How it works
No AI. A rules engine that encodes how consulting deliverables are built.
DealDesk is structured around the four tasks of the PwC management consulting simulation. Each task has a known format and a known grading standard, so each one can be systematized. Everything below runs in your browser as plain JavaScript: nothing is uploaded, nothing is paid for, and every sentence in every document comes from a rule you can read in the source code.
The four tasks
What the engine does for each deliverable
Task 1: Business Summary
The simulation asks for a short, skimmable brief: overview, financials, people, products, news. DealDesk computes growth and margins from your numbers, writes the findings from threshold rules (a 9% growth rate is called healthy because the rule says over 8% is healthy), and follows the template section order exactly. For public companies the lookup fills the description, leadership, and reported revenue from Wikipedia and Wikidata.
Task 2: Interview Guide
The simulation asks for 15 to 20 numbered questions in ordered topics with space for answers. DealDesk keeps a bank of questions per practice and per deal type. Your situation answers add targeted questions: say the company has excess capacity and the guide gains a stranded cost question; say one customer dominates and it asks about retention first.
Task 3: Synergy Presentation
The simulation asks for a three slide loop: a summary slide that signposts, then two slides that expand each category in the same order. DealDesk keeps a library of opportunity hypotheses, scores each against your situation, and puts the top three per category on the slides, with the company's own revenue and margin charted from its filings. Font sizes are computed so text always fits its box, and a geometry check confirms nothing sits outside the slide.
Task 4: Excel Model and Email
The simulation asks for a simple model with labeled inputs, scenarios, and an email that opens with the answer. DealDesk builds four linked sheets where yellow cells are the only inputs: change the growth rate or the valuation multiple and the projections, scenarios, and sensitivity grid all recalculate. Nothing is hardcoded inside a formula. The email states a recommendation that follows the numbers: proceed, proceed with caution, or run a diagnostic first.
The honest part
What it can and cannot do
The engine selects and computes; it does not reason. Two similar situations get similar hypotheses, because the hypotheses come from a written library, not from reading your memo. That is a deliberate trade: the output is deterministic, free, private, and always in the correct format. The judgment stays with the consultant, which is where it belongs.
Reliability: every release runs an automated regression of several hundred generated engagements, including real public company financials, hostile inputs, and a geometry check that verifies no slide text overflows its box.