The Idea Avalanche Problem
A 400-person innovation session generating three ideas per team produces around 120 submissions. A 48-hour open innovation challenge across a 5,000-person business can generate 600 or more. The constraint in these programmes is not generating ideas — organisations with engaged workforces have no shortage of ideas. The constraint is curation: how do you identify the 5–10 ideas worth developing further, before the decision window closes?
Traditional curation approaches — committee review, facilitator judgement, popular vote — fail at scale. A committee reviewing 120 submissions in a two-hour session is averaging one minute per idea. That is not evaluation; it is impression management. Popular votes reward the articulate and the socially networked, not the strategically sound.
The programmes that consistently surface the strongest ideas from large-scale sessions use a multi-layer filtering architecture rather than a single curation mechanism.
Filter 1: The Structured Brief Format
The first filter operates before a single idea is evaluated. A well-designed submission brief forces authors to make explicit choices that reveal both the quality of the idea and the quality of their thinking.
A submission format that consistently yields evaluable ideas includes: a one-sentence problem statement, a one-paragraph solution description, a statement of the key assumption the idea depends on, a resource estimate, and a single sentence articulating why this matters to the organisation right now.
This format does three things. First, it filters out undeveloped ideas — a participant who cannot articulate the key assumption their idea depends on has not yet thought the idea through. Second, it standardises the information available for downstream evaluation — every idea can be compared on the same dimensions. Third, it signals to participants that the programme is serious about implementation, not just ideation, which shifts the quality of what they submit.
In practice, a well-designed submission format reduces the volume of evaluable submissions by 20–30% (ideas that cannot be completed are implicitly filtered) while substantially increasing the quality of those that survive. This is the first compression.
Filter 2: Parallel Analysis
When the submission volume exceeds what a human review team can evaluate rigorously — typically above 40–50 submissions — a second filter layer is necessary. The analytical challenge is not assessing individual ideas in isolation; it is understanding the distribution of ideas across the problem space and identifying the genuinely novel submissions.
A structured parallel analysis of all submissions — processing simultaneously rather than sequentially — produces three outputs that are operationally useful:
Distribution mapping. Which problem areas attracted the most attention? Which are underrepresented? High convergence on a problem area is a signal of organisational salience. Low coverage of a problem area that is strategically important is equally informative.
Novelty detection. Which ideas represent a departure from current strategy, current technology, or current market framing? These are the ideas most likely to be overlooked in a human review biased towards the familiar.
Risk and assumption flagging. Which ideas depend on assumptions that are unvalidated, contestable, or resource-intensive? Surfacing these early allows evaluators to engage with the critical uncertainty rather than being surprised by it at the pitching stage.
This parallel analysis layer is where the economics of scale invert. A 400-submission session is not four times harder to analyse than a 100-submission session — if the analysis is structured and parallelisable, it takes the same time regardless of volume.
Filter 3: Peer Investment Voting
The third filter engages the collective intelligence of the participant pool itself. Peer investment voting — each participant allocating a fixed number of "investment tokens" across submissions — is a significantly more reliable signal than simple upvoting.
The mechanism is well-supported by research into prediction markets and collective intelligence. When people allocate a scarce resource (tokens) rather than an unlimited resource (upvotes), they engage in prioritisation rather than endorsement. They cannot vote for everything they find interesting; they must decide what they find most promising.
The statistical properties of investment voting also make it more robust against social dynamics. A charismatic team leader can mobilise their network to upvote their submission; they cannot force their network to reallocate from other submissions they have already committed tokens to.
In a 400-person session with 30 tokens per participant, the investment distribution across 120 submissions produces a reliable Pareto signal: the top 10% of submissions by investment value typically capture 40–50% of total tokens. This is not the final selection decision — it is a calibrated signal about organisational judgment that complements the parallel analysis layer.
Filter 4: Facilitator Synthesis
The fourth filter is human, and it is the most important. After the structured brief format, parallel analysis, and peer investment voting have compressed the field to a manageable set of high-signal submissions, the facilitator's role is synthesis — not further filtering, but integration.
Synthesis at this stage means: identifying the ideas that are individually strong and collectively complementary, flagging ideas with high investment scores that did not surface in the analytical layer (potential blind spots in the analytical frame), and presenting the final shortlist in a format that enables executive decision-making.
The facilitator synthesis is where domain expertise, organisational context, and strategic judgment are irreplaceable. No automated system can assess whether an idea with strong peer investment signals aligns with an acquisition in progress that is not yet public, or whether a technically novel submission depends on a supplier relationship that is being renegotiated.
A Real 400-Person Innovation Session Structure
The following structure was used by a European financial services organisation running a 48-hour internal innovation challenge.
Hour 0–4: Brief release. Structured submission format distributed with detailed brief. Teams of 3–4 formed across 6 participating business units.
Hour 4–20 (overnight): Submission development. Facilitator team available for brief clarification. Submissions accepted on a rolling basis.
Hour 20: Submission close. 127 submissions received from 38 teams.
Hour 20–22: Parallel analysis run across all 127 submissions. Distribution map, novelty flags, and risk signals produced.
Hour 22–24: Peer investment voting. Each participant allocated 25 tokens. Top 30 submissions by investment captured 62% of total tokens.
Hour 24–26: Facilitator synthesis. Cross-referencing parallel analysis outputs with investment distribution. 12 submissions identified for full development track.
Hour 26–36: Selected teams develop full business cases using a structured brief template.
Hour 36–42: Executive review panel assesses 12 business cases. Three approved for pilot funding with immediate resource commitment.
Hour 42–48: All-hands synthesis session. Full investment distribution shared with participants. Rationale for shortlist decisions explained.
The result: three ideas funded within 48 hours, seven more placed in a 90-day evaluation pipeline, and a documented record of all 127 submissions available for future reference.
The Compounding Value of Multi-Layer Filtering
Each filter layer in this architecture serves a distinct analytical function. None is redundant. The structured brief format creates evaluable inputs. Parallel analysis provides pattern recognition at scale. Peer investment reveals collective organisational judgment. Facilitator synthesis applies strategic context that no automated system can replicate.
The investment is in the architecture, not the volume management. Platforms like CoVision are built around this multi-layer model: structured submission capture, real-time synthesis across the full submission set, peer funding mechanisms, and facilitator-ready output. When the filtering architecture operates in parallel with the session rather than after it, the 48-hour window is sufficient to move from 400 ideas to three funded pilots.
