Problem statements live from the briefing
The accelerator runs on eleven numbered problem statements, seven of them captured at the orientation. Each is transcribed here with its sponsor and the pain they described, then read against the rail: who in the ecosystem answers it, and what Rakyat ONE says.
Which three to rank
The form takes three preferences and a maximum of two are assigned. Ranked on what the orientation actually established: the moat, the sponsor, whether we can demonstrate it without any bank data, and whether the cohort in this room can beat us to it.
Credit Assessment Report and CCPT ScoringHasliza Hasan, SVP Commercial and SME Credit
Bank Negara published the decision logic, nobody automates it anywhere in the world, and it is the one statement where having no bank data is an advantage rather than a handicap.
CCPT is a published rule set, not a model, and we have now read the primary documents rather than the slide. The taxonomy was issued in April 2021 with five guiding principles and a classification table that maps five categories onto three bands: climate supporting, transitioning, and watchlist. Prohibited activity is a hard gate applied first. The due-diligence questions, at version two since January 2024 and reissued unchanged in the July 2026 refresh, give five assessment categories, a yes-or-no risk question in each, remediation questions triggered by every yes, and one crisp aggregation rule: a client must have commenced at least one remedial measure in every category where harm was found, so remediating three of four failures is still a failure. Six sector guidance notes exist for small enterprises. That is a rules engine the regulator wrote for us, and it is the One Line exactly: the engine decides, the AI only extracts and explains. One nuance to get right, because a credit officer will test it: the taxonomy states in its own footnotes that holding a certification is not by itself sufficient to pass, so a tool that treats a certificate as a pass is wrong on the regulator's own terms.
Of the cohort we have identified, Finory does extraction with no scoring, CTOS is bureau and registry data, OR Technologies is pipelines, Reclimate is carbon and not a credit vendor, and Codebase is a core platform. Nobody here has a CCPT capability. The global claim survived a deliberate attempt to break it, but phrase it precisely, because a judge may have seen a headline: no vendor we could find automates CCPT classification for small-enterprise bank lending. A bond pricing agency here sells an ESG classification service for sukuk issuers that one outlet labelled CCPT, though the agency's own materials do not use the taxonomy's language, and in Europe a credit-intelligence vendor already automates the EU taxonomy in exactly this shape. The category is proven viable elsewhere; this jurisdiction's version is unbuilt. The training market tells the same story: the consultancies and the banking school sell CCPT courses, which is what a market does when there is no software.
The strongest demo needs no bank data at all, because the regulator published its own answer key. The taxonomy document carries five fully worked use cases with their correct classifications, and the July 2026 guidance notes add more, including palm oil, forestry and hazardous waste. We run the tool across those and show it reproducing the regulator's own published verdicts, which is a far better claim than any score on data nobody can check. Then the due-diligence questions walked live, category by category, each answer citing the exact evidence sentence. Then a deliberately thin case where the system refuses to classify, reports its confidence and escalates to a credit officer.
Three subsystems in five weeks, and the internal credit system's field structure is private. Ask the sponsor for the field template and one anonymised report at the first mentoring session. Be careful about one thing in the pitch: the real difficulty is evidence extraction, not the classification logic, because the rules themselves permit declarations, proxies and industry benchmarks where hard data is missing. And do not promise a public Malaysian dataset of small-enterprise accounts, because it does not exist in usable form. Listed-company reports are the wrong population and registry filings are sold as scanned images.
Fintech Driven Business FinancingMohd Azhar Md Aziz, SVP Business Intelligence and Analytics
Our existing engine ports across almost whole, into the most precisely quantified pain on the list, and the factors slide asks in the bank's own words for a white-label interface for third-party partners. The discount is the crowd, and this room is the crowd.
Micro financing is manual end to end across about fifty branches with inconsistent quality, six to ten days of turnaround, up to a hundred and eighty days of statements read by hand, and four to five hours to process a single day of work. An SVP owns it. Our conversational intake, deterministic scoring, operator console and API move across nearly unchanged, and the four-hour target rewards whoever scores off data rather than collecting documents.
This room, and the market behind it. MADCash raised about five million ringgit pre-Series A, Direct Lending has disbursed over ninety million, and beyond the cohort sit platforms with billions disbursed at default rates under one per cent. Finory has pivoted to selling exactly the statement-parsing layer we would otherwise build, which makes it a buy-or-build question rather than a rival. We will not win on funding or on loan book. We win on the clock, because the fastest verified approval in this market is forty-eight hours and the sponsor asked for four, and on an explainable engine that matches the best published Malaysian benchmark while staying auditable enough for a Shariah committee.
Same spine. Documents in, extraction to structured data, consistency and arithmetic verification, deterministic scoring. PS 001 adds the industry narrative and the taxonomy engine on top; PS 006 adds onboarding and cash-flow scoring. The factors slides make the pairing tighter still: PS 001 asks for integration with the existing credit assessment systems and PS 006 asks for an API orchestration layer with a white-label front, so both want the same plumbing. One build, two statements, two SVP sponsors, one of them carrying a revenue mandate.
Real-Time Application TrackingMortgage Financing Business Unit with the Fintech Team
The safest build on the list and a verified market gap, but it proves the least about us to a panel that includes venture investors.
Seven days to answer a status enquiry, systems that exist but are not integrated, and nobody able to say where a file is stuck. No Malaysian bank exposes a shared real-time portal to solicitors, valuers and developers. It needs no core access: an agreed milestone schema, invite links for third parties to self-report, notifications, and a stuck-file view.
Multi-role dashboards and notifications are our bread and butter, which is precisely why they demonstrate no moat. Against a rubric that rewards defensibility, and judges who include venture investors, this reads as competent delivery rather than something only we can do. Rank it as insurance in case both stronger choices are oversubscribed.
Rank PS 001 first, PS 006 second, PS 002 third. Plan the build for the likely pair, PS 001 and PS 006, as one engine with two heads.
Shared spine: documents in, language model extraction to structured data, consistency and arithmetic verification, then a deterministic scoring engine. Head A adds the industry-risk narrative and the taxonomy rules engine, and produces a credit assessment report with a confidence-scored classification and a full audit trail. Head B adds electronic onboarding and behavioural cash-flow scoring, and produces the end-to-end micro financing funnel with an embeddable interface. The cross-document triangulation that PS 003 asks for becomes a feature of the shared verification step rather than a separate bid.
One build, two statements, two SVP sponsors, and the same sentence holding it together throughout: the deterministic engine makes every financial determination, reproducibly and auditably, and the AI only extracts, explains and converses.
Why this order and not the reverse, since PS 006 now looks stronger. Two statements are assigned, so ranking PS 001 first does not cost us PS 006; we expect to be given both. What the order changes is which one we lead on and which mentor pairing we get, and there PS 001 wins on moat: it is the statement almost nobody else will pick and the one no competitor can close in five weeks. There is one risk in the choice worth stating plainly. PS 006 is the most contested statement in the cohort, so if it is rationed the team that ranked it first may take it ahead of us. We accept that risk, because losing PS 006 still leaves us the uncontested first choice and an uncontested third, whereas winning PS 006 as our lead would put us in the bloodiest field with the weakest relative position, no loan book and no funding to match theirs.
- PS 001 was wrongly declined. Reading only the slide, its language (assessment engine, natural language processing, confidence scoring) looked like a generic document-AI contest that specialist vendors would win. The opposite is true. Because the regulator published the decision logic, the defensible core is deterministic, which is our strongest ground and not theirs.
- PS 007 was wrongly promoted. It was ranked highly because the existing suite already fits it. That is our convenience, not the bank's value, and the presenter conceded that demand for those digital tools is low today.
- PS 002 was over-ranked. Buildability is not a moat. It stays on the form as insurance, not as the lead.
- The decisive fact that was missing: the bank supplies no data during the programme. That removes PS 005 entirely, weakens PS 003, and turns PS 001 into the one statement where bringing your own data is an advantage, since it demonstrates on public annual reports.
Genuine regional whitespace and the deterministic part suits us, but pixel forensics has a credibility ceiling with a bank and no forged-document corpus is available. Its best trick, cross-document triangulation, is absorbed into the shared verification layer anyway.
No public EPF interface, and open finance obligations do not begin until 2027. Two of the four fields the sponsor asked for cannot be answered without an employer network that does not exist here. An honest version reads as faster paperwork.
Behavioural signals need embedding inside the bank's own app with its mobile team, which no external team can obtain before the semi-finals. The alert-triage reframe answers a different question, and no alert data will be provided.
Easy to build, low ceiling, crowded, and the sponsor conceded low demand. Malaysian winners in this space are goal-anchored micro-saving with real yield, not game layers, and the bank already runs a merchant platform.
Four statements were not presented and are still to come from the organisers' emailed deck. Read them before submitting and re-rank only if one is clearly stronger than the first choice. Everything else on this page assumes the pair above.
The rail, layer by layer
Rakyat ONE's seven surfaces on top. Underneath, every layer a bank product needs, with the player from the room who fills it. Empty slots are the gaps we build or buy.
Slot map
Surface by surface: which partner becomes a product on it, which feeds it data or risk signals, and which rail it settles on.
Partner cards
The integration in one line each: what plugs in, what flows across, how the money works, and the ask.
Second Chance, with partners on every rung
The accelerator submission is a rejection-recovery ladder. Read with the room in it, each rung gains a partner, PS 003 and PS 004 gain a home at the intake, and PS 006 is the same ladder for merchants.
Sequence
Who we talk to, in what order, and what each conversation has to produce before demo day.
- PayNet · Lee Kok Yang. Sandbox access for DuitNow QR and JomPAY under the Community Hub MoU.Produces: the rail our QR and merchant surfaces settle on, with the hub's blessing.
- MADCash · Nuraizah Shamsul Baharin. A microfinance surface on Rakyat ONE Bisnes and a Rung 3 referral partner.Produces: the Islamic microfinance proof the bank's mandate is written for.
- CTOS · Sharvind Varma. Bureau score and CCRIS-adjacent data into Second Chance intake and Bisnes financing.Produces: verified DSR inputs, which is what Rung 1 and PS 003 both need.
- Moby · Rian Philip. Mobylize matching inside Rung 1.Produces: a declined applicant leaves with a product, not a letter.
- Level Five · Frederick Chung. NOTO 360 signals into Command's mule watch and the Individual portal's scam screen.Produces: the fraud layer PS 003's EFMS complement sits beside.
- Finory · Kee Hui Jiang. Statement parsing for the member PFM and for the payslip / EPF / bank-statement triangulation.Produces: the data engine under our PS 003 response.
- Direct Lending · Hui Yik Seong. Embedded channel for Rung 2 graduates the bank still declines.Produces: a second exit from the ladder, with takaful attached.
- OR Technologies · Ahmad Hadzramin. Lakehouse and pipelines under Command.Produces: the analytics floor the console reports from.
- ToyyibPay. Reseller or acquiring partner for Rakyat Gateway, if the bank prefers to buy reach.
- Dobin. Merchant offers on the card, once volumes exist.
- Reclimate. A sustainability-linked financing product on the portal.
- Vespid. Literacy content inside Rung 3 coaching.
- Netsec. Penetration test of the six surfaces before any live pilot.
- Codebase. The bank's core decision, not ours. We integrate either way.
How to read this. Player facts come from the public-profile research on the roster page; the placements, integration lines, commercial models and priorities are UNIVA's proposal, not agreements. Nobody here has been contacted yet. Problem statements are transcribed from the screen at the orientation briefing on 8 September 2026 and should be checked against the bank's written brief when it is issued. To add a problem statement, append to the PS array in the page script; to move a partner, change its priority in ECO.