A complete operating manual for analytical spreadsheet modeling, synthesized from MGT 6574 (Spreadsheets for Business Analytics, Module III 2025, Dr. Larry J. LeBlanc, Owen Graduate School of Management, Vanderbilt University) — fourteen classes, six homework sets, two written technique papers, and the complete in-class solution library of roughly seventy workbooks. Approximately 30,000 words across a landing page, ten module sub-pages, and an eleven-page Master Playbook.

The course in one paragraph

MGT 6574 is Owen's power-user course — the class that takes a manager who can already type =SUM(A1:A10) and turns them into someone who can build, audit, and trust a model that a real company will make a seven-figure decision on. Its working claim is that the spreadsheet is the most widely used analytical tool on earth and the most widely trusted one that nobody verifies — and that the gap between those two facts is where careers and capital quietly leak away. Across fourteen classes the course builds two intertwined skill stacks. The first is optimization and simulation: modeling business decisions as linear programs solved with Excel's Solver, and modeling uncertainty with Monte Carlo simulation. The second is the craft of robust spreadsheet construction: formal Tables, PivotTables, structured references, the relative/absolute reference discipline, text functions, form controls, conditional formatting, INDIRECT, the advanced form of SUMPRODUCT, the modern dynamic-array functions (SORT, SORTBY, UNIQUE, FILTER, XLOOKUP), and data validation. Every technique is taught the same way: a business story, a live workbook, a wrong way that looks fine on small data and silently corrupts at scale, and the robust way that survives growth, edits, and the next analyst who inherits the file.

The three anchor phrases

If you remember only three sentences from this course, remember these. Every function, every model, every formatting convention is a footnote to one of them.

1. "According to PwC, ninety percent of corporate spreadsheets contain material errors." This is the risk thesis, and it is the reason the course exists. Before spreadsheet software, when a manager needed a report they went to the IT department, where disciplined programmers were not allowed to ship code until a second programmer had verified it. Now every analyst ships their own un-reviewed software to the executive committee every week. The entire syllabus is an answer to one question: how do you build a spreadsheet you can actually bet on? The answer is never "be more careful." It is structural — SUMPRODUCT instead of hand-typed sums, lookup tables instead of nested IFs, formal Tables instead of bordering ranges, data validation instead of trust.
2. "Every LP's optimal target-cell value is unique — but the optimal solution may not be." This is the optimization truth. Solver will always return the same best objective value for a correctly specified linear program; it may return any one of several different decision-variable combinations that achieve it. The managerial consequence is the whole point: once you know the optimum is, say, 77 employees, you can evaluate alternative solutions against it — accept a more operationally convenient schedule that costs 0.1% more, reject one that costs 30% more. The model does not make the decision; it gives you the yardstick to make it well.
3. "When Excel had 256 columns, simple approaches sufficed. Excel now has 16,384 columns and 1,048,576 rows — more sophisticated approaches are needed." This is the scalability thesis, and it is the bridge between the two halves of the course. A hand-typed =B16+C16 is fine for a four-instrument credit union and catastrophic for a 50-state, 200-vendor model. Every "advanced" technique in the course — the advanced SUMPRODUCT, dynamic arrays, INDIRECT, structured references, OFFSET/MATCH lookups — is the same idea re-derived for a world where the data is too big to eyeball and too live to freeze.

The instructor's standard of proof

Three habits run through every class and are worth naming up front, because they are what separates this course from a list of Excel tricks.

• Inspect the numerical solution to debug the model. The Burger King diet LP returns 1.2 bottles of salad dressing and 0.66 cartons of orange juice — and seeing those fractions is what tells you the model is missing its integer and logical constraints. P.J.'s cash-flow LP refuses to solve consistently because it mixes 110,000 with 0.01 — and the blow-up is what tells you to rescale every dollar to thousands. The optimum is not just an answer; it is a diagnostic.
• Prefer the formulation that scales and survives editing. Every model in the course is presented twice — an "error-prone" or "poor" version that works on the example, and a "preferred" or "good" version that works at scale. The difference is almost always: did you encode the logic as data (an indicator row, a lookup table) and reference it with one copyable formula, or did you bury bespoke arithmetic in each cell?
• Build it so the file catches its own mistakes. Advanced SUMPRODUCT errors out when text sneaks into a numeric range. Data validation refuses bad input at the keyboard. Conditional formatting lights up the outlier. Checker columns (=AND(G2=H2,H2=I2)) prove that four independent methods agree. Robustness is designed in, not inspected in.

The course arc — fourteen classes, ten modules

The ten modules group the fourteen classes into natural learning arcs. The foundation (Module 1) is the analytics culture, the risk thesis, and SUMPRODUCT — the single primitive that every later model is built on. The optimization arc (Modules 2–4) builds linear programming from a two-product mix to multi-period and network models. The craft arc (Modules 5–7) covers the everyday power-user toolkit — Tables and PivotTables, references and text, interactivity and signaling. The risk-reduction engine (Module 8) is the advanced SUMPRODUCT, the course's signature technique and the instructor's own research contribution. The modern toolkit (Module 9) is dynamic arrays. And the uncertainty arc (Module 10) closes with data validation, Monte Carlo simulation, and robust design.

The ten modules

• [[spreadsheets-business-analytics-module-1]] Foundations — Why 90% of Spreadsheets Are Wrong, and SUMPRODUCT as the Universal Primitive. The analytics culture and the spreadsheet-risk thesis; sorting applications and the cardinal hazard (select the entire block or you silently scramble every row); named ranges; the IF function; SUMPRODUCT taught as "the sum of products of corresponding cells," the scalable replacement for hand-typed term-by-term sums; and the grey-cell / red-border / double-border modeling convention used in every model that follows.
• [[spreadsheets-business-analytics-module-2]] Modeling Decisions as Linear Programs. The anatomy of an LP — decision (changing) cells, the objective (target) cell, and constraints in the three-cell left-side / relation / right-side layout. The Products A & B manufacturing mix as the canonical first model (maximize 550A + 500B − 100,000 subject to three machine-capacity constraints). Setting up Solver: target cell, Max/Min, changing cells, adding constraints, non-negativity, and choosing the Simplex LP engine. The Wine Bottler and Chair Manufacturer product-mix problems.
• [[spreadsheets-business-analytics-module-3]] The LP Portfolio — Investing, Lending, Diet, and Scheduling. Four LP archetypes. The Mountain Investment portfolio (maximize expected return subject to a weighted-risk ceiling; the optimum always lands in stocks 4 and 5). The Credit Union (the error-prone hand-summed constraints vs. the preferred indicator-row SUMPRODUCT). The Burger King diet model (binary and integer staging, and how the fractional solution reveals the missing constraints). The Warehouse / Nurse set-covering schedule (the 0/1 coverage matrix, 77 employees, and unique optimum value vs. multiple optimal solutions).