RePerspective LabsDecision guide
We need to use AI. Where should we start?
Explore a customer-onboarding decision. Change the conditions, reconsider the fit and draft an opportunity brief.
RePerspective Labs · Learning and operating judgment
A small mental map
AI systems infer from inputs to produce outputs such as predictions, content or recommendations. Generative AI produces content, such as a draft explanation.
- Rules-based automation: route a complete form to its assigned reviewer using specified conditions.
- AI assistance: interpret varied information or draft a response for review.
- An AI agent: use a model to choose steps and tools toward a goal, within the permissions provided.
These approaches can be combined. A workflow can use AI while keeping its sequence and approvals explicitly defined. Evaluate outputs in the setting where they will be used.
The fictional situation
A business-services firm wants customers to begin receiving service smoothly after signing a contract. Customers repeat information, employees chase documents and the start date is uncertain. Leadership asks the onboarding manager to find a useful place for AI. The team has not agreed what success means or established what causes the delay.
All organizations, observations and changed conditions in this exercise are invented for learning.
2 / Choose
Where would you begin?
Choose an opening question to reveal its contribution, its limits and a useful next step. You can explore all three.
What outcome matters most?
Agree whose experience should improve.
Make value explicit.
A dependable start date, lower customer effort and lower delivery cost could lead to different priorities. Agree whose experience should improve and what would count as progress.
Next: trace the work that creates that outcome. An agreed ambition still needs evidence about what is happening.
Under the changed conditions
Where does work wait or repeat?
Follow the work across team boundaries.
See the whole journey.
Looking across teams can expose waiting, repeated requests and returned cases that a task-level demonstration misses. Include customers and the employees who handle exceptions.
Next: relate those observations to the outcome. Time-box discovery and keep space for a different way of delivering the service.
Under the changed conditions
What could become possible?
Imagine a different service experience.
Give imagination a useful direction.
Sketch an experience where the customer supplies information once and can see the next step. AI can help generate and challenge alternatives before the problem is fully defined.
Next: test the assumptions behind the idea with customers, employees and operating evidence. An attractive concept is a hypothesis.
Under the changed conditions
Outcome clarifies value. Tracing work exposes dependencies. Exploring possibilities challenges inherited assumptions. Move between them as you learn.
3 / Reconsider
Let evidence change the next move.
Inspect the fictional findings, then choose a next investigation. Change the conditions to see how the reasoning changes.
Initial findings
The work is waiting between teams.
- Reviewed cases wait for a weekly approval meeting.
- Sales and finance use different meanings of “complete,” causing returns.
- Customers want a dependable start date; a broader baseline is still missing.
Changed conditions
Now the difficult work is interpreting documents.
- The team has aligned completion criteria and clarified approval ownership.
- In the cases reviewed, approval waiting has fallen and reading varied documents now takes the most effort.
- Authorized source documents and reviewer capacity are available; AI performance and net benefit have not been established.
What would you investigate next?
AI document assistance
Interpret documents and prepare a draft for review.
Potential local benefit; outcome still uncertain.
Document assistance might reduce preparation or rework. These findings do not show that faster preparation will change the approval wait or the customer’s start date.
Investigate: how much of the journey this work explains, and whether added review effort would offset the benefit.
Under the changed conditions
A more promising AI candidate now.
The remaining work involves varied documents. Compare AI-assisted extraction or drafting with the current method on representative cases, including exceptions. Keep outputs reviewable against their sources.
Check: accuracy, omissions, reviewer effort, total journey time and operating cost. Use the result to decide whether to extend, revise or stop.
Clarify handoffs and approval ownership
Agree what complete means and who can decide.
The stronger next investigation in this situation.
Clarify what makes a case ready, who may approve it and whether the weekly meeting is required. Test a shared case view and an agreed review arrangement with the people who own those decisions.
Check: a broader sample, necessary controls, exception routes and total customer waiting before attributing an improvement.
Under the changed conditions
Useful maintenance; a less compelling new focus.
Keep the shared definitions and ownership working. The updated findings suggest that another broad handoff review may add less than investigating the remaining document work.
Check: whether the improvement holds across different cases. If it does, move the next experiment toward the new constraint.
An agent for the whole journey
Delegate coordination and next actions across onboarding.
The proposed scope is ahead of the evidence.
An agent could coordinate permitted actions, but it would still depend on reliable case information and agreed decision rights. Unresolved definitions and ownership would remain part of its environment.
Investigate: whether a flexible sequence of tool actions is actually needed, and which boundaries and evaluations would make a bounded trial appropriate.
Under the changed conditions
A bounded task may teach you more first.
Clearer handoffs improve the conditions for an agent, but they do not establish a need for one across the entire journey. Document assistance can test value without delegating every next action.
Investigate: what benefit requires dynamic coordination beyond a defined workflow, then evaluate that smaller scope if the evidence supports it.
Carry the principle into another situation
A useful capability can still be aimed at the wrong constraint. Follow the outcome across the whole journey, identify what currently limits it and compare ways to change that condition. Revisit the choice when the work changes.
Here, an operating model means how the people, work, knowledge, decisions and measures fit together to deliver the service. A faster task and a better customer outcome must be examined separately.
4 / Imagine
What could the customer stop doing?
What could the customer stop having to do? Choose a burden to remove, then inspect the operating changes that would support the new experience. Describe the experience you want before deciding how to build it.
Repeating the same information
Provide information once, with appropriate reuse.
Supply information once, with permission to reuse it.
The experience depends on a shared case record, agreed meanings, current information and appropriate access. Someone must own corrections when sources disagree.
Possible technology role: rules can move known fields; AI may help interpret varied documents. Investigate the fit at the specific point of work.
Under the changed conditions
Chasing people for a status update
Make the next action and its owner visible.
Make the next step visible.
A useful status explains what is complete, what is missing, who owns the next action and how exceptions are resolved. That requires an accurate record of the work.
Possible technology role: ordinary workflow software can publish structured status. AI might explain it in plain language when that adds value.
Under the changed conditions
Working out which requirements apply
Explain requirements in the customer’s context.
Explain requirements in the customer’s context.
A relevant checklist depends on agreed rules, maintained knowledge and an owner for unusual cases. Customers need a route to question or correct the guidance.
Possible technology role: use rules for explicit conditions and assess AI assistance for explaining approved guidance. Test ambiguity and missing information.
Under the changed conditions
What could a customer or colleague in your operation stop having to do? Describe the experience you want, then identify what must be true for it to work.
5 / Apply
Give the opportunity a useful shape.
Bring this back to your operation. Draft a short opportunity brief using the five prompts below. Make your assumptions visible and identify what you need to learn next.
Agree a bounded scope, an appropriate baseline, acceptance and stop criteria, and a review point with the relevant owners. Compare total effort, including review and correction. Investigate apparent improvements before claiming the change caused them.
If the change becomes normal work, continue reviewing representative cases and reported problems. Recheck the approach when requirements, documents or customer needs change. The Measurement guide can help select useful evidence; the Transformation Hub connects the next stages.
Your opportunity brief
Notes stay in this page until you leave or reload. Copy them before leaving. Nothing is submitted.
Read an illustrative opportunity brief
Illustrative answer for the initial findings. This supplies no universal targets.
The outcome that matters
A dependable service start with less repeated customer and employee effort.
The work and people involved
The journey from signed contract through sales handoff, finance approval and service delivery. Include customers, employees, the process owner and the approval owner.
Evidence and unknowns
Check how common approval waits and returned cases are. Separate ordinary cases from exceptions. Record waiting, repeated requests and reasons for returns. Distinguish observations from explanations.
Alternatives and a useful test
Compare clearer completion criteria and approval ownership with AI document assistance. Initially, test a handoff change on a bounded group while preserving required checks. If document interpretation later becomes the stronger constraint, compare AI assistance with the current method.
Ownership and continued learning
The onboarding process owner coordinates with the approval owner. Review start-date reliability, total waiting, rework, employee effort and errors before extending the change. Identify who maintains shared knowledge and who acts on feedback.
Use AI to explore your own situation
Help me examine an opportunity to improve [operation] for [people]. The outcome I care about is [outcome]. Here is what I have observed: [evidence]. Here are my assumptions and unknowns: [list]. Separate observations from assumptions. Ask one clarifying question at a time. Help me imagine alternatives involving work design, explicit rules and AI assistance. For each alternative, identify its dependencies, likely failure modes and evidence that would change the recommendation. Finish with a bounded next investigation or test, including ownership and review. Do not invent facts, results or sources. Identify claims that need independent checking.
Use an appropriate, sanitized description. Verify the output against people and records that establish what actually happens. Asking a model to avoid invention cannot guarantee accuracy.
Sources and perspective
The scenario, observations and changed conditions are fictional and designed for practice. No customer results or benchmarks are claimed. Definitions draw on OECD: the AI system definition explained and Anthropic: workflows and agents. Context and evaluation are informed by NIST AI Risk Management Framework Core. The exercise and opportunity brief are RePerspective Labs editorial proposals.