SatFinAI AdvisorySathish Susai

Process transformation & operational excellence

Fix the process first. Then automate it.

Automate a broken workflow and all you have built is a faster way of being wrong.

I turn manual, high-cost, ungoverned finance and operations processes into measured, governed, automated systems — and I bring the numbers. $57M+ in measurable business impact. $43M+ recovered. Ten transformation programmes across logistics, shipping, information management and regulated financial services.

Start a conversation See the method

One approval process at a large US-based financial-services organisation. Seventeen weeks of sequential review, measured before anything was changed, then rebuilt into three risk-tiered parallel paths.

Baseline — single sequential queue, seven reviewer functions17 weeks
High risk — full seven-function concurrent review13–14 weeks
Medium risk — three-function parallel review10–15 days
Low risk — cleared with no reviewer intervention3–5 days

Up to 97% cycle reduction on the low-risk path. Roughly 89% on the medium path. No control loosened — every routing rule maps to GLBA, NIST AI RMF, ISO 27001 and SOC 2.

$57M+Measurable business impact
$43M+Cash recovered
10Transformation programmes
65+Custom AI assistants in production
15+Years, seven countries of coverage
14Performance awards, plus a chief executive's MVP appreciation

Career aggregates. Every engagement figure further down is measured against a baseline that existed before the work started, and none of them is a projection.

The method

Three methods. The order is the method.

The work gets described as two things — process, and AI. It is three. Most consultants sell one of them. The middle one is where the deepest measured reductions sit, and it is the one that cannot be bought off a shelf.

Method one

Operational Excellence

The fix is the deliverable. DMAIC to the real root cause, operating-model and target-operating-model design, business process re-engineering, and KPI and governance frameworks built from nothing.

$43M+ recovered · disputes down 60% · DSO down up to 40% · aging improved up to 50% · bad debt down 95% · a 26-member global team built and led across 48 divisions

Method two

Operational Excellence into AI

The process is diagnosed and corrected first, then automated onto the corrected rails. Nothing here was automated before it was fixed, and that sequence is the whole differentiator.

Output: an AI-ready process

Approval cycle down up to 97% against a 17-week baseline · capacity created up to 50% · productivity up 25–40% · reporting preparation cut 75% · collector throughput roughly doubled

Method three

AI and Automation

The build is the deliverable, where the process is already sound. Automation without governance is only risk at scale, so every build ships with SOPs, routing logic and compliance designed in.

65+ custom AI assistants live · eight Microsoft Copilot Studio agents deployed across both ends of a regulated approval process · three disconnected systems into one hourly-synced view · customer outreach up 70%

The pure-AI lane is the smallest of the three. That is correct and it is not padded — most of the technology work was preceded by process work, and a thin pure-AI panel beside a full hybrid panel is the more honest claim.

What method two produces

AI readiness

An AI-ready process is one that can be handed to any automation, any platform or any new system and run without being redesigned first.

Most organisations tell me AI is a next-year conversation. That is usually the correct assessment, and it is the engagement rather than the obstacle. Getting a process to this state is operational-excellence work. It finishes before a single line of automation is built, it returns on its own, and it is the reason whatever comes afterwards works at all.

A process that cannot be described in rules cannot be automated by anyone — not by AI, not by an ERP migration, not by the next vendor. Getting it there is the work. What comes after is the easy part.

1

Documented

One current-state map, one future-state map, and the tribal knowledge written down where a system can read it.

4 BPMN process maps · 7 standard operating procedures, 14 to 17 checklist items each

2

Decision-ruled

Every branch expressed as an explicit rule with stated conditions, rather than as reviewer judgement.

12 tier rules and 25 routing conditions in a live decision matrix

3

Clean at the input

Data fields defined, owned and validated where they are entered, because automation inherits whatever it is fed.

425 data fields analysed across 4,583 historical submissions

4

Measured

A real baseline exists, so anything built afterwards has something to be scored against.

A 17-week end-to-end baseline established before any redesign

5

Controlled

A control plan and a named owner, so the process survives the system change instead of dying with it.

Control plans, RACI and tollgate governance on every programme · a receivables process carried intact across two organisations in a handover

Three disconnected systems went into one synced view at Iron Mountain, and a receivables process moved between two organisations intact at Hewlett Packard Enterprise and DHL. In both cases the process was the thing that survived, because it had been written down properly first.

Measured against a prior baseline

What the three methods actually delivered

Sorted by method, coloured by method. Percentages are improvements against the state of the process before the work started. Named clients are closed engagements; the current client appears under its sector.

Operational excellence

Bad debt reduced95%
Maersk — aged North America portfolio
Reporting preparation removed75%
XPO — 60+ minutes daily down to roughly 15
Pricing and billing disputes cut60%
XPO — $21M dispute portfolio
Aging improvedup to 50%
XPO — $50M transition portfolio
Collections increased45%
DHL Global Forwarding — Canada receivables
Days sales outstanding reducedup to 40%
XPO — $50M transition portfolio
Unapplied cash reduced40%
DHL Global Forwarding — Canada receivables

Operational excellence into AI

Approval cycle reduced, low-risk pathup to 97%
A large US-based financial-services organisation — against a 17-week baseline
Approval cycle reduced, medium-risk pathroughly 89%
A large US-based financial-services organisation
Collector throughputroughly doubled
XPO — 25–30 accounts a day to 50–70
Capacity createdup to 50%
Iron Mountain — segmentation and operating-model redesign
Productivity gain25–40%
XPO — generative AI, Excel VBA and HighRadius

AI and automation

Unallocated payments cleared100%
Maersk — reconciliation redesign, 1,200+ customers moved to a digital portal
Credit cleanup on aged accounts98%
Maersk
Customer outreach improved70%
Iron Mountain — voice and conversational bots
65+Custom AI assistants in live production
8Copilot Studio agents deployed — seven reviewer-function, one requester-facing
3 → 1Systems consolidated into one hourly-synced view
4,583Submissions analysed across 425 data fields

Selected work

Four engagements, and what each one actually proved

A large US-based financial-services organisation · via Softnotions Technologies · 2026

AI and technology intake governance, rebuilt

Every AI and technology request ran through a single sequential approval queue. Seven review teams — process excellence, AI, third-party governance, risk, enterprise architecture, legal and sourcing — each waited on the one before it. End-to-end cycle time averaged seventeen weeks. No service levels, no parallel paths, no routing intelligence, and no way for a requester to see where a submission stood. Everything downstream waited on it.

A full DMAIC cycle, evidence first: 4,583 historical submissions analysed across 425 data fields, voice-of-customer from seven stakeholder shadow sessions coded into sixteen themes, and gap analysis isolating ten missing and ten wasteful steps. Only then a redesign.

Before

One queue. Seven functions. Each waiting on the last.

Process excellence› AI review› Third-party governance› Risk› Enterprise architecture› Legal› Sourcing End to end — 17 weeks

After

Three risk-tiered paths, governed by a decision matrix of 12 tier rules and 25 conditions across 24 request types.

Low risk — 3 to 5 daysCleared with no reviewer intervention
Medium risk — 10 to 15 daysThree-function parallel review
High risk — 13 to 14 weeksFull seven-function concurrent review

Seven function-specific SOPs, four BPMN 2.0 process maps and the routing matrix were authored inside four weeks, mapped to thirty-six regulatory and policy sources including GLBA, NIST AI RMF, ISO 27001 and SOC 2 Type II. Then the automation: eight agents on Microsoft Copilot Studio — seven reviewer-function agents, one per team, each embedding that team's procedure, checklist logic and routing rules, plus a requester-facing agent sitting with the business owner raising the request. Two-directional, upstream at submission where information quality is set and downstream inside review where the decision is made.

Governance and speed are usually traded against each other. They do not have to be. The bottleneck was almost never the review itself — it was what happened before a request ever reached a reviewer.

No dollar value is attached to this engagement. The process outcomes above were measured. The financial projections were modelled and are not published.

XPO · 2023–2025

Cross-border transition of a $50M portfolio

A $50M third-party logistics portfolio moving from the United States to India, with no playbook. Aging was deteriorating, DSO was high, and pricing and billing disputes were eating the collector capacity that should have been collecting. A sub-portfolio was piloted first to prove the method and earn the trust before asking for scale.

  • $25M cleared from the $50M portfolio — a 50% recovery rate — inside twelve months
  • DSO down up to 40%, aging improved up to 50%
  • Disputes cut 60% on a $21M portfolio; $4M of bad debt recovered and $10M of revenue leakage identified and corrected
  • Productivity up 25–40%, reporting preparation cut 75%, collector throughput roughly doubled, twelve planned hires avoided
  • MVP appreciation from the chief executive, Great Place to Work Champion, six collections contest wins

The dispute backlog was not a collections problem. It was a pricing data problem wearing a collections costume — narrowed from tariff type, to specific lanes, to zip codes, then closed with a correction policy deployed globally.

Iron Mountain · via Tescra · 2025–2026

AI adoption across a $900M receivables operation

A 236-person collections organisation working a $900M portfolio faster than its analyst capacity allowed, across billing, pricing, collections, payments and disputes sitting in three disconnected systems.

  • 65+ custom AI assistants designed on a structured prompt architecture, accelerating data analysis, dashboards, business requirements, SIPOCs, charters and executive decks
  • A collections orchestrator integrating Oracle Cloud, Salesforce and Billtrust into one hourly-synced pool — one current view instead of three stale ones
  • Customer outreach up 70% through voice and conversational bots, guided by a standardised global call framework
  • Capacity created up to 50% through segmentation and operating-model redesign, with an efficiency gain of 50 positions across the 236
  • Seven Black Belts trained, who then ran their own automation projects

Governance is not the brake on an AI programme. It is the thing that lets it reach production at all.

Inchcape Shipping Services · 2022

Global credit control across 48 divisions

A global receivables book spread across 48 divisions with no KPI framework and a post-COVID cruise portfolio carrying real bankruptcy exposure. A 26-member global team was built and led, and a regional coordinator model created where none existed.

  • $10M cleared from a $30M small and medium enterprise portfolio — a 33% recovery rate
  • $10M of disputes cleared on a $20M cruise key-accounts book — disputes more than two years old, closed in 90 days, across 2,500+ invoices reviewed one at a time
  • The first KPI framework the function had, plus ADKAR change management across the divisions

The collectors hired and coached into that model were still running it three years later. That is the part of a transformation that does not show up in the first-year number.

Bought in its own right

I also train the people who run it

Seven Black Belts at Iron Mountain were trained on prompt architecture, agent creation and AI-augmented process excellence. They then ran their own automation and improvement projects with it. Alongside that: 70+ professionals coached across the United States and India, 60+ hours of structured training delivered, and ADKAR change management run at 70+ person scale across two geographies.

Taught by someone who has built sixty-five of them in production, not by someone who has read about them.

Half day to two days

AI for leaders

What to fund and what to refuse. Where AI programmes actually die — almost never in the technology, almost always in governance, data quality and adoption. What to demand from a vendor before signing, and how to tell a pilot that will scale from one that will not.

For executives, function heads, transformation sponsors and boards. Licensed by IBM Generative AI for Executives & Business Leaders.

Three to five days, cohort up to fifteen

AI-augmented process excellence

Prompt architecture as an engineering discipline. Assistant and agent design built on top of DMAIC rather than beside it. Where AI genuinely accelerates a Six Sigma cycle, and where it produces confident nonsense. Building an AI-ready process, hands on.

For Green and Black Belts, continuous-improvement teams, process excellence functions and shared-services leads.

Four to eight weeks, embedded

AI adoption and change enablement

ADKAR applied to AI adoption specifically. Measuring adoption rather than licences issued. Building the internal capability that outlasts the engagement, and handing over so the client's own people run it.

For an organisation mid-adoption with the tools bought and nobody using them.

The most under-priced part of a transformation is the part that keeps running after the consultant leaves. Deliverables depreciate from the day they are handed over. Trained people do not.

Fifteen years

Where the numbers came from

PeriodRole and organisationWhat it delivered
2026 — present Lean Six Sigma Black Belt Consultant, Operations Excellence Enablement
Softnotions Technologies · client: a large US-based financial-services organisation		 Intake governance rebuilt. 17 weeks to 3–5 days on the low-risk path. 7 SOPs, 4 BPMN maps, a 25-condition routing matrix, 8 Copilot Studio agents deployed.
2025 — 2026 Black Belt Manager, AI & Order-to-Cash Transformation
Tescra · client: Iron Mountain		 65+ AI assistants live. Collections orchestrator across Oracle Cloud, Salesforce and Billtrust. Capacity up to 50%. Outreach up 70%. Seven Black Belts trained.
2023 — 2025 Senior Specialist, Order-to-Cash Transformation
XPO		 $25M of a $50M portfolio recovered. DSO down up to 40%. Disputes down 60%. $10M of revenue leakage identified. MVP appreciation from the chief executive.
2022 Global Credit Control Supervisor, Centre of Excellence
Inchcape Shipping Services		 $10M of a $30M portfolio recovered across 48 divisions. Multi-year cruise disputes cleared in 90 days. A 26-member global team built and led.
2019 — 2022 Analyst to Senior Analyst, Lean & Agile Transformation
Maersk Global Service Centres		 $8M of a $15M aged portfolio recovered. Eight consecutive quarters at or above target. Bad debt down 95%. Unallocated payments 100% cleared.
2018 — 2019 Process Associate, Order-to-Cash
DHL Global Forwarding		 Collections up 45% and unapplied cash down 40% on the Canada book. Two salary increases inside one year.
2015 — 2018 Financial Associate, Finance Operations
Hewlett Packard Enterprise		 Canada receivables, air and ocean. The process transitioned directly to DHL with unbroken ownership.
2011 — 2014 Finance Consultant and owner
YES Advisory		 A licensed sole practice. 100+ clients, 95%+ conversion in face-to-face meetings.
2007 — 2011 Research Scholar
Loyola College		 Design of experiments, regression and hypothesis testing. A peer-reviewed publication now carrying 143 citations.

Credentials

What licenses the work

  • Lean Six Sigma Black Belt · Certified AI-Powered Lean Six Sigma Implementation ExpertSix Sigma Academy Amsterdam, April 2026. The second one certifies Six Sigma and AI implemented together, which is what method two is.
  • Generative AI for Executives & Business LeadersIBM, April 2026
  • Lean Management & Manufacturing Expert · Risk Management ExpertSix Sigma Academy Amsterdam, April 2026
  • Google Prompting EssentialsGoogle, April 2026
  • Essentials Automation CertificationAutomation Anywhere, April 2026
  • Master of ScienceBharathidasan University, 2007

Working with me

Based
Pune, India. Remote-first, global delivery.
Hours
US East Coast hours, and other global time zones based on project need.
On site
US B1/B2 visa valid to October 2034, for on-site sprints when they matter.
Languages
Tamil, native. English, full professional proficiency.
Sectors
Financial services · logistics and transportation · shipping and maritime · business services · technology

How this starts

Four ways in

Each one is fixed-scope and fixed-duration, and each ends with something you keep whether or not the next phase happens.

AI readiness assessment

Two to three weeks

Your process measured against the five conditions, with the gaps costed and sequenced. You get a documented, decision-ruled process and a baseline — which is what any automation decision later needs anyway.

Operational excellence opportunity assessment

Two to three weeks

Where the process is losing time, cash and capacity, ranked by what is recoverable and how fast. DMAIC, root-cause analysis, and a prioritised set of fixes rather than a list of observations.

AI governance and intake readiness assessment

Three to four weeks

For a regulated environment adopting AI. Routing, controls and approval design mapped to recognised frameworks, so governance stops being the reason nothing reaches production.

Cash and receivables recovery diagnostic

Three to four weeks

Portfolio segmented, root causes of leakage and dispute isolated, recovery sequenced. The proving ground for everything above, and still the fastest way to fund the rest of the programme.

Profit does not leak because people underperform. It leaks because the process lets it.

If you are carrying a manual, high-cost or ungoverned finance or operations process, send me the shape of it. Thirty minutes is usually enough to tell you where the time is going and whether it is worth a programme.

Email sathish@satfinai.com