Build operating discipline and responsible technology into the commercialisation plan.
Your learning outcomes
Distinguish repeatable growth from unsupported expansion
Assign operating, data and incident responsibilities
Evaluate AI-specific risks and prepare an evidence-led capstone
CLASS 8.1 / 60 MINUTES
Scale what works, not what looks busy
Define repeatability before expansion
Growth multiplies both strengths and unresolved problems. Establish whether the team can acquire, onboard, serve and retain similar customers without exceptional founder intervention. Use a cohort or customer-group view where useful, but define the groups and periods consistently. A large pipeline is not the same as delivered revenue, and delivered revenue is not the same as a sustainable operation. SignalBox should examine installation effort, report usefulness and support demand before expanding to machines or sectors that require different evidence and capabilities.
Build an operating cadence
Choose a small set of decision-relevant measures covering customer value, quality, delivery, cash and risk. Assign owners and a review rhythm. Record actions taken rather than producing dashboards that no one uses. Include leading indicators such as unresolved incidents or delayed onboarding, as well as lagging results such as renewals. Avoid optimising a measure in isolation: reducing support contacts is not success if customers stop asking for help because they cannot reach anyone. Review the meaning of each metric as the business changes.
Plan capacity and dependencies
Map the people, suppliers, infrastructure and knowledge needed to deliver the next stage. Identify single points of failure and practical recovery options. Document important procedures before the only person who knows them becomes unavailable. Test supplier and cost assumptions as volumes change. A service may need more training, monitoring or specialist review before it needs more advertising. State the conditions for expansion and the indicators that trigger a pause. Scaling responsibly preserves the ability to correct a problem rather than spreading it more widely.
PUT IT TO WORK
Your activity
Create a repeatability scorecard and a capacity map for ten fictional deployments. Do not assume that a successful pilot automatically scales.
Produce this
A growth gate with operational measures and named dependencies.
CLASS 8.2 / 60 MINUTES
Responsible AI and data in the product
Start with the role of AI
Ask whether AI is needed for the task and what decisions its output influences. Describe the intended use, users, data and human responsibilities before selecting a model. An impressive demonstration is not enough to establish suitability. In the running case, an AI-generated maintenance explanation might assist a trained reviewer but should not silently become a safety instruction. Compare an ordinary non-AI baseline and consider whether a simpler system is easier to validate, maintain and explain. Keep marketing language aligned with the actual role of the technology.
Identify model and tool risks
Consider inaccurate outputs, unsuitable training or evaluation data, drift, privacy exposure, prompt injection and excessive tool permissions where relevant. Distinguish a model's suggestion from an action taken in another system. Use bounded permissions, appropriate review and test cases that include failures and adversarial inputs. Do not claim that a single filter makes an AI product safe. Document the remaining limitations and the conditions in which the system should refuse, escalate or fall back to a simpler process. Evaluation must match the real task and consequences.
Treat assurance as evidence for a claim
Responsible deployment needs accountable decisions, suitable evaluation and ongoing review. Keep records of versions, data permissions, tests, incidents and changes. The DSIT introductory assurance guidance provides a route into the subject, but reading it does not certify a product or establish compliance with every applicable rule. Seek specialist advice for the sector and jurisdiction. A useful assurance question is: what evidence supports this specific claim for this intended use, and who accepts the remaining risk? Revisit that answer when the product, data or operating environment changes.
PUT IT TO WORK
Your activity
Write an intended-use statement, five failure tests and a human-escalation rule for an optional AI explanation feature. Include a non-AI fallback.
Produce this
A responsible-AI and data-risk register with testable controls.
CLASS 8.3 / 60 MINUTES
Make the final commercialisation decision
Assign authority and incident ownership
Name who can approve product changes, spending, access, claims and launch decisions. Keep responsibilities proportionate to the stage, but do not leave important decisions ownerless. Prepare a simple incident process: detect, limit impact, notify the right people, preserve useful evidence, recover and review. Avoid collecting more personal or confidential information than the task requires. For SignalBox, a misleading report or exposed customer record should have an owner and a response path before a public launch, not after the first complaint.
Review the portfolio of evidence
Bring together discovery, segment choice, rights, product tests, economics, funding, market entry and operating risks. Check for contradictions: an easy-installation claim may conflict with the pilot record; an attractive margin may omit site visits; a licensing plan may depend on unresolved ownership. Label gaps rather than polishing them away. Ask an independent peer to challenge the decision pack using the published capstone rubric. Peer review is a learning exercise, not formal professional approval or accreditation. Record which criticisms change the recommendation and which require further investigation.
Recommend a route and the next gate
Finish with a specific decision: proceed within a boundary, change the offer, license, partner, defer or stop. Explain the alternatives and the evidence that makes the recommendation reasonable. Define the next milestone, budget, owner, date and stop conditions. A commercialisation plan is a living decision record rather than a prediction that every assumption will come true. The strongest final submission shows what the learner knows, what remains uncertain and how the team can learn responsibly without making promises that its evidence or resources cannot support.
PUT IT TO WORK
Your activity
Review the capstone checklist. Write a one-page decision memo and ask a peer to challenge its weakest assumption before revising it.
Produce this
A final decision memo, evidence index and next-gate plan.
MODULE ASSIGNMENT / SELF-REVIEW
Build the evidence.
Prepare an operating and responsible-technology plan, then audit the entire course portfolio for contradictions before beginning the capstone.
Assignment self-review rubric
Criterion
Points
Repeatability and capacity evidence
25
Clear decision and incident ownership
25
AI and data risks with testable controls
30
Honest synthesis and next gate
20
Use the rubric to evaluate your own work or ask a suitable mentor. Written assignments are not uploaded or automatically marked.