A proven Touchless AP pattern, engineered into the Client's enterprise.
How BetterBrain would staff and deliver the agentic finance platform and Touchless AP MVP: starting from an approach we've already proven on AWS, then doing the integration, controls and scale work that makes it production-grade for a 1,200-person finance organization.
Proven patternTouchless AP already demonstrated end to end, so the design risk is retired up front
AWS-nativeBuilt and co-sold with AWS; deploys into the Client's own AWS account
Two pathsAn AWS-backed POC first, or a pitch demo and a Client-funded build
One teamPortage leads strategy and finance; BetterBrain leads architecture and engineering
01 · What the Client is asking for
One governed platform, with AP as the proof.
The Client wants a finance-wide agentic platform on AWS, with Touchless AP as its first live workflow: invoices go from receipt to ERP posting with no human touch, and only real exceptions reach the AP team. Every later finance agent (revenue, close, FP&A) runs on the same foundation.
The platform
Orchestration, agent registry, tool connectors to Snowflake and the ERP, an immutable audit ledger and an evaluation harness. Built once, reused by every agent.
The AP MVP
Five agents (intake, validation, matching, exceptions, approval and posting) live in production for an agreed cohort of entities and vendors.
The handover
Controls Internal Audit accepts, a trained AP team, and a platform the Client owns and can extend to the next finance agents.
02 · Our Touchless AP solution
Proof the approach works, before the Client invests.
We built our invoice automation solution in partnership with AWS, and AWS is already taking it to its own customers. It shows the Client the target experience on day one and retires the design risk. It is a reference implementation, not the Client's production system: the sections that follow cover what it takes to get there.
- ReceiveInvoices arrive by email to an AP inbox, or from Drive, SharePoint or S3.
- Classify & extractEach document is classified and its fields extracted using vision and language models.
- ReviewExtracted fields are shown for validation and correction in the document dashboard.
- Trigger rulesClient-defined rules (e.g. vendor plus amount over a threshold) pick the workflow to run.
- Duplicate checkEach invoice is checked against prior invoices before it moves on.
- 3-way matchThe invoice is matched to its purchase order and goods receipt.
- GL codingGL codes are tagged automatically.
- Budget controlOut-of-budget departmental spend is blocked.
- ApprovalRouted for approval; AP is notified and the invoice is filed in SharePoint.
- Terms & postingEarly-payment discount terms are monitored, and the invoice posts to the ERP.
Watch the full walkthrough of the solution running end to end.
Touchless AP demo video →
| Client's MVP agent | Proven in our demo | Enterprise build for the Client |
|---|---|---|
| Invoice intake & extraction | Email intake, classification, field extraction, human review | Supplier portal and EDI intake; header and line extraction across every invoice format in the pilot cohort; accuracy tuned and proven on 12 months of the Client's history |
| Validation & compliance | Field validation, duplicate check, budget controls | Live checks against vendor master, tax rules and reference data in Snowflake and the ERP, across entities that change month to month |
| Matching | 3-way match on a single entity | 2- and 3-way match with tolerances by entity, product and vendor class, tested against real historical exceptions |
| Exception resolution & routing | Rule-based routing to a reviewer | An agent that explains each exception, recommends a resolution and routes it to the right AP queue, with every override captured for audit |
| Approval & ERP posting | Single approval step, posting via API | The Client's full delegation of authority matrix, segregation of duties, and governed write-back to the production ERP |
03 · Where the work is
A demo processes an invoice. Production processes the Client's.
Every entity, vendor, exception and audit question at enterprise volume. That is where an engagement like this succeeds or stalls, and where our team spends its time.
Enterprise integration
Connectors to the Client's ERP, Snowflake, supplier portal and EDI, built inside their AWS account and through their InfoSec and change processes. Typically the largest share of the build.
Controls Internal Audit signs off on
Delegation of authority, segregation of duties, an immutable audit ledger and override capture with reason codes, designed with Internal Audit rather than added at the end.
Accuracy at scale
Golden datasets built from the Client's own history, regression gates on every change, and touchless rate raised cohort by cohort without letting errors through to payment.
A platform, not a point solution
The orchestration engine, agent registry, agent standard and tool library every future finance agent runs on. This is what makes revenue, close and FP&A agents faster and cheaper later.
Security on the Client's terms
Agent identity, tool-level permissions, encryption and private networking on AWS, with no Client data leaving their environment.
Ownership and handover
Production-grade code, runbooks and knowledge transfer so the Client's team can run and extend the platform after hypercare.
04 · Architecture fit
Their AWS blueprint, built to production standard.
Portage's proposed architecture (Bedrock, AgentCore, Snowflake, Claude) is the stack we already deploy on. Here is where proven patterns de-risk each layer, and what gets engineered specifically for the Client.
User experiencePortal, approvals, alerts
Document review dashboardAP work queues and exception reviewTeams / Slack approvals
OrchestrationWorkflows and approval steps
Configurable workflow engineRule triggers and human approval stepsDeployed on AgentCore Runtime / Step Functions
Agents & toolsThe work itself
Extraction (vision + language models)Matching and duplicate toolsVendor, PO and tax validationAgent registry and agent standard
ModelsRight model per step
Model routing to cut cost per invoiceClaude via Bedrock
Data & integrationSystems of record
SharePoint, Drive, S3, email intakeSnowflake reference dataClient ERP connector and EDI
EvaluationAccuracy you can prove
Golden datasets and regression gatesTouchless-rate and accuracy reporting
Governance & auditWhat auditors see
Source attribution on every decisionImmutable audit ledger (S3 Object Lock)Segregation of duties and override capture
DeploymentWhere it runs
Deploys into the client's own AWS accountNo client data leaves their environmentCode handover and knowledge transfer
Proven pattern we bring
Engineered to the Client's systems and rules
Built new for the Client
05 · Two ways to deliver
With AWS or without. We're set up for either.
Both paths end in the same place: Touchless AP live in production on the Client's own platform, delivered by the same team. The difference is how the Client gets its first proof point and who funds it.
Option A
AWS-backed POC, then production build
Prove it on the Client's own invoices first, with AWS funding most of that proof.
1 · Proof of concept~4 weeks
Primarily AWSClient
- Our AP solution deployed in a Client AWS sandbox
- Run on historical invoices, POs and receipts for a pilot cohort
- Touchless rate, accuracy and cost per invoice measured against today's baseline
2 · Production build~16 weeks + hypercare
Some AWS fundingClient
- Portage's full plan: platform, all five agents, ERP integration, controls, go-live and handover
- Starts from a scope and architecture already validated in the POC
Best whenthe Client wants proof on its own data before committing, the budget needs a steering-committee proof point, or AWS is a strategic relationship for them.
Trade-offAWS funding approval adds some lead time and a third party to coordinate before the POC starts.
Option B
Pitch demo, then a Client-funded build
No AWS involvement. Show the Client what's possible in the pitch, then go straight into the full engagement.
1 · Tailored pitch demo~1–2 weeks, pre-sale
No cost to the Client
- Our existing AP solution, configured on synthetic data shaped like the Client's ERP, vendors and invoice types
- Shown live in the joint pitch
2 · Full project~16 weeks + hypercare
Client-funded
- Portage's full plan as proposed
- First proof on real data comes inside the engagement, at the Week 8 checkpoint
Best whenthe Client is ready to commit and wants speed, prefers fewer parties in the contract, or has cloud commitments that make AWS funding less relevant.
Trade-offthe Client funds the whole program, and its first look at results on real invoices comes after the build has started.
Either waySame team, same architecture, same end state. Portage picks the path that fits the Client's buying process, and we can move from Option B to Option A if AWS funding becomes useful later.
06 · If we bring in AWS (Option A)
AWS has a stake in this going live.
The platform runs on Bedrock, AgentCore and the Client's AWS footprint, which is exactly the kind of production AI workload AWS funds partners to land.
Existing relationship
Built with AWS
We built our invoice automation solution in partnership with AWS, and AWS is already selling it into its own clients.
POC funding
The 4-week POC
We'd bring in AWS's generative AI proof-of-concept funding to cover most of the POC, so the Client validates value with little up-front spend.
Production funding
The production build
AWS also funds generative AI projects moving to production. We'd apply it to part of the build, with the Client funding the rest.
Funding amounts depend on AWS program approval for this opportunity. We'd register it with AWS and engage the Client's AWS account team as soon as Portage is ready to bring them in.
BetterBrain
Breadth across the stack. The right people, when it matters.
Two build workstreams running in parallel, matching Portage's plan: one builds the agentic platform, the other builds the Touchless AP agents on top of it. One architect across both, with specialists we bring in as each part of the project needs them. Senior throughout.
Leadership & architecture
Abhishek Bhargava
Engagement lead · AI architect
CMU computer science & computational finance; ex-YC, commodities trading. Design authority for the platform and agents — your main point of contact.
Michael Boyerflex
CIO advisor · IT governance
Enterprise IT, security and compliance; InfoSec review and Internal Audit walkthroughs.
Alex Broganflex
Commercial & finance strategy
Ex-Goldman IB — finance initiatives such as cost-containment tooling and the AP value case.
Workstream 1 · Agentic platform
MS
Mohit Shah
Lead platform engineer · full-stack
Orchestration engine, agent registry, audit ledger and shared services in the Client's AWS environment.
+
Platform & integration engineer
Forward-deployed eng · data & ERP
Connectors to Snowflake, the ERP, supplier portal and EDI; ERP posting integration.
Ilona Litvinova
Quality & evaluation eng
Evaluation harness, golden datasets, regression gates and touchless-rate reporting for both workstreams.
Workstream 2 · Touchless AP agents
+
AI engineer · agents
Forward-deployed eng · agent development
Builds and tunes the intake, validation, matching, exception and posting agents on the Client's invoices.
DK
Dima Kyrychuk
Forward-deployed eng · app & UX
AP portal, work queues, exception review and approval screens.
+
BetterBrain benchflex
Additional engineers on demand
More forward-deployed engineers join if volume, entities or integrations expand the scope.
Design authority · both workstreams
Abhishek
Accuracy & evaluation · both workstreams
Ilona + Mohit
Security & controls
Abhishek + Michael
BetterBrain · Prepared for Portage Point
One engagement, many disciplines — we bring in depth wherever the project needs it.
How the team maps to Portage's plan
Portage Point
Strategy & finance
- Executive and program leadership
- Finance transformation lead
- AP process and controls SMEs
- AI operating model and governance
- Training and change management
BetterBrain
Architecture & engineering
- AI architect and design authority
- Platform, data and integration engineering
- AP agent development
- Evaluation, QA and controls testing support
- Portal and UX, hypercare and support
| Phase | BetterBrain team | Focus |
|---|---|---|
| Phase 1 · POC | Abhishek (architect), Mohit, Ilona, AI engineer | Deploy the solution, run the historical invoice test, co-design the architecture and agent standard with Portage |
| Phase 2 · Production build | Workstream 1 · Platform: Mohit, platform & integration engineer, Ilona Workstream 2 · AP agents: AI engineer, Dima Abhishek across both; Michael for security and controls | Orchestration, audit ledger, evaluation harness and ERP integration in parallel with all five agents, the AP portal, UAT and controls testing |
| Go-live & hypercare | Mohit plus one support engineer; Abhishek on escalation | Monitoring, fixes, knowledge transfer and handover to the Client's team |
Final team size is set after the Week 0 workshop confirms scope. Starting from a proven pattern means the team's time goes into the Client-specific work (integration, controls, accuracy at scale) rather than reinventing the basics.
BetterBrain
Meet the experts. Built to deploy, not just advise.
Research-grade AI depth meets operators who've shipped in production — across finance, robotics, enterprise IT and energy. We're backed by leading funds, and by individual investors and advisors from the very companies building the models and data platforms this program will run on.
Academia & research
Enterprise & finance
Y Combinator
Industry & operations
Backed by
Funds
SAMSUNGNEXT
hustle fund
Individual angels & advisors — people who work at
Snowflake
Individuals investing & advising in a personal capacity — not corporate investments by these companies.
Core capabilities
StrategyRoadmap & prioritization · use-case discovery · eval design · governance & risk.
ImplementationDocument intelligence · workflow orchestration · context-aware agents · ERP & data integration.
Industry expertiseFinance & accounting · legal & compliance · manufacturing & logistics · energy & infrastructure.
BetterBrain · Prepared for Portage Point
Academic depth and production scars — the same team that builds the foundation ships it in your environment.
Relevant work
Invoice reconciliation
Live reconciliation of invoice records against ledger data for a large enterprise in Asia.
Audit-grade AI
A source-attribution platform tracing every AI answer to its source document, built for a global bank's AI governance.
Finance data agents
Text-to-SQL agents cutting analyst time on data requests by 80%, in production for finance and operations teams.
09 · Next steps
A joint pitch the Client can see working.
- Confirm scope inputs. Pilot cohort, invoice volume, ERP and sandbox access, AWS stack preferences and code-handover requirements.
- Tailor the demo. We re-run our AP solution on synthetic data shaped like the Client's ERP, vendors and invoice types for the pitch.
- Choose the path. Decide with Portage whether to lead with an AWS-backed POC (Option A) or go straight to a Client-funded build (Option B). For Option A, we register the opportunity with AWS and line up funding before the Client meeting.
- Pitch together. Portage leads the finance and transformation story; BetterBrain shows the solution running and walks through the architecture.