Skip to main content

AI-Accelerated User Needs Leadership

· 5 min read
Dave Hulbert
Dave Hulbert
Builder and maintainer of Wardley Leadership Strategies

In our previous post, we explored how to use LLM-driven competitor simulations to anticipate and prepare for the moves of our rivals. But a purely external focus is not enough. To create lasting value, we must also have a deep and evolving understanding of our users.

Leaders default to visible requirements, yet competitive advantage emerges when you stretch beyond the backlog to hypothesise the needs users can’t articulate. Wardley Mapping, and its user needs-focused cousin, remind us that "what people ask for" is only the top layer. AI now gives us leverage to work the deeper layers without guesswork.

Anchor Wardley Maps on layered needs​

Wardley reminds us to map user needs before we map value chains. The practice of user needs mapping sharpens this discipline by distinguishing between expressed wants, expected needs, and hypothesised needs (see User Needs Mapping for a deeper dive). Expressed wants are like solution requests ("I want a mobile app"), expected needs are the basic capabilities that users assume (secure login, reliable search), and hypothesised needs are the strategic bets that create differentiation. Leaders can drift because the first two tiers are easy to quantify and put in a backlog, while the third is hazy.

Use your map reviews to test the balance across the tiers. For every initiative, ask yourself: which layer are we serving, and what signals prove it? If the board report only contains metrics for expressed and expected needs, you are funding predictability but starving discovery. Rotate your metrics to include measures of hypothesised learning, such as the time it takes to run an experiment, the number of latent needs validated, and the share of engineering capacity that is spent on discovery.

Combine LLM insight mining with situational awareness​

Large language models can now process user-generated content at scale. Recent work on LLM-driven analysis of product reviews has shown how models can extract product attributes, connect sentiment to satisfaction, and feed prioritisation frameworks like IPA-Kano to spotlight hidden demand patterns (Wei et al., 2025). Treat these systems as co-analysts that can surface the needs that are obscured by the backlog data. Pipe the findings back into your map: annotate components with the needs layer they reinforce and whether sentiment signals rising dissatisfaction or untapped appetite.

This feedback loop can shift doctrine conversations. Instead of arguing over whether a feature matters, your leadership team can inspect concrete user narratives that the model has clustered, ranked, and linked to satisfaction outcomes. This keeps the map grounded in the lived experience of users, rather than in internal assumptions.

Simulate the hypothesised layer with synthetic research​

LLMs are equally valuable before real data exists. They can be used to construct rich personas, draft proto-journeys, and stage conversational interviews that explore unmet aspirations. To guard against hallucination, it's important to seed the prompts with ethnographic notes, support transcripts, and market signals. The goal is not to replace field research, but to generate plausible hypotheses faster than you could schedule workshops.

Blend these synthetic insights with your Wardley Map: sketch the evolutionary path of the components required to serve the hypothesised need, mark the inertia that will resist change, and outline the doctrinal shifts (e.g., "Bias for the new" or "Focus on user needs") that will be required to respond. When the model highlights a recurring aspirational theme—say, "real-time guidance during critical incidents"—you can task product teams with small-scale experiments to validate or refute the need before committing significant capital.

Operational rhythms for AI-informed need discovery​

  1. Weekly signal review – Pair a narrative Wardley Map walkthrough with a quick, LLM-generated digest of emerging user sentiments and hypothesised needs. Decide what moves one stage closer to delivery, learning, or retirement.
  2. Monthly persona refresh – Regenerate synthetic personas and journey fragments using the latest market signals, and then sanity-check them with real customers or front-line staff.
  3. Quarterly doctrine check – Audit whether your leadership rituals still prioritise user needs. Are teams rewarded for retiring features that no longer meet expected needs? Are discovery sprints funded in proportion to the hypothesised bets on your map?

Leadership guardrails​

  • Stay evidence-led. Demand citations, transcripts, and source snippets to accompany every model insight. Don't allow "the AI says" as a justification.
  • Use multi-model triangulation. Run the same prompts through at least two models and compare the variances to spot any instability.
  • Keep humans in the loop. Make user researchers and service designers the stewards of the hypotheses; AI can accelerate their craft, but it can't replace their judgment.

Great leadership keeps the organisation obsessed with users even as AI accelerates analysis. By combining Wardley Mapping discipline with model-assisted discovery, you create a system that serves current expectations without losing the differentiating edge hidden inside unspoken needs.

References​