PowerSolver pilot guide — the consultant's field manual

Who this is for: any consultant, including a junior one, who meets a customer to understand their planning problem, collects their data, prepares the PowerSolver workbook, validates it and runs the pilot. How to use it: read §1-§3 once. In the customer meeting, follow §4 (the interview script) with §6 open for the customer's problem type. Before any solve, run §8 (validation) and §11 (acceptance gate). After the pilot: hand the tool over with HAND_OVER_CUSTOMER.md.

Everything below describes what PowerSolver does today (wizard, workbook, engine as released in October 2026). What it does not do is written as such (§2.4, §13). Never promise anything that is not in this guide.

Contents

1. PowerSolver in one page
2. What to say about limits
3. The pilot, step by step
4. The customer interview — script
5. Decision trees
6. Problem types — what to ask and collect for each
7. Data discovery and the Excel inventory
8. Filling the workbook
9. Validating the data before planning
10. Exceptions and edge cases
11. Troubleshooting
12. Acceptance gates
13. Known gaps and escalation

1. PowerSolver in one page

A plan is described by one Excel workbook (or the equivalent problem.json file). The consultant's job is to fill it correctly. Six notions are enough:

Pin or reference list? A pin fixes the decisions of some items; a reference list holds something that is never decided for any item. Details and examples: CONCEPTS_GUIDE.md §5b. A pin fixes all decisions of the item at once (machine and slot).

The workbook (built for every demo in powersolver-demos/*/input/*.xlsx; the wizard's step 1 Download workbook template gives the current plan or a worked example):

The wizard (https://planning.planningpowertools.com, sign-in required): step 1 Load a plan from Excel, then Fill all steps now (or step by step), step 7 Can this work? (checks the data before planning), Plan it, then the Results page (Check before you use it, Rules & score, all items, Changes, Export to Excel).

The 19 rules (exact settings: RULE_REFERENCE.md):

2. What to say about limits

Say these at the first meeting. They are facts of the product, not excuses.

2.1 What PowerSolver does

• Decides one to four things per item (who, when, where, with whom) or the order of items on a resource (routes).
• Checks the data before planning (Can this work?) and tells, in plain words, when a valid plan is impossible and why (not enough rooms, a chain of operations that cannot fit before its due date, a teacher given more hours than their service, …).
• Replans: keeps what is done (pins), moves as little as possible (wishes on the previous values), shows exactly what changed (Results → Changes; Compare Solutions for several successive plans).

2.2 Operating limits

2.3 Honest statements

• PowerSolver finds a very good plan, not a proven optimum. Promise: valid, measurably as good or better, much faster to make and to change.
• Two runs can give two different, equally valid plans (the search stops on time).
• "No valid plan found in a quick plan of 20 s" never means impossible: only a red Can this work? verdict does.

2.4 Needs PowerSolver does not cover today — check every customer rule against this list

Any rule that does not fit is written down in the pilot report as a limitation, never silently dropped.

3. The pilot, step by step

Success criteria — agree 3 or 4 in writing at step 2: (1) every must-follow rule holds on the real period (0hard); (2) on their top-ranked wishes, PowerSolver's plan is at least as good as theirs; (3) full plan in under N minutes, replan in under M; (4) the planner would publish it with fewer than K manual corrections.

Pilot report: the problem in their words and volumes; each rule as modelled (sentence + rule kind) and the rules that could not be modelled; results per success criterion; their plan vs PowerSolver's, rule by rule; limitations met; proposal.

4. The customer interview — script

Run it with the person who makes the plan today, and the decision maker for parts A and J. Write the answers in the interview sheet (§4.11). Each question says why it matters and what it becomes in the workbook.

4.1 Part A — the problem (15 min)

4.2 Part B — the items to plan (20 min)

4.3 Part C — the decisions (10 min)

4.4 Part D — the options lists (20 min)

For each decision: "List every possible option." Then per option:

4.5 Part E — reference lists (10 min, only if C3 or B10 answered yes)

"For each fixed person or thing: when are they away? What can they do? How much can they take?" These columns are read by Respect unavailability, Capability must match and Stay within capacity through resource_property. If the data itself breaks one of them (a teacher given more hours than their service), Can this work? says so before planning.

4.6 Part F — the rules (40 min)

1. Ask: "List every rule a plan must follow — things you would never publish." Write them in their words.
2. Ask: "List what makes a plan better — your wishes."
3. For every rule, read it back as a PowerSolver sentence and the rule kind (§1 table): "No two operations on the same machine at the same time — correct?"
4. For every must, ask the publication test: "Would you publish a plan that breaks it once?" If yes, it is a wish.
5. For every rule, ask which column holds the information. A rule without data cannot be checked: add the column to the data request.
6. For rules that do not fit any kind: check §2.4; write it down as a limitation with the customer, never drop it.

4.7 Part G — ranking the wishes (10 min)

"If two wishes conflict, which wins?" Order them; then give weights that keep the order with margin (e.g. 1, 3, 10). Let the planner decide the order, not the consultant. In replanning, "keep the previous value" wishes usually get 2-5 (demo 18: keep machine 3, keep time 2).

4.8 Part H — exceptions (15 min)

Use §10 as a checklist: holidays, partial availability, subcontracting, overtime, emergency items, items that may be left out, one-off events.

4.9 Part I — volumes and periods (5 min)

Number of items, options per decision, reference rows; planning horizon; how often they replan. Check against §2.2.

4.10 Part J — the baseline and success criteria (10 min, with the decision maker)

"Give us the plan you actually made for this period." It becomes the comparison (§11.4). Then write the 3-4 success criteria (§3).

4.11 The interview sheet (copy it into the meeting notes)

5. Decision trees

5.1 Which problem type?

Are items placed in an ORDER on each resource (routes, tours, a machine's job sequence with travel/set-up)?
├── yes → vehicle-routing (decision of kind "sequence")
└── no → What is decided?
    ├── a time AND a room/place for each item, groups that must not clash → school-timetabling or conference-scheduling
    │      (a school: classes, teachers, subjects;  a conference: sessions, speakers, tracks)
    ├── a person for each shift/position (the shifts are given) → employee-rostering
    ├── a machine AND a time for each operation, routing order → job-shop-scheduling
    ├── a technician AND a time for each job at a site → maintenance-scheduling
    ├── a server/host for each workload, capacities → cloud-optimization
    ├── a person (and possibly a time) for each task, skills → task-assignment
    └── none of these fits → generic (same engine, no type-specific form fields)

The problem type changes only the wizard's form fields and labels; the rules and the engine are the same for all.

5.2 Resource, time or sequence?

Is the decision "when"?                       → Kind = time      (the options list has startTime, endTime)
Is it "who / what / where" (no order)?        → Kind = resource
Is it "in what order on each vehicle/machine"? → Kind = sequence (one per plan at most; options = vehicles/machines)

5.3 Decision, pin or reference list?

For a column such as "technician" or "teacher":
Is it chosen by the planner for at least some items?
├── no, never chosen, for any item          → reference list (+ a column on the item naming it)
└── yes
    ├── for every item                       → decision
    └── for most items, some already fixed   → decision + pin those items (pins fix ALL their decisions)

5.4 Must or wish?

Would the customer publish a plan that breaks it once?
├── never        → Must (hard). Then check: does the data allow it? (§9.3)
└── yes, if needed → Wish (soft), with a weight from the ranking (§4.7)

5.5 Can a rule be modelled?

Find the rule kind in §1 whose sentence matches → is the needed column in the data?
├── yes → write the Rules row (§8.4)
└── no column → add it to the data request
No kind matches → §2.4 → workaround agreed with the customer, or written down as a limitation

6. Problem types — what to ask and collect for each

The wizard offers 9 problem types. For each: the columns its form template asks for (mandatory ones in bold: an import without them is refused with "Required column … is missing"), a model verified on a demo, the specific questions, the usual rules, and the traps. Any extra column is accepted and kept ("new" column) — rules can use it.

The template columns come from PowerAdmin (step 2 = items, step 3 = options). duration is labelled hours in the form; the engine reads ISO durations (PT2H, PT55M) natively and the import accepts them. Use ISO durations to avoid any ambiguity (§13).

6.1 Job shop — job-shop-scheduling (demos 09, 18)

6.2 School timetabling — school-timetabling (demos 01, 12)

6.3 Conference scheduling — conference-scheduling (demo 14)

6.4 Employee rostering — employee-rostering (demos 05, 15)

6.5 Maintenance scheduling — maintenance-scheduling (demos 07, 13)

6.6 Cloud optimisation — cloud-optimization (demos 08, 17)

6.7 Task assignment — task-assignment (demos 04, 11, 16)

6.8 Vehicle routing — vehicle-routing (demo 06)

6.9 Generic — generic

Use it when no type fits. Template columns: items id, name, duration, priority, skills · options id, name, skills, capacity, hourlyRate, availability, location. Everything else is identical. Demos 19 (exams: 3 decisions) and 20 (operating theatre: 4 decisions) use their own type names and show models with 3 and 4 decisions.

7. Data discovery and the Excel inventory

7.1 The inventory table (fill one row per data element before building anything)

Mark each element mandatory (a rule or the template needs it), optional (improves the plan) or not needed.

7.2 Questions about existing files

1. "Where does this information live today? Excel, ERP, HR tool, paper?" — ask to see the file.
2. "Who updates it, and how often?" — the owner becomes the handover contact for that sheet.
3. "Is there one row per item, or one row per something else?" — e.g. one row per order with "3 operations" in a cell must become 3 rows.
4. "Are ids stable? Do two files use the same ids?" — joins between files fail on ids written differently (T01 vs T1).
5. "Are there hidden rows, merged cells, colour codes that carry meaning?" — colours and merged cells are lost on import; turn them into columns.
6. "Is the file complete for the period, or is part of it in someone's head?" — write the missing part down now.

7.3 Reusing customer files

• A customer sheet can be imported as it is in steps 2-3 when its first row holds the column names and each row is one item. Column names can be the template's field names or labels (case and spaces ignored); unknown columns are kept as extra properties.
• Prefer copying the columns into the PowerSolver workbook (one sheet per list, named like the list), so that one file carries the whole plan and Fill all steps now works.
• Keep the customer's original file untouched; record in the inventory which columns were copied and transformed.

8. Filling the workbook

Start from Download workbook template (wizard step 1) — it gives a worked example (3 jobs, 2 technicians, 2 slots, 5 rules) when no plan is loaded — or from the demo workbook of the customer's type (§6), and replace the data.

8.1 Setup sheet, row by row

Columns: Part, Name, Label, Decision, Kind, From, Formula. First row = these headers.

Bad: Items to plan | Operations list (space: refused); two Items to plan rows; a Decision without Kind (read as resource — wrong for a time decision: the calendar views need time).

8.2 List sheets (steps 2 and 3)

• Name the sheet exactly like the list in Setup (operations, not Operations 2026).
• First row = column names; first column = id, unique in the sheet, never empty.
• One row per item / option. No merged cells, no subtotal rows, no blank rows in the middle.

Value formats (the import converts exactly as the engine reads):

Empty cells: an empty cell means "no value" — an empty needs = no requirement; an empty unavailable = always available. A column with a name but no value in any row is kept, empty, on every row (a rule may read it).

8.3 Pins and replanning columns

For an item already decided: pinned = TRUE and one column per decision id with the decided option id (e.g. machine = F2, slot = Mon-3). All decisions of a pinned item must be filled. To keep changes small in a replan, add previous_machine, previous_slot and wishes Honour preferences (entity_property previous_machine, value_property code) — as demo 18 problem.wednesday.

8.4 Rules sheet (step 4)

Columns: Rule, Must / Wish, Weight, Kind, then one column per setting. Write it in the wizard (step 4) and use Download rules (Excel), or copy the Rules sheet of the matching demo. Example (job shop):

Each setting names a decision id or a column of the step 2-3 sheets. The meaning of every setting is in the workbook's How to fill sheet and RULE_REFERENCE.md.

8.5 Loading

Wizard step 1 → Load a plan from Excel → the dialog lists what each step will import → Fill all steps now (all or nothing; every problem listed by sheet, row and column) → step 7. Or Go through the steps one by one and press Import on each step's banner (useful in front of the customer).

9. Validating the data before planning

Run 9.1-9.3 in Excel, 9.4-9.5 in the wizard. Do not solve before gate 1 (§12.1).

9.1 Structure checks (Excel)

9.2 Value checks

9.3 Arithmetic — is it even possible? (do it before Can this work?, to discuss with the customer)

If demand exceeds capacity, say it on day one: the right plan leaves items unplanned, and the customer must decide (add capacity, relax a rule to a wish, accept unplanned items).

9.4 Wizard import checks

Each import shows the problems by row and column; Import stays disabled until the file is corrected. Typical: "Required column "duration" is missing" (§6 template), "PT2H" is not a number (only for templates that predate ISO support — reload the wizard), "ID is empty", "ID … is already in the list", "Column 7 has values but no name".

9.5 Can this work? (step 7)

Findings and what to do:

10. Exceptions and edge cases

Ask each question; record the answer and the model decision.

Questions to raise before proceeding (do not build until answered): who decides when two wishes conflict; what to do with unplanned items; whether the plan may change once published, and how much; which data owner corrects errors.

11. Troubleshooting

11.1 Recognising unclear requirements

11.2 The plan is not valid (red on Results)

1. Read Check before you use it: which must rule is broken, how many times, for which items.
2. Run Can this work? again (step 7): red findings name the cause in the data.
3. If it is green and the plan is still not valid: give more time (step 6), or use Try a quick plan to see whether a valid plan exists at all.
4. Still not valid: turn one must rule at a time into a wish (step 4) to find the one that cannot hold, then find the data rows behind it; turn the rule back to must afterwards.

11.3 Data errors on load (The problem cannot be read)

Each message has a plain title, the engine's exact sentence and where to correct it ("Correct it in step 4: Rules › rule 'No clash' › setting 'group_property'"). Typical: a rule names a column no item has (spelling), a value has the wrong kind (text in a number column), an id is used twice, a decision has no options.

11.4 Comparing with the customer's own plan

Put their plan into the workbook as pins (every item pinned with their values), plan it: the score is their plan measured by our rules. Then plan without pins. Compare in Results → Rules & score, and with Compare Solutions.

11.5 Replanning after a change

Copy the published plan's decisions into the items sheet (Results → Export to Excel gives every item with its decisions): pin what is done, add previous_* columns and the "keep" wishes for the rest, apply the change (absence = unavailability, breakdown = down slots, rush order = new row), load, plan. Results → Changes compares with the previous plan (same browser tab, or load the baseline problem and solution files). The wizard has no one-click "replan from this plan" yet (§13).

11.6 Performance

Iterate with 30-60 s; use the long time limit only for the final plan. Above ~1,000 items, test the time first.

12. Acceptance gates

12.1 Gate 1 — data ready for planning (all must be ticked)

• ☐ Interview sheet complete (§4.11), signed by the customer planner.
• ☐ Every must rule mapped to a rule kind with its columns; every unmodelled rule listed as a limitation.
• ☐ Inventory table complete; every mandatory element has a source and an owner.
• ☐ Workbook: Setup complete; every sheet named like its list; ids unique and non-empty.
• ☐ §9.1 and §9.2 passed; §9.3 arithmetic discussed with the customer.
• ☐ Fill all steps now imports without any problem.
• ☐ Can this work? is green or amber, or every red finding is understood and accepted by the customer.
• ☐ The customer's own plan for the period is in hand.

12.2 Gate 2 — plan ready for review

• ☐ Results: "Ready to use" or every red point explained.
• ☐ Every must rule at 0 (Rules & score).
• ☐ Wishes compared with the customer's plan.
• ☐ Problem, solution and workbook saved with the date in their names (§ handover guide §7).

13. Known gaps and escalation

13.1 Gaps found while writing this guide (check before promising)

13.2 Escalation

Escalate to the PowerSolver product team when: the engine answers "internal error"; a red finding looks wrong for the data; a solve never finishes; a customer rule is essential and not in §1/§2.4. Send: the workbook, the problem_*.json and solution_*.json files the wizard downloaded, a screenshot of the message, and the step you were on. Never send customer personal data outside the agreed channel.