Industry · Technology
Runbooks and release routines that new engineers can follow and leaders can trust.
Operational know-how lives in a few engineers and a chat history, and customer commitments depend on it.
Example: Brightwater Software Services (fictional). Fictional organisation, fictional data, scripted answers.
What makes this sector hard
- 01Operational know-how lives in a few engineers and a chat history, and customer commitments depend on it.
- 02An incident review teaches the same lesson twice because nobody turns it into a runbook.
- 03Cloud bills grow faster than usage, and nobody owns the idle resources.
- 04A customer asks about their own data, and the company must say exactly what its assistant may read.
See it work on Brightwater Software Services’s own data
Simulation. Five short examples, one for each working area. The wording is scripted and no model is called; the figures are computed from what you decide, and a person’s approval covers exactly what you decided.
Draft a runbook from an incident review, using only what the review shows
- What happened
- After a three-hour outage of the booking service, the lead wants a runbook that the next on-call engineer can follow.
- The question
- Which steps can go into the runbook because the incident record supports them? Asked by: Site reliability lead.
Draft a runbook from an incident review, using only what the review showsChoose the steps that go into the draft runbook. A step the incident record does not support is refused; an estimate or a calculation is labelled as one.Open
Choose the steps that go into the draft runbook. A step the incident record does not support is refused; an estimate or a calculation is labelled as one.
Choose which statements go into the brief. A statement with no source is refused; each one you include carries its sources with it.
Booking service outage: runbook steps
Choose the statements to include. Each one carries its sources into the brief.
0 of 6 statements included · 1 held back for lack of a source.
0 steps are ready with evidence; 1 was left out because the review does not show it.
A runbook should contain what the record shows happened and worked [S1]. The belief that a rollback always fixes the problem is not in the record, and the outage began with a configuration change that rolling back the release would not have undone.
0 step(s) are labelled as calculated or estimated: the failover time comes from the drill [S2], and the escalation rule comes from the support policy [S3].
Relies on: S1 · Incident timeline and chat export (fictional)S2 · Failover drill record (fictional)S3 · Customer support escalation policy (fictional)
The site reliability lead with the on-call engineers decides. Approving releases this text as a draft runbook for review. It is not the approved version until a second person approves it, and it runs nothing.
Not ready. Choose at least one supported statement for the brief first.
Not done here. It does not run any step or change any system, and it does not add a step the incident record does not show.
Choose what the support assistant may read for one customer
- What happened
- A customer wants to use the company's support assistant on its own tickets, and security must set what the assistant may read.
- The question
- Which data sources may the assistant read for this customer, and which must stay out of its reach? Asked by: Head of security.
Choose what the support assistant may read for one customerChoose the sources the assistant may read. Restricted material cannot be chosen; a classification the analysis is unsure of must be confirmed by a person first.Open
Choose the sources the assistant may read. Restricted material cannot be chosen; a classification the analysis is unsure of must be confirmed by a person first.
Choose what the platform may read. Restricted material cannot be put in scope here, and a classification the analysis is unsure of must be verified by a person.
Read it? Source Holds Sensitivity What the analysis thinks it is Public help articles
Help centre640 documents open Articles clear This customer's support tickets
Helpdesk4,200 documents internal Tickets check Engineering design documents
Wiki1,800 documents internal Design notes clear Other customers' support tickets
Helpdesk61,000 documents restricted Tickets clear Source code repositories
Code host380 documents restricted Code clear Billing and payment records
Finance system9,300 documents restricted Invoices clear Release notes
Public website150 documents open Notes clear - Sources in scope
- 0 of 7
- documents in scope
- 0
- Restricted, kept out
- 3 sources
- To verify
- 0
0 of 7 sources are open to the assistant for this customer, 0 documents; 3 restricted source(s) stay out.
The assistant may read help articles, release notes, design documents and this customer's own tickets, and never other customers' tickets, source code or billing records [S1]. 0 included source(s) still need a person to confirm what they are before the scope is ready.
Relies on: S1 · Data source inventory for the support assistant (fictional)
The head of security decides. Approving sets what the assistant may read for this customer's pilot. It does not switch the assistant on, and it can be narrowed or revoked at any time.
Not ready. Choose at least one source to read first.
Not done here. It does not read other customers' data, source code or billing records, and it never answers from a source a person has not classified.
Find the cloud spend nobody is using
- What happened
- The monthly cloud bill rose 22 percent while customer usage rose 4 percent, and the finance lead wants to know where the difference went.
- The question
- Which cloud resources look idle or duplicated, and what does keeping them cost each month? Asked by: Cloud finance lead.
Find the cloud spend nobody is usingEach flagged resource has a reason. Mark it to reclaim, or to keep; the monthly cost is recomputed from your decisions.Open
Each flagged resource has a reason. Mark it to reclaim, or to keep; the monthly cost is recomputed from your decisions.
Review each flagged record: confirm it or dismiss it. The totals and the answer below follow your decisions.
Resource Owner Seen in use Monthly cost Finding Decision Production database cluster Platform team Used every day 18,400 Matched None needed Rehearsal cluster, a copy of production Platform team No traffic for 41 days 6,200 No traffic for 41 days Search service, old version Search team Replaced in March; still running 3,100 Replaced in March but still running Nightly batch workers Data team Run every night 4,700 Matched None needed Test environment, Project Falcon No owner recorded Last used in July 2,900 No owner and last used in July Object storage, backups Platform team Written to daily 5,200 Matched None needed Load balancer, retired API API team No requests this month 780 No requests this month Analytics warehouse Data team Queried every day 8,800 Matched None needed - Flagged
- 4 of 8 · GBP 12,980
- Reclaim
- 0 · GBP 0
- Keep
- 0 · GBP 0
- Undecided
- 4 · GBP 12,980
4 of 8 resources look idle or duplicated; you chose to reclaim 0, costing GBP 0 a month.
A resource that looks idle may be a disaster-recovery copy or a seasonal job; each is a question for its owner before anything is switched off [S1].
0 you chose to keep stay out of the total and 4 are undecided; the one with no owner is the first to find an owner for [S2].
Relies on: S1 · Cloud billing export and usage metrics (fictional)S2 · Resource owner register (fictional)
The cloud finance lead with the owning teams decides. Approving sends the reclaim requests to the owners. It does not switch off or delete any resource.
Not ready. 4 flagged resources are still undecided. Confirm or dismiss each one first.
Not done here. It does not switch off, resize or delete anything: owners do, after reading each request.
Settle the changes to the customer terms of service
- What happened
- The company is changing its terms of service to cover its new AI assistant, and customers will be told next month.
- The question
- Which proposed changes alter what customers have agreed to, and which are only wording? Asked by: Commercial counsel.
Settle the changes to the customer terms of serviceAccept or reject each change. The terms below are rebuilt from your decisions, and the changes that alter a commitment to customers are marked.Open
Accept or reject each change. The terms below are rebuilt from your decisions, and the changes that alter a commitment to customers are marked.
Accept or reject each change. The text below is rebuilt from the changes you accept, and changes that alter what people must do are marked.
Terms of service, section 9: data and the assistant (fictional)
The Customer's data remains the Customer's property.
Reworded
The Customer's data stays the Customer's property.
Wording only; no change to the commitment.
The Provider may use the Customer's data only to provide the services.
Added Changes what people must do
The assistant reads only the data sources the Customer approves, and the Customer may withdraw approval at any time.
Adds a commitment about what the assistant may read, and a right to withdraw.
The Provider will tell the Customer of any security incident affecting its data without undue delay.
Reworded Changes what people must do
The Provider will tell the Customer of any security incident affecting its data within seventy-two hours.
Replaces a flexible duty with a fixed deadline.
The Provider may change these terms by giving thirty days' notice.
Reworded Changes what people must do
The Provider may change these terms by giving sixty days' notice.
Doubles the notice period customers receive.
The text after your decisions (4 lines)
- The Customer's data remains the Customer's property.
- The Provider may use the Customer's data only to provide the services.
- The Provider will tell the Customer of any security incident affecting its data without undue delay.
- The Provider may change these terms by giving thirty days' notice.
4 proposed changes: 0 accepted, 0 rejected; 0 accepted change(s) alter what customers have agreed to.
A fixed incident deadline and a withdrawable approval are commitments the company must then be able to keep, so security confirms it can before they are accepted [S1].
The wording-only change carries no new commitment [S2]. The terms below are rebuilt from your decisions and have 4 lines.
Relies on: S1 · Proposed change list for the terms (fictional)S2 · Current terms and the assistant design note (fictional)
Commercial counsel with the head of security decides. Approving settles the wording to take to the executive team. It does not publish the terms or notify any customer.
Not ready. 4 changes are still undecided. Accept or reject each one first.
Not done here. It does not decide whether a commitment is wise or lawful: counsel and security do.
Decide whether to build or buy the customer assistant
- What happened
- Product wants an AI assistant for customers, and the executive team must decide whether to build it or buy a platform.
- The question
- At what customer volume does building beat buying, on the assumptions we can change? Asked by: Chief technology officer.
Decide whether to build or buy the customer assistantChange the four assumptions. The three-year cost of each route is recomputed, and the difference says which is cheaper on those assumptions.Open
Change the four assumptions. The three-year cost of each route is recomputed, and the difference says which is cheaper on those assumptions.
Change the assumptions. Every result and the answer below are recomputed from them, and a warning appears when a result crosses a stated limit.
- Cost of buying over three years
- GBP 2,160,000
- Cost of building over three years
- GBP 1,548,000
- Buying cost less building cost
- GBP 612,000Positive means building is cheaper over three years.
Over three years buying costs GBP 2,160,000 and building costs GBP 1,548,000; buying less building is GBP 612,000 (positive means building is cheaper).
The answer turns on how many customers use the assistant and what the licence costs per customer [S1].
The build figure is an engineering estimate and the run cost per customer is the least certain number in it [S2].
Relies on: S1 · Customer pipeline and vendor quotation (fictional)S2 · Engineering estimate for the build (fictional)
The chief technology officer decides. Approving records these assumptions for the executive team's discussion. It does not commit budget or sign anything.
Ready.
Not done here. It does not choose a vendor or commit spend: it shows how the answer moves with the assumptions.
Where a person decides, and what is never done
Every example above ends with a named person approving it. Approving means exactly what is written here, and nothing more.
- SOP IntelligenceDecides: The site reliability lead with the on-call engineersApproving lets: Approving releases this text as a draft runbook for review. It is not the approved version until a second person approves it, and it runs nothing.Never: It does not run any step or change any system, and it does not add a step the incident record does not show.
- ComplianceDecides: The head of securityApproving lets: Approving sets what the assistant may read for this customer's pilot. It does not switch the assistant on, and it can be narrowed or revoked at any time.Never: It does not read other customers' data, source code or billing records, and it never answers from a source a person has not classified.
- Financial IntelligenceDecides: The cloud finance lead with the owning teamsApproving lets: Approving sends the reclaim requests to the owners. It does not switch off or delete any resource.Never: It does not switch off, resize or delete anything: owners do, after reading each request.
- Legal IntelligenceDecides: Commercial counsel with the head of securityApproving lets: Approving settles the wording to take to the executive team. It does not publish the terms or notify any customer.Never: It does not decide whether a commitment is wise or lawful: counsel and security do.
- Executive IntelligenceDecides: The chief technology officerApproving lets: Approving records these assumptions for the executive team's discussion. It does not commit budget or sign anything.Never: It does not choose a vendor or commit spend: it shows how the answer moves with the assumptions.
Always a person in technology
- The on-call engineer decides when to roll back a release; the platform never does.
- The head of security decides what an assistant may read.
- Counsel decides what is said to a customer after an incident.
The capabilities behind these examples
- Create a procedure from the sourceWriting a clear, complete procedure from scratch is slow, and the written version often differs from what people actually do.
- Review and approve, as peopleA procedure that nobody accountable has approved is a liability, however well it reads.
- Transition AssuranceA handover is usually a conversation and a folder.
- Govern the AI your organisation usesTeams adopt AI tools faster than anyone records them, who approved them or what they may touch.
- Permissions, approvals and an audit trailWhen something goes wrong, nobody can say who could do what, who approved it and when.
- AutoDiscover from a folder you approveNobody has time to organise years of files before they can benefit from any tool.
- SaaS and Cloud Waste IntelligenceAnalyse authorised software usage, subscriptions, commitments, cloud resource usage and spend, and identify optimisation opportunities with evidence and permissioned…
- Financial Anomaly InvestigationsConnect unusual financial records with the invoices, payments, vendors, contracts and evidence around them, and support investigation, human decisions and a documented…
- Legal Drafting and Negotiation SupportReviewed first drafts, redlines, clause comparisons, playbooks and negotiation questions.
- Privacy and Data RightsData-rights requests, data-processing agreements, breach-response support, retention, transfer rules and privacy obligations, joined to the privacy workspace and the…
- Technology Investment AdvisorA technology portfolio with cost, owner, use and value evidence.
- What-If Decision SimulatorA person changes stated assumptions and sees the effect on verified baselines through documented formulas.
All industries · All demonstrations · Ask for a demonstration on your own sector
The other industries
- HealthcareClinical routines that are written down once, followed everywhere and checked in practice.
- ManufacturingLine procedures that survive a retirement, a new shift and a new product.
- BankingControls and procedures that a supervisor can trace, from the rule to the person who did the work.
- InsuranceClaims and underwriting routines applied the same way by every handler.
- RetailStore routines that are the same on a busy Saturday in every shop.
- LogisticsDepot and driver procedures that hold across sites, subcontractors and seasons.
- EducationSafeguarding and operational routines every member of staff can find and follow.
- Professional servicesEngagement methods captured once, so quality does not depend on who is staffed.