Deloitte. Horizon · AI Transformation App blueprint →
Product Concept — Internal Working Draft

Flare.

An internal task marketplace that converts unassigned bench capacity into an open, merit-based delivery engine — work is posted, competed for, reviewed, and rewarded in a transparent point economy.

Status
Draft v0.4
Date
August 12, 2026
Origin
E. Rodrigue × A. Savard
Product
Flare
Currency
Lumens
Pilot scope
Polaris · AI Transformation
1.0

Executive summary

Customer runs ~600 practitioners at roughly 70% utilization; the remaining capacity — 60 to 150 idle practitioners at any time — represents thousands of hours per week with no visibility, no direction, and no measurable output. Work attribution happens behind closed doors and rewards proximity over merit. Flare replaces that opaque allocation with an open marketplace: any practitioner or project can post a fully specified task with a bounty; anyone on the bench can compete to deliver it; the poster accepts the best submission and the bounty transfers as points. Leaderboards and performance-review integration make participation genuinely worthwhile — and the scoreboard doubles as the firm's talent radar. The concept pilots on Horizon's own backlog (Polaris, AI Transformation) before any wider rollout.

2.0

Problem statement

~600
practitioners in Customer
the population in scope
~70%
utilization
the missing 30% is invisible today
60–150
idle at any time
a continuously rotating pool
1000s
of unaccounted hours / week
"Where are these hours going?"
Allocation

Work is assigned behind closed doors

Senior managers and directors originate work and hand it to their own people. No one checks whether capacity or better-suited talent exists elsewhere; the operating driver is patronage, not efficiency.

Meritocracy

Networking outperforms competence

Careers advance by attaching to two or three patrons. Equally capable practitioners without those relationships stay parked on low-value work. Two years of promotion stagnation has broken the traditional incentive lever entirely.

Transparency

Freed hours absorb invisibly

Non-billable hours are quietly redirected — in some cases to conceal delivery costs on undersold engagements. Leadership has no instrument to see or steer where liberated capacity flows.

Friction

Small work items never execute

Commissioning one or two hours of a developer's time requires a WBS, negotiation, and meetings. The transaction cost exceeds the task, so the work simply doesn't happen.

Scalability

Junior capacity cannot be micromanaged

No manager can direct 30 rotating juniors by hand; even superstar teams underperform when roles blur (the York precedent). A churning bench pool needs fixed structure and self-service, not heroic supervision.

Quality

The bench is a black box

Bench practitioners upskill — or don't — with no track record and no way to prove capability. When bench labor does reach projects it churns (developers rotating every two weeks) and ships unrequested work with no quality gate.

3.0

Product overview

Flare is an internal, open task marketplace governed by game mechanics. It democratizes access to work and makes the decision of what gets worked on transparent. Positioning is deliberate: Flare is a tool that occupies the bench, not a new owner of the bench — resource owners keep their people, their billing, and their rake; HR structures are untouched.

Marketplace

An open task board

Any practitioner or project posts a task publicly — criteria, template, objectives, deadline, bounty. Everyone in scope sees everything available. Nothing is allocated in private.

Workforce

A self-directed bench

Bench and PRD-time practitioners select tasks that match their skills and work them fully asynchronously — the posting is complete enough that no meeting is ever required.

Incentives

A point economy worth playing

Accepted work earns currency; leaderboards rank contributors; the record of won work is what converts into outcomes practitioners actually value.

Intelligence

A live talent radar

The scoreboard is a continuously updated map of who actually delivers — surfacing strong practitioners without a staffing request, an interview, or a favor.

"Démocratiser l'accès au travail, centraliser la décision du travail… il faut que tout le monde puisse voir qu'est-ce qui est à faire."

Source conversation, ~1:07 — the founding principle
4.0

Operating model

Release 1 is deliberately minimal — post, submit, accept, transfer. Every additional capability layers onto that loop without changing it.

Post

A fully specified task — criteria, template, steps, expected result — with a bounty. Posting draws on the poster's budget; it is never free.

Pick up

Practitioners browse the open board and signal what they are working on, preventing blind duplication of effort.

Submit

One or several practitioners deliver against the stated criteria — at any hour, with no coordination overhead.

Review & accept

The poster reviews submissions and accepts the best one. Low-quality volume submitters are filtered out reputationally.

Transfer

The bounty transfers to the winner; the leaderboard updates; the transaction enters both parties' track records.

CapabilityDescriptionRelease
Open task boardPublic postings visible to all of Customer or a defined sub-offering; fully specified tasks ("the 15 steps, the template, the objectives — go for it").R1
One-click acceptanceAcceptance is the payment event: the poster accepts a submission and points transfer instantly.R1
LeaderboardRunning contribution ranking per scope (initially: Polaris contribution leaderboard).R1
Skills profile & track recordEvery practitioner maintains a living CV — skills, preferences, differentiators — validated over time by won tasks.R1
Work-in-progress signalingPractitioners declare what they are attempting; hard tasks surface collaboration organically (reference model: the Mario Maker community board).R1
Priority boostingPosters raise the bounty to raise urgency — priority becomes a visible price rather than a favor.R1
AI onboarding gateA Polaris module generates a ~3-hour onboarding course from a project's primary material; passing it qualifies a practitioner to compete on that project's tasks.R2
Reverse auctionsTasks posted without a fixed bounty; practitioners bid the points they would accept and present their skills; the poster selects.R2
Private postingsDirect assignment to a chosen practitioner with escrow-style acceptance (reference model: RentACoder).R2
Budget integrationTask bounties drawn from and reconciled against real project budgets (see §5).R2
5.0

Incentive & currency model

The currency — Lumens — is an invented unit deliberately pegged to real project value (≈ $1,000 per point). The anchor already exists: as AI absorbs the junior-consultant layer, engagements carry a real, measurable token cost (e.g., $20–40k on a representative project). The design requirement is that posting a task draws on the poster's project budget — otherwise the marketplace becomes free internal labor.

FlowMechanism
EarnWin a task — the bounty transfers on acceptance. Seeding variants: points for posting use cases to the hub, plus points per page visit earned.
SpendPosting escrows the bounty from the poster's budget while the task is live — refunded in full on cancel, no fees; raising the bounty purchases priority. Ultimately, a project could be estimated and budgeted in points.
Review integrationPoints read alongside utilization at mid-year and year-end: 50% utilization with strong points = highly productive PRD; 50% utilization with zero points and no pursuits = managed out.
The rakeResource owners retain their per-hour rake, so bench-to-billable conversion through the marketplace always benefits them. The platform takes a commission on hours it routes; HR ownership is unchanged.

"Je te donne 5 h, mais ça se peut que ça t'en prenne 20, puis je te paie 5… c'est exactement comme ça que ça fonctionne en ce moment — mais informellement."

Source conversation, ~1:09 — why the currency must remain carefully structured (see Risk R-1)
6.0

Governance & safeguards

Market integrity

Posting locks budget

Posting escrows the bounty from the poster’s own Lumen balance while the task is live — refunded in full on cancel, never a fee. The finite budget prevents an engagement from being AI-decomposed into a thousand simultaneous micro-tasks, and counterweights the risk of 500 bench-hours converging on a single 10-hour job.

Quality

Curated intake

Postings flow through a curating team rather than a free-for-all; the quality of task definitions determines the marketplace's credibility.

Conduct

Reputational anti-spam

Volume submitters of low-quality work are filtered out by posters and discounted by the system. Review capacity is a recognized operating cost that scales with task granularity.

Oversight

Moderated, observed, steered

Leadership actively watches the market — which tasks attract effort, which stall, which solution paths emerge. Five practitioners attempting one problem down three paths is treated as intelligence, not waste.

7.0

Implementation roadmap

PhaseScopeDescriptionExit signal
0 — PilotPolarisMinimal release on Horizon's own backlog: open postings, submissions, one-click acceptance, point transfer, a Polaris contribution leaderboard. Free-ish posting.Tasks reliably picked up and accepted; review loop sustainable.
1 — DogfoodAI TransformationReal AI Transformation needs posted (e.g., two-pagers from transcripts). Deliberate price discovery — an unanswered task signals an underpriced bounty.Repeat posters; stable price signals per task type.
2 — SeedDeclared AI ambitionsPractitioners who publicly state AI ambitions in year-end aspirations (workshop QR-code intake) are plugged into the leaderboard and hub.Organic participation beyond the seeded cohort.
3 — ScaleCustomer benchBench chief-of-staff mandate routes idle practitioners into the funnel; business-development tasks added. Resource owners remain advantaged throughout.Measurable bench-hour engagement; first bench-to-billable conversions.
4 — InstitutionalizeThe Horizon methodTop scorers who demonstrate vision become pod leads; the leadership panel forms from the leaderboard, superseding the Carson group. Points enter formal performance review.Majority of eligible PRD hours flowing through the marketplace.

Success measures (proposed — to be validated)

Adoption

Weekly active solvers · tasks posted vs. completed · repeat posters

Efficiency

Median time-to-completion · review latency · % tasks with no takers (price signal)

Outcomes

Bench hours engaged · bench-to-billable conversions · pod leads sourced from the leaderboard

8.0

Strategic value

Talent

A self-selecting pipeline

High scorers identify themselves — no staffing request required. They convert from bench to billable Horizon work while the resource owner's rake keeps flowing; every party is structurally aligned.

Performance

A defensible filter

A premium on efficiency and AI capability: strong practitioners rise immediately; those with neither points nor billable hours become visibly unstaffable, giving the firm an evidence-based path to workforce renewal.

Intelligence

The market as sensor

Task uptake, stalls, and emergent solution paths tell leadership more about real capability and demand than any reporting cycle.

Position

The allocation layer

Whoever operates the marketplace operates the bench. The machinery through which idle capacity, upskilling, and talent discovery flow — with the stated ambition that within one to two years, the majority of the practice works the Horizon method.

9.0

Risks & mitigations

IDRiskSeverityDescriptionProposed mitigation
R-1Compensation exposureHIGH Points mapped to real hours or money create a written record of under-compensated work — "codified wage theft" (paying 5 for work that took 20; documenting 80-hour weeks on 40-hour salaries). Fixed salary covers standard hours; points stay symbolic — a record of contribution, not compensation. Legal review before any currency-to-money linkage.
R-2Market failureMEDIUM Underpriced tasks get no takers (especially with deadlines); overpriced tasks burn disproportionate bench-hours (50 practitioners × 10 hours on one 10-hour job). Finite escrow-based posting budgets, observed price discovery in the pilot, and moderation. Accept that some parallel effort is valuable exploration.
R-3Review burdenMEDIUM Granular tasks multiply submissions; the acceptance step demands real management capacity — the tool's principal operating cost. Task-granularity guidelines; reputational filtering; review capacity planned per phase.
R-4Residual favoritismMEDIUM Posters can still select their preferred people; some preference is legitimate (context, continuity). Open visibility of every award; selection patterns auditable on the board itself.
R-5Political positioningMEDIUM The tool must never read as a land-grab on resource owners' people. Broker positioning: owners keep billing and rake; HR ownership unchanged; platform takes a commission on flow only.
R-6FundingMEDIUM Prize pool and budget linkage are unfunded; sponsorship required for review-policy changes (points counted at mid-year/year-end). Executive sponsorship secured before Phase 2; no prize-pool funding is sought.
10.0

Open decisions

DecisionOptionsCurrent leaningStatus
Product nameFlare · ConstellationFlare — formally "Deloitte's Flare."Decided (Aug 12, 2026)
Currency nameStardust · Photons · LumensLumensDecided (Aug 12, 2026)
Currency realismSymbolic · budget-pegged unitSymbolic in R1; budget-pegged in R2 pending R-1 reviewOpen
Reward mixBounty + XP record · bonus · vacationBounty and XP only in R1 — draws, Green Stars and util relief withdrawn Aug 2026Decided
SponsorshipReview-policy changePoints count at reviewNot yet raised
Intake curationCentral curating team · per-offering curatorsCentral (pilot)Open
A

Appendix — Reference models

Bug bounty platforms (Bugcrowd)

Tasks posted with bounties; contributors build verifiable track records from solved bounties, independent of who they are.

RentACoder-class bid sites

Public projects, private bids, escrowed funds released on delivery, deadlines — the Release 2 auction model.

Mario Maker community (server shutdown)

Self-organized boards of who-is-attempting-what; top performers sharing techniques on the hardest problems; effort de-duplicated without central direction.

Karaoke queue tipping

Priority as an honest, visible price rather than an informal favor.

Source: working conversation, "Leadership | Horizon & AI Transformation" (E. Rodrigue / A. Savard). Full structured extraction with timestamps: transcript-notes.md. Interactive walkthrough of the application structure: app blueprint.