An AIPM community award
AI Project
of the Year
Edition 1 (2026). Working-team review for Ricardo Vargas and Antonio Nieto-Rodriguez.
30 minutes · 5 blocks · one decision to leave with
Everything in this deck traces to the operating rules, the single source of truth. Where this deck and the rules ever differ, the rules prevail. Press the right arrow to advance, "g" for the slide grid, "t" to start the timer.
What we will cover
Five blocks, and the last one is the reason we are meeting this week.
Block 1 · The gap
Two kinds of awards already exist.
Neither asks the whole question.
What AI awards doThey judge the product
Business excellence, model novelty, innovation narrative. Strong on what was built.
Silent on whether it was governed, adopted, or delivered.
What project awards doThey judge the delivery
Planning, governance, risk, benefits realisation. Strong on how it was run.
No nuance at all for how the AI itself was applied.
The question in the middle
Was the AI responsibly governed, genuinely adopted, and delivered with real project discipline?
A team that did all three has nowhere to go today. That is the whole opportunity, and it is the only thing this award is trying to be.
Not a model. Not a tool. Not a 24-hour prototype. Delivered work, with evidence.
The design, on one page
Block 2 · How we got here
We did not design this from taste.
We benchmarked 25 programmes first.
- Brazilian PM chapters and international PM bodiesPMI Project of the Year, PMI PMO of the Year, APM, IPMA, and seven PMI chapters
- AI and innovation awards, and hackathons and commercial recognitionStevie, BIG, AI Breakthrough, Webby, ITU AI for Good, Australian AI Awards, GitHub Stars, and the credential-mill end of the market
The headline finding
Across all 25, no award owns both: recognising responsible AI inside delivered project work, and returning feedback that builds the practitioner.
- The PM bodies own rigorous delivery rubrics and say nothing about AI.
- The AI awards reward products, judged against generic business-excellence language.
- The hackathons judge greenfield builds under time pressure.
- Only IPMA returns a written feedback reportand it is paid, on-site, and costs EUR 600 to 4,000 to enter
That intersection is unowned, and it is time-limited. It is the reason to launch in 2026 rather than 2027.
One decision we can win on immediately
PMI declares its Project-of-the-Year weights proprietary.
We publish ours, before the call opens.
Publishing a weighted rubric is the cheapest trust-builder in the entire benchmark. It costs nothing but the discipline of deciding the weights first and then not moving them.
Rules clause 13.6: once the window opens, rules, rubric, weights and eligibility do not change for entries already in progress.
Nine design decisions
All nine are decided for edition 1.
| # | Decision | Where it landed |
|---|---|---|
| 1 | Categories | One overall award plus two special recognitions, single funnel, single rubric |
| 2 | Eligibility and window | Beyond proof of concept, global, vendor-neutral, fixed calendar floor |
| 3 | Nomination model | Self or open peer, one Lead Nominator, draft then final |
| 4 | Jury and conflicts | Mixed panels of about seven, calibrated, auto-decline plus disclose-and-recuse |
| 5 | Community vote | Jury only. No public vote in edition 1 |
| 6 | Scoring rubric | Six weighted criteria, 1 to 10, weights published |
| 7 | Platform | Built in-house on Cloudflare, no third-party entry SaaS |
| 8 | Recognition and sponsorship | Free end to end, no cash, owner-funded, no sponsor |
| 9 | Geographic structure | Regional to global, three macro-regions |
The four that were left open
All four are now resolved, and the page says so.
- Eligibility window. A fixed calendar floor of 1 January 2025, not a rolling window.Checkable by the entrant with one comparison and no arithmetic
- Sponsorship. Owner-funded. No sponsor, and no sponsor juror.
- Community Choice. Jury only. No public vote.
- Person-level award. Project only for edition 1, deferred to a future edition.
Each is recorded in the rules with the reasoning, so a future edition can reopen it knowing why it was closed.
The core team reviewed it, and two things changed
The review worked. Both amendments are recorded and numbered.
Amendment 13.8.5The calendar was re-baselined
The old plan launched in mid-August, inside the European summer break the ambassador network had already flagged, and ran the pitches and ceremony into late December, when unpaid juror attention cannot honestly be asked for.
Now: opens 8 September, ceremony 17 February 2027, with a scheduled shutdown from 12 December to 10 January written into the calendar as downtime rather than hidden risk.
Amendment 13.8.6Eligibility stopped moving on its own
The window was a rolling 18 months measured back from the deadline, so the deadline was doing two jobs at once. Every schedule change silently moved the eligibility floor and could disqualify entries already in progress.
Now: a fixed floor of 1 January 2025. A schedule change is a schedule change and nothing more.
Block 3 · What an entrant lives
One entry. One rubric. Up to three ways to be recognised.
- AI Project of the YearThe overall award, always conferred
- Excellence in Responsible AIDesignated from the finalist pool
- Excellence in Agentic AIOnly where a finalist clears the bar. A bar to clear, not a slot to fill, and not conferred if nobody qualifies
You never apply for the specials. Entering once puts you in the running for all three.
Can you enter
Four tests, and the entrant can check all four alone.
- Completed or operationalized on or after 1 January 2025, and on or before 5 November 2026One published date, one comparison
- Beyond proof of concept. Delivered, piloted in a real setting, or operationalized, with evidenceIdeas, models, tools and demos alone are not eligible
- AI is a meaningful component of the work, not a decoration
- Global and vendor-neutral. Any sector, any country, any size of organisation
The window is applied strictly: no discretionary extension, and no case-by-case exception for a project that falls just outside it, however close.
What an entry actually is
Bounded on purpose, so a small team can win it.
- A 180-character first gate. What the project did, in one lineBorrowed from GitHub Stars: it forces clarity before effort
- Six questionnaire sections, each mapped to a rubric criterion and printing its own point value
- An evidence pack with per-criterion guidance, accessible without a login, capped and typed
- A short video, so judges see the real application rather than the prose about it
- Draft, then final. Resume by emailed magic link, submit once
What it costs
Free end to end. Including the proof.
What is freeEverything
No entry fee. No judging fee. No charge for the badge, the certificate, the logo use, or any proof that you won.
Why we say it out loudThe credential-mill model
The benchmark's sharpest negative case is free to enter and free to "win", then charges USD 99 to USD 499 for a certificate and paywalls the logo licence.
Free entry is table stakes now. Free proof is the wedge.
No cash prize, by design, and that matches the field: no benchmarked nomination-and-rubric award pays a purse.
Block 4 · How it is judged
Six weighted criteria, published before the call opens.
The first three carry 60 percent. The rubric rewards durable change and real adoption over model novelty, which is the design position, stated in the rules rather than left to each juror.
Two tiers
Regional first, then a global round that starts from zero.
Tier 1 · 6 Nov to 11 Dec 2026Regional
Three macro-regions: Americas, EMEA, Asia-Pacific. Mixed panels of about seven, ambassadors alongside external experts.
Each region promotes up to three entries that clear a published cut-off.
The cut-off is a floor of quality, not a seat count. A region promotes nobody if nobody clears it.
Tier 2 · 11 Jan to 5 Feb 2027Global
About six to nine finalists pitch live to the global jury, which then re-scores every finalist from zero under the same rubric.
Regional scores do not carry forward. A strong region cannot buy an advantage in the final round.
Why an outsider should trust the result
Governance is cheap, and it is what separates an award from a mill.
- Conflicts are structural, not polite. Automatic decline on a canonical match, human review on a near match, per-entry attestation, disclose and recuse
- No contact with judges once scoring opens, and scores are never altered quietly
- Claims versus evidence is surfaced at the moment of scoring: an unsupported assertion scores in the bottom band
- AI-written entries are barred, which for an AI award is not irony but necessity
- An entry needs at least three jurors before its aggregate is valid, and ties break on a published deterministic rule
The differentiator
Every finalist leaves with written, criteria-anchored feedback. Not a result. A read on their own practice.
- It is affordable because the raw material already exists: every juror records a strength and an improvement area per criterion at scoring time. Nobody is asked to write anything twice.
- It is private, finalists only, and it is never a communications asset.
- The publishable output is the Story of Success case study, under consent recorded separately from the entry.
This is what turns a once-a-year ceremony into a pipeline of community case studies that feeds the webinar cadence between editions.
Block 5 · Calendar, state, decision
The edition, end to end.
Where the build actually is
The software is done. The account configuration is not.
Built, tested, committedThe platform
- Public entry form, first gate, questionnaire, evidence upload with type and size enforcement
- Draft and resume by magic link, five consent gates captured as timestamped records
- Admin screening behind Cloudflare Access, with the organizer conflict wall
- GDPR erasure reaching the stored files, retention purge, audit export
- The award page, the operating pack, and the public downloads
Not yet configuredThe production account
- The transactional email path. Without it the magic link cannot send, so an entrant cannot resume a draft
- The real bot-protection keys (a test key is in place)
- Both Cloudflare Access applications, and the named administrators
- The award page wired into pmairevolution.com
- Ricardo's production account, not yet started
The decision we need today
Which Cloudflare account is production for edition 1?
This has exactly two branches, because once an entrant submits, the account holding their data is fixed for the whole edition. There is no DNS cutover of a live edition.
Ricardo's account
The originally planned path. Requires provisioning, DNS, two Access applications, and an authenticated sending domain, all standing up and smoke-tested by 28 August.
Vitor's account is production
pmairevolution.com points at it from the start. No entrant data ever crosses accounts. The structural handoff happens between editions, against an empty database.
Ricardo is unreachable 30 August to 7 September, and nominations open on the 8th. That leaves six business days. Deferring this choice selects Option A by omission, which is the riskier of the two.
What we need from you
Four answers, and the week can run itself.
- Confirm the production accountDue 29 August. Everything else in the infrastructure track waits on it
- Confirm the jury baseline: seven seated plus three reserves per region, with a floor of five that is never waivedIt sets the panel publication gate
- Name the two or three administrators for the maker-checker actionsPromotion and designation require a second pair of eyes by design
- Approve publishing the call for judges on 8 SeptemberJury formation runs August to October, so the call cannot wait for the judge portal to be finished
What exists today, and where it lives
The pack is complete and ready to read.
The operating packEight documents
Operating rules (the source of truth), nomination form, evidence-pack guide, judge guide, scoring-sheet specification, feedback-report template, ceremony outline, communications plan.
Plus the launch award page, its four public downloads, and the privacy notice.
How it stays honestEverything derives from one document
Every document, the page, and the graphic derive from the operating rules. A change goes into the rules first and propagates outward, and an automated check reads the finished PDFs to catch any that did not follow.
The four previously open decisions are closed, the amendments are recorded and numbered, and the pack has been rebuilt from the current rules.
Who built this
The award core team, and what each of them changed.
Every line above is a recorded amendment, not a courtesy. The review changed the rules you are being asked to approve, and each change is numbered in Section 13.8 with the reasoning that produced it.
Thank you
Ready when
you are.
Nominations open 8 September 2026. The ceremony is 17 February 2027.
The one thing to leave with: the production-account decision, by 29 August