An AI system could read incoming maintenance emails, categorise them by trade and urgency, pull up property details, and draft the initial response — all before a human even looks at it. That might save 15 minutes per request across hundreds of requests a year.

Steps 4–6 involve coordination with external people. AI can assist (drafting messages, suggesting available times) but a human should oversee the communication, at least initially.

Common Mistakes to Avoid

Mapping too broadly. "Our sales process" is not a workflow — it's a department. Get granular.

Ignoring the messy bits. Real workflows have exceptions, workarounds, and edge cases. Map those too. They're often where AI struggles most, and knowing that upfront saves pain later.

Assuming AI replaces the whole workflow. It rarely does. AI typically handles specific steps within a workflow, not the entire thing. Think assistance, not replacement.

Skipping the people. The person doing the work knows it best. If you map workflows from a management perspective without talking to the people who actually do them, you'll miss critical details.

---

Key Takeaways

1. Start with the work, not the tool. Map your actual workflows before evaluating any AI solution.
2. Get granular. Break workflows into individual steps — AI helps with specific tasks, not vague processes.
3. Categorise by AI-readiness. Look for information-heavy, repetitive, lower-risk steps as your best starting candidates.
4. Prioritise by impact and feasibility. Not every opportunity is worth pursuing first. Focus where the return is clearest.
5. Talk to the people who do the work. They know the reality that process documents miss.

---

Practical Exercise

Map one real workflow from your organisation.

1. Choose a workflow you're personally familiar with — something that happens at least weekly.
2. List every step from trigger to completion (aim for 5–10 steps minimum).
3. Rate each step using the 🟢🟡🔴 system described above.
4. Identify your top 2–3 steps where AI could add the most value.
5. Estimate the time currently spent on those steps per week/month.

Write it up. You'll use this map in later lessons when we move into implementation.

---

Knowledge Check

Question 1: What is the recommended first step before selecting an AI tool for your business?

A) Research the most popular AI tools in your industry B) Map your existing workflows and identify specific steps where AI could help C) Hire an AI consultant to assess your organisation D) Start a pilot project with the cheapest available AI tool

Correct Answer: B — Always start by understanding your actual work processes before evaluating tools. The workflow map tells you what you need; only then can you find the right tool.

---

Question 2: Which type of task is generally the strongest candidate for AI assistance?

A) Tasks requiring novel creative judgment every time B) Tasks that depend heavily on in-person relationships C) Repetitive, information-heavy tasks with lower risk of error consequences D) Tasks that happen rarely but are extremely important

Correct Answer: C — AI excels at processing information consistently across repetitive tasks. Lower-risk tasks also make better starting points because the cost of errors during implementation is manageable.

---

Question 3: When mapping workflows, why is it important to document exceptions and workarounds?

A) They make the documentation look more thorough B) AI handles exceptions better than standard processes C) They reveal where AI is likely to struggle, helping you plan for human oversight D) Exceptions should be eliminated before implementing AI

Correct Answer: C — Edge cases and workarounds are often where AI performs least reliably. Knowing about them upfront lets you design appropriate human oversight into your AI-assisted workflow.