AI Opportunity Backcasting
Start with business friction, imagine ideal assistance, then work backward.
- Difficulty
- Moderate
- Time to result
- ~weeks to results
- Steps
- 6
- Confidence
- 94%
The framework begins with present business friction rather than a preferred model or product. For customer-facing work, imagine that every user could speak with a capable human who understood the desired outcome, then work backward to an AI-assisted version of that experience. For internal work, identify what makes the company feel inefficient: too few capable people, capable people assigned to low-value tasks, or teams failing to share context. Convert those observations into candidate workflows where AI assists, augments, or amplifies people. The initial implementation can be narrower than the ideal because model capabilities are improving. The output is a prioritized set of experiments tied to better products, smoother operations, stronger collaboration, or improved margins rather than an arbitrary list of AI features.
Origin
Extracted from Marketing Against The Grain, where Gabriel Hubert described how he guides founders toward AI opportunities by questioning their problems, envisioning ideal human assistance, and working backward.
Core principles
- 01Start with a real problem rather than an AI product.
- 02Imagine the ideal human-assisted experience before considering constraints.
- 03Separate customer-facing opportunities from internal operating friction.
- 04Use AI to augment scarce capable people before attempting replacement.
- 05Accept a degraded first version when the underlying technology is improving.
How to run it
- 1
Surface the real problem
Ask which product or operating problems feel most urgent, costly, or frustrating. Do not begin by prescribing a model or AI product.
Pro tip Ask what repeatedly makes leadership feel that the company is not running smoothly.
Watch out Starting with a fashionable tool can produce a technically impressive workflow with little business value.
- 2
Separate external and internal value
Classify each problem as customer-facing product improvement or internal company operation. Define what becoming better means in that context.
Pro tip For products, distinguish better performance from better perception of value.
Watch out Do not treat increased activity as proof of improved customer or business outcomes.
- 3
Imagine ideal human assistance
Describe how the experience would work if a knowledgeable human were immediately available to help the user accomplish the job. Remove unnecessary forms, navigation, and rigid interaction patterns from the imagined experience.
Pro tip Focus on the outcome the user wanted before software imposed its current interface.
Watch out Do not merely insert a chatbot into the existing workflow.
- 4
Work backward to feasibility
Break the ideal interaction into tasks that current models, software, and humans can perform. Preserve human review where uncertainty or consequences demand judgment.
Pro tip Launch a narrower degraded version if it still creates measurable value.
Watch out Do not promise autonomy beyond the model's demonstrated reliability.
- 5
Diagnose operating friction
Determine whether internal performance suffers from too few capable people, capable people working on the wrong tasks, or inadequate collaboration. Target assistance to the specific constraint.
Pro tip Look for recurring unstructured information that employees manually translate, route, or summarize.
Watch out Avoid framing augmentation as immediate employee replacement.
- 6
Run an outcome-based experiment
Test one candidate workflow and compare its effect on time, quality, customer experience, or margin. Use the result to refine or abandon the opportunity.
Pro tip Include user trust and adoption alongside efficiency metrics.
Watch out A faster workflow that creates more correction work is not a successful result.
In the wild
A travel product imagines each customer speaking to an agent who understands dates, flexibility, and destination preferences. The team works backward to a conversational intake that extracts structured trip criteria, asks only necessary follow-ups, and sends uncertain details for confirmation rather than forcing users through multiple date selectors.
→ Customers reach suitable options with less interface friction while retaining control over critical booking details.
A company identifies manual transfer of Slack discussions as an internal collaboration problem. An AI assistant detects a feature discussion, classifies it, summarizes the relevant context, and creates a properly formatted item in the product system for human review.
→ Ideas reach the correct workflow without employees repeatedly copying and restructuring conversations.
Common mistakes
Beginning with the solution
Choosing an AI product first encourages teams to retrofit it onto problems that may already have better solutions.
Automating the existing interface
Placing a model inside a poor workflow preserves assumptions that conversational or proactive software could eliminate.
Confusing augmentation with replacement
Early systems create more dependable value by helping capable people handle scarce time and information than by removing them wholesale.
Is it for you?
Best for
It is best for leaders redesigning customer experiences or internal knowledge-work processes around generative AI.
Not ideal for
It is not ideal for teams seeking a predetermined AI vendor before identifying a meaningful business problem.
From the transcript
“what problems they have right now that they feel top of mind”
“start from the ideal situation and then sort of roll backwards and break it down”
“what are the things that make you mad”
From the episode
AI Founder Reveals How AI Exposes Lazy Employees (#157)