Insite-Ed/ Heuristic/Praxis

In this video, I take a real-world Excel problem — extracting email addresses from messy text — and show how to solve it using modern Excel functions and AI. We start by asking ChatGPT for a solution, and it delivers a powerful formula using functions like LET, TEXTSPLIT, and FILTER. But when we paste it into Excel… nothing happens. This leads to a key lesson: sometimes the problem isn’t the formula — it’s how Excel is handling it. A simple fix reveals that the formula was correct all along. From there, we explore two different approaches: • A transparent, step-by-step Excel method using dynamic array functions • A more compact and precise solution using REGEXEXTRACT This video is not just about extracting emails — it’s about understanding how Excel works, how AI fits into your workflow, and how to troubleshoot when things don’t go as expected. Key takeaways: • How to extract email addresses from messy text in Excel • Why formulas sometimes don’t activate when pasted • The difference between traditional Excel logic and regex • When to use each approach If you work with messy data, this is a practical technique you can apply immediately.

Exporting from Microsoft Project to Excel Microsoft Project organises project data across several built-in views and tables, three of the most commonly used being the Entry table, the Cost table, and the Resource table. Each of these surfaces a distinct slice of your project data: the Entry table captures task-level scheduling information such as task names, durations, start and finish dates, and completion percentages; the Cost table presents the financial dimension of each task, including budgeted, actual, and remaining costs; and the Resource table holds information about the people and materials assigned to the project, including resource names, types, rates, and availability. To move this data into Excel, each table is copied directly from Microsoft Project and pasted into a separate worksheet within an Excel workbook. It is important to note that this is a static, point-in-time transfer — there is no live or dynamic link between Microsoft Project and Excel. Once the data has been pasted, it is a snapshot that reflects the state of the project at that moment, and any subsequent changes made in Microsoft Project will need to be manually copied across again. Each worksheet should be clearly named to reflect its source table, so that the data remains organised and traceable back to the original Project file. Understanding the Data Relationships Before importing into Power BI, it is important to understand how these three tables relate to one another — or in one case, do not. The Entry table and the Cost table share a natural relationship: both are task-centric, meaning each row in the Cost table corresponds directly to a task that also appears in the Entry table. This relationship is typically established through a common unique identifier, such as the task ID or task name, and it allows Power BI to link task scheduling data with task cost data seamlessly. The Resource table, however, operates independently. While resources are assigned to tasks within Microsoft Project itself, the exported Resource table contains resource-level summary information rather than a direct task-by-task breakdown, and as such there is no direct relationship between the Resource table and either the Entry or Cost tables in this export structure. The Resource table therefore stands alone in the data model as a reference table, useful for resource-specific analysis but not connected to the task-level tables through a formal relationship. Importing into Power BI via Power Query Once the Excel workbook is prepared, the data is brought into Power BI Desktop through Power Query, which serves as the transformation and loading layer between the Excel source and the Power BI data model. Using the Get Data function and selecting Excel as the source, Power Query opens each of the three worksheets — Entry, Cost, and Resource — allowing you to inspect, clean, and shape the data before it is loaded. This step is valuable precisely because the data arrives as a static paste from Microsoft Project, and Power Query provides the opportunity to remove any unnecessary rows or columns, correct data types, rename fields for clarity, and ensure consistency across tables before they enter the model. Once the transformations are applied and the data is loaded, Power BI will attempt to auto-detect relationships, but it is advisable to review and confirm these manually in the Model view. The relationship between the Entry table and the Cost table should be defined on the shared task identifier, typically as a one-to-one relationship since each task appears once in each table. The Resource table, having no relational link to the other two tables, remains as a standalone table in the model — perfectly valid in Power BI, where it can still drive its own visuals and slicers independently, without filtering or cross-referencing the task-level data from the Entry or Cost tables.

One of the most useful developments in AI tools is memory. But memory is only useful if it is managed properly. Just as we do not want a human assistant remembering every random sentence we ever said, we do not want an AI remembering every passing detail either. The trick is to curate the memory — to decide what should be kept, what should be ignored, and what should occasionally be removed. Here are four practical principles. 1. Save only durable facts and preferences The first principle is this: save only durable facts and preferences. By durable, I mean information that will still be useful next week, next month, or even next year. For example, if you regularly ask AI to create training documents, it is useful for the AI to remember your preferred formatting style. Perhaps you like blue and white banded tables, a particular font, a certain tone, or a specific way of structuring course outlines. It may also be useful for the AI to remember your business context, your teaching style, recurring projects, important clients, or the kind of documents you often produce. But not everything belongs in memory. If you mention that you are having coffee, or that you are annoyed with the traffic, or that you are briefly working on a small one-off task, that probably does not need to be remembered forever. So the question to ask is simple: Will this information still help the AI serve me in future? If the answer is yes, save it. If the answer is no, leave it in the conversation, but do not make it part of long-term memory. 2. Use clear memory instructions The second principle is: use clear memory instructions. AI tools are improving, but they are not mind-readers. Sometimes they may infer that something matters, but the safest approach is to be explicit. Use simple phrases like: “Remember this.” “Save this for future use.” “Use this as my preferred style going forward.” “Forget this.” “Don’t treat this as a permanent preference.” These instructions help the AI distinguish between a passing comment and a lasting instruction. For example, there is a big difference between saying, “This table looks good,” and saying, “Remember this as my preferred table style for future training documents.” The first statement is feedback for the current task. The second is a memory instruction. Likewise, if something changes, tell the AI. You might say, “Forget the old branding style,” or “From now on, use the new course name.” Good memory depends on good signals. The clearer your instruction, the better the AI can separate temporary conversation from long-term preference. 3. Anchor important work in files The third principle is: anchor important work in files. AI memory is helpful, but for serious projects, memory alone is not enough. If you are working on a course, a report, a business proposal, a task list, or a project plan, keep a proper file. That might be a Word document, an Excel workbook, a PowerPoint deck, a project brief, or a saved PDF. Why? Because files create a stable reference point. Memory can remember that you prefer a certain style, or that you are working on a particular project. But the actual content — the chapters, tables, dates, tasks, sections, decisions, and revisions — should live in a document. Think of AI memory as the assistant’s general familiarity with you. Think of files as the official record. For example, if you are developing a training course, the AI might remember your general teaching approach. But the current version of the course outline should still be kept in a Word document. Then, when you come back later, you can upload the latest file and say, “Continue from this version.” This avoids confusion, duplication, and accidental drift. 4. Review and prune memory periodically The fourth principle is: review and prune memory periodically. Memory is not only about adding things. It is also about removing things. Over time, old details can become misleading. A project may finish. A client may no longer be relevant. A formatting preference may change. A temporary workflow may no longer apply. If the AI remembers too much outdated information, it may start applying old assumptions to new situations. So every now and then, review what the AI remembers. Ask: Is this still true? Is this still useful? Should this still influence future answers? If not, remove it. You can say things like: “Forget that project.” “That style is no longer my preference.” “Replace the old instruction with this new one.” Good memory is not a junk drawer. It is more like a well-kept filing cabinet. The value comes not from storing everything, but from storing the right things.

Here’s the kind of prompt you should use: PROMPT: I need you to write a formal workplace environmental and occupational health report examining fumes and dust generated by furnace operations at [Company Name], a metals processing facility located at [Location]. Context and operations: [Describe the furnaces — type, number, what metals are processed, typical operating temperatures, production volumes, and shift patterns. Include any existing extraction, ventilation, or PPE controls already in place.] Monitoring data: [Paste or describe the air quality and particulate measurement results — substance names, recorded levels, units, measurement dates, and locations. Include both personal exposure readings and area monitoring if available.] Regulatory benchmarks: [Specify the applicable occupational exposure limits and environmental standards — e.g. OEL/TWA values for manganese fume, respirable silica, iron oxide, etc. — and the jurisdiction or standard they come from.] Workforce and health observations: [Note the number of workers exposed, roles, duration of exposure, and any occupational health findings, complaints, or incidents related to fumes or dust.] Report audience: [State who will receive this report — e.g. health and safety committee, regulatory body, senior management, external auditor.] Using all of the above, write a professionally structured report with the following sections: an executive summary; operational context; measurement methodology; findings and data; comparative analysis against applicable standards; risk assessment for employees and the external environment; and recommendations. Write in formal but clear prose — no bullet points in analytical sections. Where the data has gaps or limitations, state them plainly rather than glossing over them. The recommendations should distinguish between immediate actions and longer-term measures. The more specific the data you drop into those placeholders, the sharper the report. If you leave sections thin — say, the monitoring data is limited — Claude will flag that in the findings rather than inventing numbers, which is exactly what you want in a document that may face scrutiny. LET’S EXPLAIN The task here is specific: produce a written report that examines the levels of fumes and dust generated by furnace operations at a metals company, assesses the impact on employees and the surrounding environment, and presents findings in a form that is credible, structured, and professionally formatted. Microsoft Word is the natural delivery format — it handles headings, section breaks, tables, and pagination cleanly, and it produces a document that can be signed off, printed, and filed. Claude is the engine that drives the content. The workflow between them is direct. You begin by giving Claude the raw material. That means uploading or pasting in whatever source data you have — monitoring records, air quality readings, occupational health observations, equipment specs, regulatory thresholds, any incident logs — and telling Claude what the report needs to accomplish and who it is for. A report destined for a health and safety committee reads differently from one prepared for a regulatory submission or an internal management review. Claude adjusts register, depth, and emphasis accordingly, so being explicit about the audience at the outset saves considerable revision downstream. Once the parameters are established, Claude drafts the report structure and begins populating it section by section. A report of this kind would typically move through an executive summary, a description of the operational context (the furnace types, the metals being processed, the production volumes), a methodology section explaining how fumes and particulate levels were measured and over what period, a findings section presenting the data with commentary, a comparative analysis against applicable occupational exposure limits and environmental standards, an assessment of risk to both employees and the external environment, and a recommendations section proposing engineering controls, procedural changes, or monitoring improvements. Claude handles all of this — not as boilerplate, but as substantive prose shaped by the data you have provided. The interaction is iterative rather than one-shot. You review what Claude produces, flag what needs sharpening or redirecting — a section that overreaches the data, a tone that is too clinical for the audience, a recommendations paragraph that needs to be split into immediate actions versus longer-term measures — and Claude revises. This back-and-forth is where the value compounds. Claude does not forget earlier context within a session, so it adjusts consistently rather than starting from scratch with each prompt.

In this video I show you why the LET function in Excel is a genuine upgrade — not just syntactic sugar — using a construction site materials dataset with XLOOKUP embedded inside an IF formula. The problem with IF + XLOOKUP on its own When you write a conditional formula that depends on a lookup result, you end up writing the XLOOKUP twice: =IF(XLOOKUP([@[Element Code]],tblElements[Item],tblElements[Unit Weight (kg)])＜500, XLOOKUP([@[Element Code]],tblElements[Item],tblElements[Unit Weight (kg)]), "CRANE LIFT") Excel has to evaluate that XLOOKUP twice — once to test the condition, once to return the value. The formula is long, hard to read, and if your lookup range ever changes you have to update it in two places. In a more complex model with nested lookups, this compounds fast. What LET does LET lets you assign the result of the XLOOKUP to a named variable — unit_wt — and then reference that name wherever you need it: =LET(unit_wt, XLOOKUP([@[Element Code]],tblElements[Item],tblElements[Unit Weight (kg)]), IF(unit_wt＜500, unit_wt, "CRANE LIFT")) One XLOOKUP. One place to update. The formula reads like a sentence: define unit_wt, then ask whether it's under 500. This is particularly powerful when the lookup itself is complex — Power Query outputs, multi-condition XLOOKUPs, nested IFs — because LET lets you break the logic into named steps that are easy to audit and debug. What's in the dataset I use a two-table construction model: a site materials requisition (tblSiteWorks) and a structural elements register (tblElements) containing 12 element types from lightweight hollow sections (28 kg) to precast staircase flights (1,200 kg). The Lift Class column classifies each element as either its actual unit weight — for manual or forklift handling — or "CRANE LIFT" if it exceeds 500 kg. The same column is built twice side by side: once the traditional way, once with LET, so you can see the contrast directly in the sheet. Who this is for If you use Excel for project management, engineering, or construction reporting and you're still writing the same lookup reference multiple times inside a single formula — this is the video that fixes that habit.