I want to confess something.

Back in 2019, before Docsumo, I built a spreadsheet I was kind of proud of.

It tracked our entire sales pipeline. Lead source, deal velocity, all colour-coded with conditional formatting that would make any data analyst weep with joy.

I showed it to everyone. “Look how automated this is!”

One day the sales ops person sent me this meme lol!!

I realised nobody else could use it. Including me, after about three months of not touching it.

Last week, I saw a survey that stopped me cold. 58% of finance leaders said Excel is their primary automation tool.

I laughed. Then I felt bad for laughing. Because I’ve been that person.

Why Does Nobody Admit This About Spreadsheets..???

You start with something simple. It can be tracking quarterly invoices.

Then the next quarter hits. Instead of rethinking your approach, you tack on another sheet.

Leadership wants trend analysis. Fine, you add a pivot table tab.

Your boss needs vendor risk metrics. Time to dust off those VLOOKUP skills.

Half a year passes, and suddenly you’re babysitting a multi-tab creature that lives in your head. And you’re burning entire days each quarter just to keep the thing breathing.

The Cost You’re Ignoring.

On average, AP teams capture just 58% of available early payment discounts. The other 42% is gone.

MONEY LEFT SITTING THERE!!

If you’re processing invoices worth ten million monthly. If 30% of suppliers offer 2% early payment discounts, that’s $60,000 in potential monthly savings. Over a year, that’s $720,000 in savings just sitting there.

Doesn’t it blow your mind that most teams only grab about half of it?

That’s $360,000 annually that just... disappears. Because by the time your team finishes entering data and routing it for approvals, the discount window has closed.

Teams with automation capture around 85% to 95% of available discounts.

I spoke with a finance director whose AP team dedicates over thirty hours weekly to manually entering invoice information into their system.

She told me, “My people are Excel wizards. They’re just not getting to apply their intelligence to anything that matters.”

When I asked why they hadn’t automated, she said they’d looked into it eighteen months ago. Too complicated. Their documents were “too messy.”

Meanwhile, her best analyst just resigned. Burned out from data entry.

The Comfortable Lie

“Excel does what we need.”

Does it really?

I’ve noticed a pattern: when people say their Excel workflow is fine, they typically mean “it hasn’t exploded catastrophically yet.”

They’re not counting the discounts that expired or the vendor who’s now demanding faster payment because they’ve been consistently late. The contract renewals that happened on autopilot were because nobody had time to review the terms.

What Happens When You Actually Solve It

Recently, I talked to a manufacturing CFO processing 2,500 supplier invoices monthly. Their AP team was four people spending 25 hours weekly on data entry.

There were constant vendor complaints about late payments.

After automation, the data entry time dropped to 3 hours weekly. Discount capture jumped to 92%. One major supplier even offered better terms because payments were now consistently early (pretty cool perk imo).

Three Questions Worth Asking Yourself

Open whatever spreadsheet runs your most important process right now. Your monthly close. Your reporting system. Whatever keeps your operation running.

Consider these:

If your spreadsheet expert disappeared tomorrow, could anyone else execute their work?

If you discovered an error in a key formula today, what decisions from recent months would need revisiting?

If any of those made you shift uncomfortably in your chair, you already have your answer.

Look, I understand completely. Something urgent always demands attention. Why disrupt a process that technically works?

What if Excel could be what you thought it was?

Rather than being a data entry prison, it functions as the analytical tool it was designed to be. I think Claude for Excel comes really close. It lives inside your spreadsheet, traces formula dependencies, and debugs errors in seconds.

Excel should be for analysis. The invoice processing, the document chaos, that’s what automation like Docsumo handles. So your team can use Excel for thinking, not typing. Over a quarter of finance operations use zero automation tools.

Remember my sales pipeline masterpiece? I’m simply pointing out that there’s a meaningful difference between using a powerful tool and being trapped by it.

Until next week,
Rushabh

P.S. If you’re thinking, “but my situation is different because...” I want to hear your story. Reply to this and walk me through it. What’s functioning well? What’s falling apart? I personally read and respond to most messages I receive.

True story from this week.

A Fortune 500 vendor sent us an invoice. Here’s the journey it took to reach my inbox:

PDF → printed → faxed → photographed → emailed as JPEG.

Five formats. One document.

I called them. “Why are you doing this?”

Nobody knew. “That’s just our process.”

Before I tell you why this one invoice matters, here are three things I came across recently that made me realise this isn’t a quirky vendor issue. It’s a billion-dollar problem:

• The document processing market has hit $7.89 billion this year. It’s expected to reach $66 billion by 2032.

• UiPath (the automation company) just got named a Leader specifically for handling “high-complexity document workflows.

• Datamatics became one of the first worldwide to get ISO certified for AI document systems. They patented a way to handle unstructured documents.

Anyway, back to my story. 80% of your company’s data is in formats like the invoice I got, and most companies can use only about 20% of it.

So Here’s What I See

Last month, a CFO told me his AP team forwards the same invoice 4.7 times before processing. He tracked it for a week.

Here’s the journey:

9:00 AM - Generic AP email
9:45 AM - Dallas forwards to Mumbai
2:30 PM IST - Printed for signature
4:00 PM IST - Scanned, sent to ERP
Next day - Wrong GL code (scan cut off digit)
Two weeks later - Vendor calls, nobody finds it

This costs 20-30% of operational expenses before counting missed discounts or duplicates.

Small companies have messy docs because they’re scrappy. Big companies have 200 vendors, five ERPs, and nobody remembers why it works this way. You’re processing 5,000 invoices from 200 vendors in 47 formats.

Every Time I Talk To Someone About This, They Say The Same Two Things

“We’re looking into automation”, which means someone saw a demo 18 months ago and decided it was too complex.

“Our documents are too messy for automation”, which is backwards. Messy documents are exactly why you need it.

Enterprise ops teams manually process unstructured data for 80% of their work hours. Four days a week, typing numbers from PDFs into systems.

What we’ve built at Docsumo handles the mess. Handwritten notes. Faxed documents. Any photo-of-a-printout situation from Monday. We’ve trained our models on 20+ million documents specifically so they work when formats are all over the place.

The companies getting this right aren’t asking “can we automate?” They’re asking “what should our team actually be doing?”

What Changes When You Fix It. FOR REAL

National Debt Relief processes debt settlement letters - which vary “a lot” and maintains 95%+ accuracy. They’re saving 2,500+ hours yearly. That’s human attention redirected from “is this a 6 or an 8?” to “does this application make sense?”

Biagi Bros handles 11,000+ bills of lading monthly. The system processes. Humans review exceptions.

Arbor settles 75,000+ insurance claims with 99% accuracy on ACORD forms. If you know ACORD forms, you know there are about 47 variations, and they’re all slightly different.

None of these are clean operations. They’re dealing with real chaos that doesn’t disappear. The bottleneck does.

Look, Here’s What This Comes Down To

Your AP team shouldn’t spend 32 hours weekly typing invoice data. They should spend 8 hours on exceptions and 24 on analysing vendor terms, catching discounts, negotiating contracts.

We’re focused on the space between receiving a document and making a decision from it. Getting the right docs to the right people at the right time - processed, validated, flagged if something’s off.

One Last Thing About That Invoice

That photo-of-a-fax invoice from Monday? When I called them, nobody knew why they did it that way. They just did.

Sometimes the system’s broken because nobody questioned it in ten years.
Rushabh

P.S. If you’re dealing with this and want to talk through what’s possible, just reply. Genuinely curious what patterns you’re seeing.

Someone built a 47-tab spreadsheet and I can’t stop thinking about it:

Last week, I was on a sales call with a VP at a mid-sized insurance firm.

He shared his screen to show me their quarterly reporting process.

47 tabs.

Each one color-coded. Cross-referenced. Dependent on data from at least three other tabs.

“I know it looks complicated,” he said, “but once you understand the logic, it makes perfect sense.”

That’s exactly what everyone says right before their spreadsheet breaks.

I’ve heard this line maybe fifty times now. And every single time, the same pattern plays out.

How We Got Here

Here’s how it always starts.

You need to track Q1 invoices. 10 rows. A few formulas. Done in 20 minutes.

Then Q2 arrives. You add another tab instead of rethinking the structure.

Someone needs year-over-year comparisons. Another tab with a pivot table.

By Q4, you’re maintaining a 47-tab monster that only you understand.

And you’re spending 8-10 hours every quarter just keeping it alive.

Research analyzing 15,000 Excel spreadsheets found that 24% contained formula errors. Another survey revealed 33% of large UK businesses reported poor decision-making due to spreadsheet problems.

JP Morgan lost $6 billion partly because someone divided by sum instead of average in a spreadsheet.

TransAlta spent $24 million on what their CEO called “literally a cut-and-paste error.”

These companies had smart people.
Good processes.

The documentation chaos just outgrew its purpose.

How does this happen?

That VP I mentioned?

He spends 8-10 hours quarterly maintaining his spreadsheet.
Updating references.
Fixing broken links.

That’s not counting his team’s time feeding data into it.

TLDR: he’s become the “spreadsheet person.”

You know the type.

The one person who built the system. The one everyone’s terrified will quit or go on vacation.

Research shows 94% of financial spreadsheets contain errors.

Last month, I worked with an accounting team processing 3,000+ bank statements monthly across 50+ layouts.

Five people spent their entire day extracting data from PDFs and entering it into spreadsheets.

By the time it was 6 pm, they were only halfway done.

Over 6,000 hours per month on just moving data from one place to another.

When they automated the document workflow - processing speed went up by 10x and error rate dropped to nearly zero; ensuring the organization makes accurate accounting decisions.

It costs real money

Those quarterly maintenance hours compound.

The institutional knowledge locked in one person’s head.

The 3 AM panic when formulas break before a board meeting.

And here’s the part that haunts me: when 94% of spreadsheets contain errors, you’re probably working with bad data right now.

You just don’t know where the mistakes are.

Think about the decisions you made last quarter.

Was your data clean?

How would your choices change if you discovered a formula error tomorrow?

My two cents

I’m not telling anyone to abandon spreadsheets.

They’re brilliant for quick calculations, scenario modeling, and analysis.

But there’s a difference between using a spreadsheet and being held hostage by one.

I think there’s a need to remove the need for them to be the single source of truth.

Let documents flow into systems automatically.

Let data be validated at entry.

Let spreadsheets be where humans add judgment, not where they spend hours copying and pasting.

The tragedy is that all that intelligence is trapped in a system requiring constant manual feeding to stay alive.

Until next week,
Rushabh

P.S. If you’re thinking “wait, my spreadsheet only has 43 tabs,” I’d love to hear about your setup. What’s working? What’s breaking? Hit reply and let me know. I read every response.

I’ve been thinking about a demo call from last week.

Not because of the technology discussion, but because of what happened about 20 minutes in.

Before I tell you that story, here are three data points that suddenly made a lot more sense:

• 76% of workers spend 1-3 hours daily just moving data from one place to another.

• Manual data entry costs U.S. companies $28,500 per employee annually.

• 56% of employees experience burnout due to repetitive tasks.

Coming back to the demo call. And context about these stats…

Emily is a debt specialist at a financial services company.

On a demo call, she walked me through her workflow:

1. Open the settlement letter PDF.
2. Switch to the spreadsheet.
3. Paste into Column A.
4. Back to PDF.
5. Switch.
6. Paste.

About 10 minutes into watching this, she stopped mid-sentence.

Her voice had changed (she was exhausted).

She processes about 200 settlement letters a week.
Each one takes 5-7 minutes of manual data entry.
That’s 17 hours a week, almost half her job, just copying and pasting.

“I’m actually good at this,” she said. “I’ve gotten fast.
But I have a finance degree. I didn’t think this was what I’d be doing.”

We built Docsumo to solve exactly this problem.

After Emily’s call, I started seeing the pattern everywhere.
The loan processor with carpal tunnel from scrolling through bank statements.
The property manager who had nightmares about utility bills.
The AP specialist who knew every invoice template by heart.

They’d all built coping mechanisms.

Colour-coded tabs.
Highlighting systems.
“Focus music” playlists.
One person timed how many documents they could process before their coffee got cold (that’s insane!!).

Leadership often views “document processing” as a line item.
They rarely see the exhaustion and the people who’d stopped mentioning their job at parties.

What most people do not tell you about manual document processing:

It wastes both people and time.

Emily is smart. Finance degree. She understood debt restructuring, cash flow analysis, and negotiation strategy.

She should have been analysing portfolios and making strategic decisions.

Instead: Cell A1, Cell B1, Cell C1. Repeat 200 times. Go home. Do it tomorrow.

When we automated her workflow, 5-minute documents became 40-second extractions.

First week: The team kept checking the data. How could it be this fast?

Second week: She called me. Laughing this time.

“I processed last month’s backlog in 90 minutes. So I actually read through the settlement terms. Found three accounts where we could negotiate better. Called the clients. They were shocked I had time.”

That’s the thing. When you give people their time back, they don’t just work faster. They start working like humans again.

What matters is what happens after.

The loan officer is approving applications in 10 minutes instead of 60.
The property manager spends time on tenants instead of utility bill entry.
The debt specialist was using her brain for the job she was hired to do.

How many people in your organisation are one bad day away from crying over a spreadsheet?

You probably don’t know. They’ve gotten good at hiding it.

If someone on your team is drowning in document chaos, copying, pasting, validating, and reformatting, that’s not a productivity issue. It’s a human one.

— Rushabh

P.S. We turn document chaos into clean data automatically. For efficiency metrics and the people doing the work. If your team is buried in manual data entry, let’s talk.

P.P.S. Emily got promoted, by the way. Turns out, when you have time to think strategically, people notice.

Last Tuesday, I got an email from Bikram, my co-founder.

Subject: “Quick sync?” He wanted to discuss an “urgent vendor payment” before EOD.

Looked legit, except Bikram never says “vendor,” he says “supplier.” And he doesn’t email, he walks over to my desk.

I walked over. He had no idea what I was talking about.

If I’d been careless, I would have become a statistic.

I run a document AI company. We help businesses turn messy documents into clean data automatically. If I almost fell for it, what’s happening to finance teams handling hundreds of requests daily?

I polled our team: “Have you received a spoofed email from leadership in the last 6 months?”

34 out of 47 said yes. That’s 72%. Eleven people (23%) admitted they almost acted on it.

One colleague showed me what they’d received:

Email phishing is just the entry point.

And the next step isn’t another email. It’s documents: invoices, statements, contracts that look real, reference real names, and even match real company formats.

The Moment Fraud Crossed a Threshold

A UK fintech caught something their manual review missed. 15 applications, 3 days, perfect bank statements. Traditional verification cleared everyone.

An intelligent system noticed that income patterns were statistically improbable for the business types (the numbers didn’t add up). Transaction sequences followed identical timing across all fifteen.

All 15 were fraud. Prevented losses: over €200,000.

The issue? Most companies are still verifying documents the old way:

Modern document fraud doesn’t fail format checks. The complete financial picture needs to make sense, and most verification systems don’t look at the bigger picture.

So What’s Working?

The companies that aren’t getting wrecked? They stopped trying to manually verify every document. They started asking completely different questions:

A European fintech firm now processes loan applications in minutes instead of days. Their fraud detection got smarter because they shifted from document-by-document verification to cross-document intelligence.

That’s the same lens we apply at Docsumo every single day.

Documents are the surface.

Patterns are the truth.

I almost fell for a spoofed email from my co-founder, and we build AI for document intelligence. So if it’s happening here, it’s happening everywhere.

Fraud isn’t sneaking in through fancy hacks.
It’s sneaking in through perfect paperwork.

What are you seeing in your organisation? Hit reply. After writing this, I’m genuinely curious what patterns others are catching.

Rushabh

P.S. My team obsesses about turning messy documents into clean, structured data that operations teams can actually trust. If you’re rethinking document workflows, let’s talk.

P.P.S. Know someone dealing with document chaos? Forward this their way.

Hey folks,

I was looking at a company called Vertikal last week.

They’ve been in the insurance verification business since 2004.

Their platform CertFocus helps construction companies and property managers track contractor insurance.

Basic idea: before you let a contractor on your job site, you need proof they have insurance. If they cause damage and aren’t insured, you’re liable.

Vertikal processes about 10,000 of these insurance documents every month.

Each one took 45 minutes.
Do the math.
That’s 7,500 hours of manual work monthly.

What The Documents Look Like

The insurance industry has standard forms for this. They’re called certificates of insurance. Two main ones: ACORD 25 and ACORD 28.

Standard format. Name of contractor. Name of insurance company. Policy number. Coverage amount. Expiration date.

Sounds simple to automate. Just read the form and extract the data.

Except it’s not that simple.

Why Computers Struggle With These

The structured parts work fine. The policy number is always in the same box. Expiration date is always in the same place. Software handles that.

The problem is in the description sections.

Let me show you what I mean. A simple entry looks like this:

“General Liability: $2,000,000”

Easy. The contractor has $2 million in general liability coverage.

But real entries look like this:

“General Liability: $2,000,000 per occurrence, $4,000,000 aggregate. Additional insured status applies to ABC Construction per endorsement CG 20 10. Coverage excludes professional services. 30-day cancellation notice required. Waiver of subrogation in favor of certificate holder when required by written contract.”

Now the computer needs to understand:

Every insurance company writes these differently. Every state has different requirements. Every contract has different conditions.

The computer can read the words. But it doesn’t know what matters.

Does “per occurrence” mean something different from “per claim”? Yes. Is “CG 20 10” different from “CG 20 26”? Yes, critically different. Does the order of conditions matter? Sometimes.

This is why Vertikal’s software was only getting it right 30-40% of the time. It could read text. It couldn’t understand insurance requirements.

What Happens At Scale

At 1,000 documents a month, you can handle this manually. Hire a few people who understand insurance. They review each one. Slow but accurate.

At 10,000 documents a month, you can’t keep up.

A construction manager needs to verify a contractor’s insurance before they start work. Documents go to the outsourced team. Team reviews them. Sends results back three days later.

Meanwhile, the contractor is waiting. The job is delayed. Or worse, you let them start work before verification comes back.

Competitors were offering same-day verification. Three days wasn’t competitive.

The Headcount Trap

Most companies hit this problem and add more people. Hire a bigger team. Move more work offshore. Add quality control layers.

You end up spending more money, but you don’t solve the speed problem. You just make the slow process more expensive.

Vertikal realized they had a technology problem, not a staffing problem.

What Changed

But here’s what mattered more than the metrics.

Construction managers could now verify insurance before letting contractors on site, not after. Property managers could check vendor coverage in real time instead of waiting days.

The product promise changed. Used to be “we track your insurance certificates.” Became “we verify compliance instantly.”

Same business. Different capability. Because the operational bottleneck was gone.

Download Free Ebook

The Pattern

I see this in a lot of businesses.

You build something valuable. Clients love it. Volume grows. Operations become the constraint.

You can’t promise real-time anything when your backend takes days instead of minutes.

The first response is usually to add people. More processors. Bigger offshore team. Project managers to coordinate with the processors.

But you’re not solving the problem. You’re making it more expensive.

The companies that win are the ones who recognize when an operational limitation is actually a product limitation.

Three Questions

Vertikal had maybe a few years before competitors would have caught up. Maybe less. When clients can see your operational constraints, the clock is ticking.

What I’m Watching

This pattern shows up everywhere documents are involved.

Expense reports. Loan applications. Medical records. Legal contracts.

Any time you need to extract structured information from documents that weren’t designed for computers.

The technology for this is getting better. Not just reading text, but understanding what it means in context. What’s important. What’s an exception. What makes one version different from another.

Insurance has been dealing with this for decades. But the underlying problem is universal.

The winners aren’t the companies with the most people reviewing documents. They’re the ones who saw that operational bottlenecks define what your product can be.

Talk soon,
Rushabh

P.S. If you’re dealing with document processing at scale, I’m curious, what’s working for you? Hit reply.

Welcome back, folks!

This week, I have an interesting story of a UK payroll company that discovered its biggest competitor was a single government form that cost thousands in penalties. I’ll show you how they turned that nightmare into their acquisition advantage.

But first, here’s some intel you cannot miss out on:

• UK employers face £100 penalties per 50 employees for each month P11D forms are filed late, with additional penalties up to £3,000 per incorrect form

• Class 1A National Insurance on employee benefits rose from 13.8% to 15% in April 2025

• Labor costs represent 70.4% of total business expenses, making payroll accuracy business-critical

When Promise Depends on Government Forms

Pento is a London-based payroll automation company serving over 400 UK businesses. Their pitch was straightforward: we’ll eliminate your payroll headaches. You run your business, we’ll handle the complexity and ensure compliance.

They delivered brilliantly, customers loved them, and the business was growing. They were doing well enough to catch HiBob’s attention, a major HR platform player.

But there was a problem that showed up like clockwork every July. P11D forms were creating compliance time bombs.

The Impossible Form

P11D forms are deceptively simple in concept but brutal in execution. They report employee benefits and expenses to HMRC, company cars, private healthcare, accommodation, and loans. Every UK employer providing benefits must file them by July 6th annually.

For Pento’s customers, these were material financial hits that destroyed trust in payroll software.

P11D forms are completely unstructured. Unlike payslips with predictable formats, P11D forms vary wildly. Benefits data is scattered across multiple pages. Different employers organise information differently. Handwritten amendments. Supporting documentation attachments (the nightmare list is endless if you ask me!).

Processing them manually meant days of work per client. Missing details meant penalties. Getting data wrong meant bigger penalties. You can see the cycle of problems here!

Why Standard Solutions Failed

You’d think this would be a solved problem by now, right? Payroll software exists specifically for this kind of thing.

But traditional document processing couldn’t handle P11D’s chaos. These systems needed templates, predictable layouts, and data in consistent positions; basically, everything P11D forms didn’t offer.

Most platforms required manual intervention where someone had to review multi-page forms, consolidate information, and then validate everything before submission.

Which, let’s be honest, defeated the entire purpose. Pento’s customers weren’t paying for software that required intensive manual work during the year’s most critical compliance deadline.

The Strategic Decision

Learning P11D Logic: The system studied how P11D forms work as business documents, understanding that benefit values, employee details, and tax calculations follow business logic regardless of where they appear.

Handling Multi-Page Chaos: Instead of requiring templates, the platform learned to identify and consolidate relevant data automatically across multiple pages.

Eliminating Manual Review: The intelligence layer caught inconsistencies and validated calculations before submission, removing the manual intervention competitors required.

Download Free eBook

The Results!

Pento achieved 98.71% accuracy across P11D processing. More importantly, they eliminated tax filing penalties for customers (pretty cool, right?).

Not reduced. Eliminated.

During filing season, while competitors’ customers scrambled with manual reviews and faced potential penalties, Pento’s clients filed confidently. No penalties. No compliance anxiety. No last-minute panic.

This operational certainty became a competitive differentiator. When HiBob acquired Pento in February 2024, they were acquiring proven compliance infrastructure that worked when it mattered most.

What This Really Means

Pento’s story illustrates a crucial principle: in compliance-driven businesses, operational excellence isn’t optional but the product itself.

Your Next Move

In businesses where every form counts, operational excellence is what keeps you competitive.

Stop hoping manual work will save you. Stop relying on “good enough” automation. Start thinking of your processes as part of the product. Do it now, or risk losing trust (and money) later.

Docsumo’s document AI exists to handle the messy, unstructured chaos so you can focus on delivering real value.

PS: Pssst… here’s a secret, the forms that scare everyone? They don’t stand a chance against Docsumo’s AI.

PPS: 98.71% accuracy. ZERO customer penalties. Real results from a company that recognised the strategic importance of getting compliance right.

Welcome back!

This week, we’re diving into a silent leak in banking. The billions lost not to fraud, but to customers who quit before they even open an account.

But first, let’s see what concerns every financial institution this week:

• Up to 40% of UK customers abandon onboarding that’s too slow or complex.

• Over $6.7B wasted every year on incomplete financial onboarding.

• In 2023 alone, regulators fined institutions $6.6B for KYC, AML, and sanctions lapses.

• The identity verification market will double by 2030, from $14.3B to $29.3B.

Source (State of Intelligent Automation Report | November 2022)

The Customer Who Never Was

Sarah downloaded your banking app on Monday morning. She needed a business account for her new consulting firm.

Your marketing team spent $127 to acquire her as a lead. She made it through the welcome screen, entered her email, and hit “Get Started.”

Then your KYC process began.

Your competitor onboarded her in 12 minutes with automated verification.

Her lifetime value? $15,000+. Gone.

The business account with expected monthly transaction volume of $50,000? Never materialized.

When Manual Checks Slow Everything Down

Sarah’s application lands in your compliance queue. She’s #394 of 847.

Your KYC officers are manually checking ID formats, cross-referencing databases, and verifying signatures.

Each file takes hours.

And it’s not your team’s fault, they’re drowning in PDFs, screenshots, and half-legible photos.

87% of banks say poor data management and siloed processes drive clients away.

The Compliance Catch-22

You can’t go faster; fines are brutal.

In 2024, TD Bank paid $3.09B for AML failures, one of the largest in history.

“Move too fast and risk fraud. Move too slow and lose customers.”

Manual KYC creates a false choice: speed or accuracy.

Neither side wins.

When Eight Documents Become Too Many

What did Sarah actually need to upload?

Each in a different format, language, or quality.

Every mismatch, blur, or missing stamp sends her backward.

68% of applicants abandoned onboarding because it simply took too long.

When Mistakes Get Expensive

Ironically, it’s not just real customers who suffer. Fraudsters win too.

Manual reviewers miss things, especially on their 200th application of the day.

Fraudulent accounts slip through, leading to:

• Direct financial losses

• AML/KYC penalties

• Brand damage

• Costly investigations

AI-powered systems, by contrast, detect forgeries and inconsistencies far more accurately and instantly.

How Document Intelligence Changes Everything

The smartest banks aren’t just digitizing old steps; they’re also rethinking them.

AI-based document processing can:

• Verify IDs and proofs in seconds

• Handle any file format (photo, scan, PDF)

• Detect tampering or fakes automatically

• Cross-check info across documents

Case in point:

Building a Better KYC Flow

Manual KYC is creating three big problems:

• Customers quit when onboarding drags on.

• Regulators are tightening oversight and handing out billion-dollar fines.

• Competitors are already experimenting with a host of document automation platforms for faster verification.

Transforming KYC isn’t about buying software. It’s about re-engineering your workflow. Hiring more people won’t fix this. The real shift happens when document checks move from manual review to intelligent automation.

Winning teams do this:

• Accept documents from any source: app, email, or upload.

• Build validation logic that knows a valid driver’s license from a fake one.

• Create clear workflows for exceptions and edge cases.

• Keep audit trails for regulators.

Your Next Move

At Docsumo, we’ve helped banks reach 95%+ accuracy on complex IDs by treating verification as a semantic task, not just text extraction.

The system “understands” what a passport means, not just what it says.
Modern platforms can verify IDs in seconds, check multiple formats, and scale as your business grows. They flag suspicious cases for human review while clearing genuine ones instantly.

Two-thirds of digital users abandoned a banking application last year.
Fintech challengers are winning customers by making onboarding take minutes, not days.

Every delay means another lost lead.
Customers compare you to the fastest experience they’ve had anywhere, not to other banks.

“The best product doesn’t always win. The easiest one does.”

Every delay costs customers. The institutions moving fastest on KYC automation are already winning that market share.

If document checks are slowing you down, Docsumo can help you move faster, with fewer errors and happier customers.

Welcome back!

This week, I’m sharing the story of a UK bills management company that discovered its growth ceiling wasn’t market competition. It was hidden in thousands of water bills sitting on their operations team’s desks.

But first, the critical intel you need this week:

• Manual utility bill processing costs businesses an average of 45 minutes per bill, with error rates climbing as volume increases - creating bottlenecks in service delivery

• The utility billing software market reached $6.13 billion and is projected to hit $14.07 billion by 2032, growing at 12.07% CAGR.

• UK property management firms are under unprecedented pressure in 2025, with rising operational costs and evolving legislation forcing them to streamline every process or risk profitability

When Your Service Model Becomes Your Constraint

Launched in 2015, Huddle aimed to handle household utilities so landlords don’t have to. Huddle grew into one of the UK’s leading bill management companies.

But by the time they hit scale, Huddle had developed a problem that threatened everything they’d built.

The Daily Grind

Each bill required identifying the provider, finding usage charges buried in different sections, identifying environmental fees and wastewater costs, and then manually entering everything.

Simple bills with straightforward usage? Maybe 10-15 minutes. Complex multi-property accounts with seasonal variations and adjustments? 30-45 minutes per bill.

Tenants questioned delays. Landlords grew frustrated. Every processing delay chipped away at Huddle’s reputation.

The Multi-Provider Reality

Every water company structures their bills differently. Rate structures differ based on property type and location.

For Huddle, managing properties across the UK meant dealing with an ever-expanding universe of bill formats.

Traditional document processing relies on pattern recognition. Huddle couldn’t build templates for every water bill variation. Even if they could, water companies update systems periodically, breaking existing templates.

They can pull “£143.67” from a bill, but can’t validate whether that indicates a potential leak.

The Turning Point

Huddle’s leadership demonstrated forward thinking that separates sustainable businesses from those that hit scaling walls.

Understanding Billing Patterns: It learned to recognise usage data whether it appeared in detailed tables, summary paragraphs, or mixed formats.

Adapting to Provider Variations: Instead of creating rigid templates for each water company, the system learned to identify the same business information across presentations.

Validating in Context: The platform cross-checked extracted data against Huddle’s property records and historical patterns. Inconsistent charge structures were caught before reaching customers.

The system doesn’t care where the usage data is. What matters is understanding that certain pieces of information represent meter readings and others represent calculated charges.

Read the full case study here!

The Results That Matter

After implementing intelligent document processing, Huddle achieved 95.46% accuracy across their 12,000+ annual water bill volume. Processing that previously consumed days of operator time now happens in a fraction of the time with minimal manual intervention.

Customer experience improved measurably. Billing turnaround times dropped. Error complaints declined. Tenants got accurate bills faster. Landlords gained confidence that utility management was being handled genuinely.

But most importantly, document processing stopped being a constraint on growth. Huddle can now onboard new properties without worrying whether their water provider’s bill format will create operational chaos.

Your Next Move

If you’re running a property management business, utility service, or any operation that processes high volumes of varied bills, Huddle’s story should resonate.

The technology exists. Huddle has proven it works across format variations. The only question is timing - will you address the constraint before or after it becomes a crisis?

PS: If water bills, energy statements, or any utility documents are eating your operations team’s time, Docsumo can help.

PPS: Every number here is real. 12,000+ bills. 95.46% accuracy. Two operators freed for higher-value work. True results from a company that chose to solve its problem rather than accept it.

Welcome back, folks!

This week, we’re exploring why insurance compliance has become the silent deal-killer in commercial real estate, and how the smartest operators can turn document chaos into an advantage.

But first, the critical intel you need this week:

• Commercial property insurance premiums jumped between 5-25% , with carriers reporting commercial price increases of 6.6% consecutive quarters.

• ACORD reports that digital reinsurance transactions via the ADEPT platform have surged by ~73% in 2025 and have been >300% since 2023.

• ACORD has provided standardised forms since 1971, yet 90% of agencies still struggle with format variations that break traditional processing systems

This week, I want to share a story about a company that was drowning in a documentation nightmare every real estate firm faces, but found a way to turn its operational headache into its strongest moat.

The Problem Hidden in Plain Sight

Meet Arbor Realty Trust, a New York commercial real estate powerhouse. Arbor processed 6,000+ insurance applications monthly while managing 75,000+ yearly claim settlements. This includes certificates of insurance and specialised flood certificates for waterfront properties, all feeding into their loan origination process. Each demanded meticulous documentation review and data extraction.

The Scale Trap

Their operations team faced a familiar nightmare:

The brutal workflow: identify document types, extract vendor and policy details, cross-reference coverage against property requirements, validate against loan covenants, route through approvals, then update multiple databases.

Time per document ranged from 15 minutes for simple certificates to over an hour for complex submissions. Error rates climbed faster than their hiring ability.

Why Traditional Solutions Failed

The fundamental challenge was that templates looked similar. (spoiler: THEY WERE NOT!)

The ACORD Paradox

ACORD forms have represented standardisation since 1971. But insurance carriers and property managers interpret “standard” differently, adding custom fields, changing layouts for legacy systems, and using tools that create variations.

The Template Recognition Trap

Traditional document processing systems excel in controlled environments with predictable formats.

Production environments tell a different story. When real insurance certificates hit these systems, with handwritten notes, varying table structures, and overlapping e-signatures, accuracy rates crash from the 90%+ demo results to 60-65% in actual operations.

The Strategic Breakthrough

Arbor’s leadership demonstrated strategic thinking that separates industry leaders from followers. Instead of accepting bottlenecks as an operational reality or hiring more staff, they recognised this required solving the core intelligence challenge.

They understood that scaling needed semantic understanding of business relationships, systems that could understand insurance logic regardless of presentation format.

The Solution Architecture

Learning Arbor’s Business Logic: The system studied a few sample documents from Arbor to understand their requirements and compliance mandates. Whether processing certificates for waterfront properties or standard certificates of insurance for office buildings, it learned what coverage looked like for each asset class and could flag compliance issues.

Handling Any Format: Now, whether insurance data appeared in scanned broker communications or carrier-specific custom formats, the Document AI platform flagged compliance issues before they reached underwriters, avoiding costly delays. The intelligence layer recognised the business relationships that mattered for operations.

Reducing Manual Checks: When extracting coverage limits from any certificate of insurance, the platform automatically validates against all regulatory mandates, significantly reducing the manual checks needed in the process.

Read the detailed case Study here!

The Transformation Results

Processing dropped from hours to under 5 minutes per document with 99% accuracy. They achieved 96% faster processing, handling 6,000+ monthly applications across multiple formats.

Most impressive: 95% straight-through processing - 95 of 100 certificates flowed directly into databases without human intervention.

The Strategic Lesson

Arbor’s story shows a core principle about operational excellence in document-heavy industries: the companies that win, preempt when traditional approaches hit their limits.

What This Means for Your Strategy

When you’re managing insurance compliance across hundreds of properties with dozens of carriers, success depends on building intelligence that adapts to business requirements.

The competitive advantage will always belong to those who solve problems their competitors consider unsolvable. Semantic document intelligence can handle complexity that breaks traditional template-based approaches.

Stop accepting document processing bottlenecks as an operational reality (by now you know, they are not). The technology exists to eliminate these constraints - companies like Arbor have proven it works at enterprise scale.

PS: If insurance compliance is becoming a bottleneck in your deal flow, you’re facing the same challenge Arbor solved. The question is whether you’ll recognise it as a technology problem or keep treating it as a staffing problem.

PPS: Every metric here comes from real operational results, not marketing projections.

Welcome back, folks!

This week, we're exploring why even the most sophisticated accounting teams collapse under invoice variations, and why companies that crack this have a competitive advantage.

But before we dive in, quick intel you should read this week:

• Manual invoice processing costs up to $16 per document versus as low as $3 with automation, according to the Institute of Finance & Management data

• Global e-invoicing market reached $13.5 billion in 2023 and is projected to grow at a 17.7% CAGR through 2032, driven by government mandates worldwide

• 55 countries worldwide have adopted or are considering e-invoice mandates as governments push digital transformation and tax transparency initiatives

This week, I want to share a fascinating case study that perfectly illustrates why traditional document processing approaches hit a wall at enterprise scale.

The Growth Trap

By 2023, Valtatech was serving over 100 enterprise customers. Each client came with a supplier ecosystem, approval workflows, and, most problematically, invoice formats.

The company was processing over 20,000 invoices monthly, and the diversity was killing them.

Their typical Monday looked like this:

More than 20 employees were dedicated to invoice processing tasks.

The manual workflow was brutal: download each format across different client portals, manually identify vendor details across layouts, extract line items from tables or paragraphs, cross-reference against 100+ different purchase order systems, validate against individual client contracts, route through separate approval workflows, then update multiple client accounting systems.

Time per invoice: Several minutes to hours, depending on complexity
Error rate: Rising as volume increased

Why They Were Struggling

The root cause was the fundamental mismatch between how traditional document processing works and what enterprise service providers need.

The Technical Reality

Traditional OCR systems are built around template recognition. They work well when you have standardised documents from a limited number of sources. But Valtatech's business model created the opposite scenario: unlimited format variations from hundreds of vendors across dozens of clients.

Each new client brought their own supplier ecosystem with unique invoice formats. The traditional approach of creating templates for each format became mathematically impossible when dealing with 100+ vendors across multiple client accounts.

The Business Logic Gap

More critically, invoice processing at enterprise scale requires understanding relationships between vendors, contract terms, and approval workflows. It's not enough to extract "Net 30 payment terms" from a document if you need to validate that against specific cash flow implications.

Traditional systems treated each invoice format as a separate extraction problem. They could pull text from documents, but couldn't understand the business context. Things like which vendor belongs to a client or how approvals should route were not things that the system could understand.

Why Conventional Solutions Failed

When Valtatech tested traditional document AI solutions, demos looked promising. Clean, standardised invoices showed 90-95% accuracy in controlled environments.

But production reality was different. Accuracy crashed by 35% when real client invoices hit the system. Every vendor format variation broke the extraction logic because the systems lacked semantic understanding of business relationships.

The Solution: Semantic Intelligence Over Format Recognition

Instead of trying to standardise vendor invoice formats across 100+ suppliers, we implemented semantic document intelligence that understood invoice logic regardless of presentation format.

How It Works

The new approach treats each invoice as standard business data through different presentation layers:

This semantic validation catches errors that pure format recognition misses.

The Results

After implementing Docsumo's semantic processing, the transformation was dramatic:

• Processing time: From hours down to under 5 minutes with 99% accuracy

• Volume handling: 20,000+ monthly invoices processed automatically across multiple client formats

• Efficiency gain: 3x faster data extraction, enabling capacity increases without proportional staffing

Business impact went deeper:

• Client onboarding was no longer limited by invoice format compatibility

• Teams were freed for strategic client relationship management instead of data entry

• Accuracy improvements eliminated costly error correction workflows

• The company could scale capacity without proportional workforce increases

Why This Approach Works

The success came from solving the right problem. Format variations become irrelevant when the system understands business semantics rather than visual patterns.

Whether vendor information appears in headers, footers, or watermarks across different document types, the intelligence layer recognises the data relationships for business operations. Line item extraction works across presentations because the system understands procurement logic rather than layout templates.

Financial validation happens automatically because the platform learns client-specific business rules and contract terms.

The Broader Lesson

Valtatech's story illustrates a critical principle: successful document automation at scale requires understanding business context.

What This Means for Your Strategy

When you're managing multiple client relationships with hundreds of vendor formats, success depends on building intelligence that adapts to business requirements.

The competitive advantage belongs to companies that solve document processing problems their competitors consider impossible. Semantic intelligence can handle complexity that breaks traditional template-based approaches.

For companies implementing intelligent document processing solutions like Docsumo, the lesson is clear: focus on business logic integration rather than format standardisation. The technology can handle unlimited format variations when the system understands your operational context.

PS: If your finance or ops team is spending hours fixing OCR errors or dealing with supplier templates, you’re solving the wrong problem. You need to invest in smarter automation that understands your business.

PPS: Every metric here comes from real customer accounts, not marketing slides. If Valtatech can kill their document chaos, chances are your team can too.

Welcome back, folks!

This time, we’re digging into why even the smartest document processing systems fall apart the moment messy tables or odd formats land in your inbox.

But before we dive in, quick intel you should read this week:

• Enterprise IDP adoption surges 22% driven by e-invoicing mandates and Gen AI integration across enterprise systems

• Global logistics automation market is projected to grow from USD 34.55 billion to USD 90 billion by 2030

• Natural language processing segment disrupts traditional OCR with 23.80% CAGR, adding contextual understanding beyond text extraction

I came across something this week that got me in deep thoughts about document processing.

C.H. Robinson, a $23 billion logistics giant, just achieved what most CTOs have given up on: processing complex business documents buried inside emails at enterprise scale.

The Email That Started Everything

Picture this: You're managing logistics for a Fortune 500 retailer.

An email lands in your inbox from a supplier with a freight quote request. Embedded in three paragraphs of conversational text is a complex shipping table with pickup locations, delivery windows, weight specifications, and special handling requirements.

Your current process:

Time required: hours.
Scale: thousands of emails daily.
This is exactly what C.H. Robinson faced.

Tens of thousands of emails daily containing freight specifications buried in unstructured text, shipping requirements scattered across attachments, and critical logistics data hidden in conversational language that standard OCR couldn't touch.

When they tested document AI platforms, the demo looked promising.

Clean freight tables showed promising extraction accuracy during demos, but once the system hit production reality, accuracy collapsed.

The Table Extraction Death Trap

Here's what killed their first attempt: logistics emails contain tables that don't look like tables.

A shipping request arrives as conversational text: "We need pickup from Chicago warehouse on Tuesday between 2-4pm, 15 pallets of automotive parts totaling 12,000 lbs, delivery to Detroit facility by Thursday 6am, requires temperature control and special handling for fragile components."

Standard document AI sees unstructured text. But there's actually a complex data table hidden inside:

Every email contains multiple invisible tables with hierarchical relationships that determine pricing, routing, and operational feasibility.

The breakthrough came when C.H. Robinson stopped treating this as an OCR problem and started treating it as a semantic understanding problem.

Engineering Semantic Table Extraction

Mark Albrecht, VP of AI at C.H. Robinson, explains what changed: "Big picture, our tech makes it possible to automate virtually any kind of email transaction and capture efficiencies in global supply chains that just couldn't be achieved before."
They built a system that understands freight semantics embedded in conversational text.

Using Azure AI Foundry and Azure OpenAI, they created models that could parse the hidden data structures within unstructured emails.

The technical architecture treats each email as containing multiple implicit tables:

• Freight Specification Table: Weight, dimensions, quantity, freight class, special handling requirements

• Route Planning Table: Origin address, destination address, pickup windows, delivery requirements, transit constraints

• Service Requirements Table: Equipment type, temperature control, security requirements, documentation needs

• Pricing Variables Table: Distance, fuel costs, accessorial charges, market rates, customer agreements

Instead of extracting text and hoping to find patterns, their system understands that "15 pallets of automotive parts totaling 12,000 lbs" represents structured freight data that maps to specific operational parameters.

The validation layer proved critical. When the system extracts "delivery by Thursday 6am," it validates against:

This semantic validation catches errors that pure extraction misses, enabling production accuracy.

The 32-Second Transformation

The results reveal what's possible when you solve semantic extraction correctly. Price quotes now average 32 seconds versus hours previously required.

The system processes over 2,000 daily pricing requests automatically. Over 10,000 routine email transactions now complete in seconds rather than hours.

But here's the real insight: the speed improvement came from understanding document structure, not improving text extraction speed.
Meghan Hughes, Key Account Manager, describes the impact: "Employees are now freed up to do higher-value work, solving greater challenges for 2,720 of C.H. Robinson's customers so far."

The financial results validate the approach. 15% productivity increase in 2024. Q4 gross profits up 10.4% to $672.9 million. Operating margin expanded 940 basis points to 26.8%.

What This Means for Your Document Processing Strategy

1. Stop optimizing OCR accuracy. Start building semantic understanding. Your documents contain implicit data structures that traditional extraction completely misses. Focus on understanding document semantics rather than improving text recognition.

2. Domain expertise beats generic algorithms. C.H. Robinson succeeded because they built logistics intelligence into their processing workflow. Generic document AI platforms failed because they lacked freight industry knowledge. Your success depends on encoding business logic into extraction processes.

3. Validation enables automation. Production accuracy comes from understanding what extracted data should look like within your business context. Build validation rules that understand operational constraints, business relationships, and domain-specific requirements.

The competitive advantage belongs to companies that solve document processing problems their competitors consider impossible. C.H. Robinson proved that semantic extraction can handle complexity that breaks traditional approaches.

PS: If your business documents contain hidden data structures that standard OCR can't extract, your automation strategy should prioritize semantic understanding over text extraction improvements. At Docsumo, we help enterprises achieve 95%+ accuracy on complex documents by building domain-specific intelligence into processing workflows rather than relying on generic OCR improvements. The companies that solve semantic extraction first will capture disproportionate competitive advantage while their competitors struggle with traditional approaches.

PPS: A lot of the case study was simplified through examples that are made up as their dataset isn’t public

If you need help with anything document processing, hit reply and let us help :)

👋 Quick recap: I’m Rushabh, the Co-founder of Docsumo and you’re getting this email because you signed up on our platform at some point in time.

Before we dive in, some quick things I think you should read this week 👀

• Chelsea NHS Trust trials AI for patient discharge summaries automating medical records

• President Trump signs "Fair Banking" Executive Order targeting debanking practices

I was scrolling through Reddit, and came across this:

That made me dig into what’s up with Metlife.

And as I dug deeper, it got crazier.

In 2019, Sean Nicolello made a discovery that probably kept MetLife's executives awake at night: $100 million in potential savings was trapped in their document processing operations.

TLDR: Operational incompetence at a 152-year-old company that should have figured out paperwork by now.

The Customer Horror Stories

While MetLife's executives were discovering their $100 million problem, customers were living the nightmare:

• Disability claims taking 30+ days instead of the promised 5-10

• Hospital claims sitting in queues for weeks beyond their 3-5 day promise

• Life insurance claims delayed for months despite meeting basic requirements

• Case managers who never returned calls during critical periods

For a company generating $73.49 billion in annual revenue, this revealed systemic document processing failures.

The Paper Mountain Problem

To understand MetLife's $100 million opportunity, picture this: Millions of insurance contracts dating back over 100 years, customer onboarding involving "dozens of documents with hundreds of pages," and group customer data arriving in dozens of insurance-specific invoice forms.

Think about it, customer onboarding alone involves dozens of documents with hundreds of pages, while group customer data arrives in dozens of insurance-specific invoice forms with variable formats due to customization.

If you’re thinking this could be solved by OCR, nope.

This was an unstructured data nightmare costing hundreds of thousands of hours annually.

The Robot Army Deployment

MetLife's response was ambitious: deploy 130+ Blue Prism robots across 11 processes in U.S. claims operations.

The initial results were encouraging:

But the real breakthrough came in 2025.

The AI Breakthrough

This June, MetLife announced a global partnership with Sprout.ai that would accelerate document processing transformation across the US, Asia, and Latin America.

Sprout.ai's platform integrates directly into MetLife's existing systems, enabling automated claims decisions that are both swift and fair.

The technology had already proven itself in Latin America, where it demonstrated measurable improvements in claims turnaround times and customer satisfaction.

The Unexpected Bonus

MetLife's document processing transformation delivered a surprise benefit: dramatically improved fraud detection. Their implementation of Shift Technologies' Force AI solution in Japan resulted in:

This dual benefit of faster processing and better fraud detection showed how intelligent document processing creates compound value.

The Industry Wake-Up Call

MetLife's transformation comes as the entire insurance industry faces a document processing crisis.

Accenture research reveals that $170 billion in global insurance premiums are at risk by 2027 due to poor claims experiences.

The competitive pressure is real:

• 65% of insurers plan investments of $10 million or more in AI technologies over the next three years

• The intelligent document processing market is growing from $10.57 billion in 2025 to a projected $66.68 billion by 2032

My two cents

MetLife's journey offers lessons for enterprise document processing:

MetLife's $100 million document processing transformation continues expanding globally. But challenges remain as customer reviews still reflect processing delays, suggesting that technology alone cannot solve systemic service problems, but in my opinion it’s a matter of time.

Something to think about

As the insurance industry faces pressure for faster, more accurate claims processing, MetLife's transformation is just testament to the need for intelligent document processing. 

PS: If a 152-year-old company can find $100 million in their document processing, start thinking of what's hiding in yours.

If you need help with anything document processing, hit reply and let us help :)