AI Prompts for Project Management: Business Cases, PRDs and RAID Logs

A business case compares the options for solving a problem, including doing nothing, on their costs, benefits and risks, and ends with a funding request. Net present value (NPV) converts future cash flows into today’s money at the required return, 12% at Corbel, and payback is the time needed to recover the investment.

• The cost of the problem: 48 of 118 cancellation surveys in the last year named offline working, from accounts paying $2.8M a year, and 19% of job completions fail at 170 commercial accounts.
• The recommended option: offline job completion costs $1.4M to build and $0.2M a year to run. From 2028 it protects $1.4M of revenue and saves $0.3M of support costs a year: an NPV of $1.6M over three years and payback in 1.4 years.
• Stage-gated funding: the budget is released in stages. $0.5M funds the sync engine and a field pilot; the remaining $0.9M is released only if failed completions at pilot accounts fall below 5%.

A business requirements document (BRD) lists what a system must do as numbered requirements, each with a priority, a source and an acceptance test, so IT or a vendor can build and test against it; a product requirements document (PRD) covers the same ground as user stories for a product team. Dispatch 3.0 had neither: it started with 14 user stories and finished with 31.

• MoSCoW priority: each requirement is marked Must, Should, Could or Won’t this release. Offline scheduling, invoicing and parts ordering are recorded as Won’t, so adding any of them later goes through change control.
• Business rules: the conditions the system must enforce. BR-04 decides what happens when a dispatcher edits a job while the technician is offline, the kind of data conflict behind 9 of Dispatch 3.0’s 41 pilot rollbacks.
• Traceability matrix: a table that links each business objective to its requirements, and each Must requirement to its acceptance test, so every Must requirement serves the business case and can be tested.

A RACI chart names who is Responsible, Accountable, Consulted and Informed for each decision, and on Dispatch 3.0 three executives were each Accountable for scope, so 17 additions went in without anyone pricing them.

• Tolerance sets the escalation point. The project manager can absorb up to 2 weeks and 10% of the $1.4M budget; beyond that the COO decides. Dispatch 3.0 drifted 11 weeks and $460K before anyone did.
• The 100% rule found missing work. A work breakdown structure (WBS) splits the project into deliverables and work packages, and its 100% rule requires every level to add up exactly to the one above. Applying it turned up app store review and a field pilot: 4 weeks that were in nobody’s estimate.
• One stakeholder has promised $1.1M of pipeline. On the power/interest grid, which places each stakeholder by influence over the project and how directly it affects them, the VP Sales is high on both: he has told 3 prospects, worth $1.1M, that full offline arrives in Q1. The project needs him on side by week 2.

Cost of delay is what the business loses for each week an item waits, and weighted shortest job first (WSJF) divides it by the size of the job, which put offline job completion first at 6.8 and a reporting rebuild eighteenth at 0.9.

• Most of the backlog was opinion. 174 requests had a requester but no usage data, revenue or support volume behind them. They are listed as unscored rather than given guessed numbers.
• Capacity draws the line. 14 engineers over a 12-week quarter covers the top 9 items. The other 29 scored items are marked “not this period”. A RICE cross-check, which scores reach, impact, confidence and effort, agreed with the WSJF order on all 9.
• An override has a written price. The largest customer, paying $900K a year, wants custom invoice templates. Moving them above the line pushes photo compression, rank 4, back 5 weeks, and the log names the COO as the person who approves that trade.

An outcome-based roadmap lists the results the company is trying to move rather than a calendar of features, and Corbel’s has three: fewer failed job completions, fewer cancellations that name offline, and faster invoicing at commercial HVAC accounts.

• Confidence decides what gets a date. A Now / Next / Later roadmap groups work by level of commitment instead of by month: Now is in delivery, Next is being shaped, Later is a problem worth solving. Only the Now column carries a quarter with a stated confidence level. Next carries a quarter only, and Later, where offline scheduling sits, carries no date at all.
• The board promise survives as scoped. A commitment check tests every promise already made against capacity. “Offline in Q1 2027” fits 14 engineers if it means job completion, the part supported by the 19% failure rate in the logs.
• The sales promise does not. Full offline for 3 prospects worth $1.1M needs offline scheduling and invoicing, about 2 more quarters of work. The roadmap includes the message the VP Sales sends them now, not in March.

The critical path is the longest chain of work in which each step waits for the one before, and Corbel’s runs through the sync engine, conflict handling, a field pilot and app store review, with no slack anywhere along it.

• Three-point estimates show the spread. Each work package gets an optimistic, likely and pessimistic figure. The sync engine ranges from 5 to 11 weeks, the widest on the plan, which makes it the package to watch first. The dependency map, which records what each package waits on, puts it at the start of the critical path.
• History sets the buffer. Dispatch 3.0 took 37 weeks against 26 planned, 1.4 times its estimate, so the committable date carries 4 weeks on top of the likely finish of 15 March.
• Two weeks of scope closes a 12-day gap. Batch photo upload, sending more than 10 photos at once, is 2 weeks of work on the critical path. Moving it to Next makes 29 March a date engineering will commit to.

A RAID log tracks Risks, Assumptions, Issues and Dependencies in one place, and a pre-mortem fills it by imagining the project has already failed; Corbel’s produced 23 entries, and the two highest scores were things the team knew but had never written down.

• One person is a single point of failure. Priya wrote the sync layer that every offline job completion depends on. If she left, the critical path would stop: probability 4 of 5, impact 5 of 5, a score of 20.
• The pilot collides with the busy season. Commercial HVAC contractors are at full stretch in January and February, exactly when the pilot is scheduled. A pilot account that cannot spare the time scores 16 and would push the date.
• Every response has a cost. Each high-scoring risk gets a response strategy: avoid, mitigate, transfer or accept. Mitigating this one by pairing a second engineer from week 2 costs 6 engineer-weeks, set against the 8 weeks the critical path would stand still while a replacement was hired and trained.

Change control prices every request before it is scheduled, and offline invoicing was already recorded in the requirements as Won’t this release, so it could only come in as a change with its cost attached.

• The cost falls on the critical path. Classified as new scope, rather than a clarification or a defect, the request is priced in full. Invoices need the same conflict handling as jobs, which adds 3 weeks to the longest chain of work and moves release from 29 March to 19 April.
• Three weeks breaches tolerance. Two changes approved earlier had already used 9 days. This one takes the total past the 2 weeks the project manager may absorb, so the decision goes to the COO.
• Deferring still keeps the deal. Northgate’s peak is the summer cooling season, so a Q2 release with a written date still reaches its technicians in time. Approving it now moves the date for all 640 customers.

Earned value compares the work finished with the work planned and the money spent, and at week 10 Corbel had earned $442K of value against $520K planned while spending $541K.

• Two ratios make it plain. The schedule performance index (SPI), value earned divided by value planned, is 0.85: 15% behind. The cost performance index (CPI), value earned divided by actual cost, is 0.82: each dollar buys 82 cents of work.
• The forecast breaches tolerance. At a CPI of 0.82, the $1.4M budget finishes at $1.71M, 22% over against a 10% tolerance. Under the report’s RAG rules, where red, amber and green each map to a tolerance, that is amber now, and red on the day the breach happens.
• The team’s green rating matched none of the rules. The sync engine, on the critical path, is 9 days late: the same kind of early slip Dispatch 3.0 reported as green before it finished 11 weeks late.

A post-implementation review measures benefits realisation, whether a release delivered the benefits in its business case and the metrics in its requirements, rather than whether it launched on time, and Corbel launched on 29 March at $1.52M, inside its 10% tolerance, but moved failed job completions from 19% to 4.1% instead of under 3%.

• No evidence counted as not met. 11 of 13 go/no-go criteria were met. The missing support runbook and 2 open severity-2 defects became named conditions with owners, not a reason to slip the date.
• Five whys found a process cause. Five whys means asking “why” repeatedly until the answer is a process the company can change. The failures cluster at 31 accounts running Android phones older than 2021, which run out of storage for queued photos. The pilot was drawn from the 5 largest accounts, whose technicians all carry recent iPhones, and no rule required it to match the customer base.
• The lesson changes a rule. Pilot selection must now match the customers’ mix of devices, owned by the head of product operations, and the next release adds a low-storage mode targeting under 3% by Q3 2027.