- Everyday AI
- Posts
- AI & Trust: When 98% accuracy won't cut it and how Sage can fix it
AI & Trust: When 98% accuracy won't cut it and how Sage can fix it
Sage CTO on AI and Finance, Google Labs' interactive chart visualizations, Perplexity hits 780M monthly queries, AMD acquires Untether AI and more!
Everyday AI
June 06, 2025
š Subscribe Here | š£ Hire Us To Speak | š¤ Partner with Us | š¤ Grow with GenAI
In Partnership With
Sup yāall! š
No joke. When I saw what Sage Copilotās AI could do, I instantly emailed my CPA.
Whether youāre a small business owner, accountant or CFO, thereās a chance you or someone on your team is often trying to close the book.
And it often means a TON of manual spreadsheet, data crunching and number anxiety.
(I personally hate it.)
Sage Copilot solves this.
It uses AI to automate the busyworkālike approvals, reports, and reconciliations.
Itās like having an extra set of hands (and a brain) to keep your companyās number game flowing smoothly.
Best part?
You can see the work itās doing with Sageās new AI Trust Label. (Check out todayās show for more on that.)
Try Sage Copilot and see how smarter financial operations can be with transparent and trustworthy AI.
Anyone else breaking up with Quickbooks? Or just me?
āļø
Jordan
Outsmart The Future
Today in Everyday AI
7 minute read
š Daily Podcast Episode: 98% accuracy sounds great right? For Sage, thatās not enough. We learn from Sageās CTO on how you can build trust with AI and your finances. Give it a listen.
šµļøāāļø Fresh Finds: ChatGPT used for malicious intent by Chinese groups, Microsoft Copilot now includes visuals in answers and Cursor parent company raises $900M. Read on for Fresh Finds.
š Byte Sized Daily AI News: Google Labs gets interactive chart visualizations, Perplexity hits 780M monthly queries and AMD acquires Untether AI. For that and more, read on for Byte Sized News.
š§ Learn & Leveraging AI: We break down how Sage is setting new standards in AI accuracy and trust for financial tasks, ensuring near-perfect precision and transparency. Keep reading for that!
ā©ļø Donāt miss out: Did you miss our last newsletter? We talked about Gemini 2.5 Pro getting a coding update, Anthropic's AI models for U.S. Gov., Alphabet CEO pushing back on AIās workforce threat and more. Check it here!
In Partnership With
AI & Trust: When 98% accuracy wonāt cut it and how Sage can fix it š”
Your CFO just lost sleep over a single missing penny... again.
Here's the thing about finance teams: they'll hunt for days to find ONE CENT that's off in their books. Because in accounting, even 98% accuracy = complete failure.
So when it comes to your company's finances and AI, there's a HUGE elephant in the room: trust.
Sage is changing the conversation around AI, trust and your books.
Sage is a global leader in cloud-based accounting, financial management, and business management solutions.
Sage CTO Aaron Harris joins us to show us the new recipe for trust they're cooking up.
Also on the pod today:
ā¢ Transparent Trust Labels in AI Usage š·
ā¢ Sage Copilot's Accounting AI Accuracy šÆ
ā¢ AI Implementation Strategy in Accounting š”ļø
Itāll be worth your 27 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Hereās our favorite AI finds from across the web:
New AI Tool Spotlight ā Chat4Data is an AI web scraper plugin for Chrome, Xavier AI is an AI strategy consultant and Google Whisk generates images using images as prompts.
OpenAI ā OpenAI has found more Chinese groups using ChatGPT for malicious use.
Microsoft ā Copilot now answers your questions with visuals, including images and videos.
Got a question? I have answersānow with visuals. Get real images, videos, and more with cards that make searching smarter.
ā Microsoft Copilot (@Copilot)
4:00 PM ā¢ Jun 6, 2025
A leaked Microsoft org chat shows the team Jay Parikh created to lead CoreAI.
Money in AI ā Anysphere, the maker of AI coding assistant Cursor, has raised $900M at $9.9 billion valuation.
Trending in AI ā Builder.ai has filed for bankruptcy after creditors seize accounts.
Anthropic ā Anthropic has appointed national security expert Richard Fontaine to Anthropicās Long-Term Benefit Trust.
Future of Work ā Some experts are claiming AI may cause a broken career ladder for college graduates.
AI in Healthcare - Stanford Medicine has unveiled ChatER, new AI software that can expedite chart reviews and other tasks.
1. Google Labsā AI Mode Adds Interactive Finance Charts š
Google is rolling out new interactive chart visualizations in AI Mode, now available in Labs, aimed at transforming how users analyze stock and mutual fund data. This feature uses advanced AI to deliver custom graphs and detailed explanations, allowing quick comparisons over time without manual research. Users can even ask follow-up questions like dividend payments, with the AI understanding and expanding the analysis.
According to Soufi Esmaeilzadeh, Director of Product Management for Search at Google, this update leverages Geminiās sophisticated reasoning to make financial insights more accessible and actionable for investors and professionals alike.
2. Perplexityās AI Search Hits 780M Queries in May šļø
Perplexityās CEO Aravind Srinivas revealed at Bloombergās Tech Summit that the AI search engine is growing over 20% monthly, reaching 780 million queries in May and aiming for a billion queries weekly within a year. The company plans to boost this momentum with its upcoming Comet browser, designed not just to surf the web but to act as a ācognitive operating systemā that completes tasks through AI-driven browsing sessions.
According to Srinivas, this approach also opens doors for new revenue streams through premium ads by tracking user activity beyond the search app
3. AMD Snags Untether AI Team to Boost Edge and Data Center Chips š¤
AMD has quietly acquired the engineering talent behind Untether AI, a startup known for its energy-efficient AI inference chips designed for edge and data center use, according to CRN. This move strengthens AMDās foothold in AI hardware, aiming to improve its AI compiler, kernel development, and SoC design just as competition with NVIDIA heats up.
Untether AI will cease its product support, but its teamās expertise is poised to accelerate AMDās AI capabilities at a critical time when power-efficient AI processing is in high demand.
4. Appleās AI Challenges Spotlighted Ahead of WWDC 2025 š¬
As Appleās annual Worldwide Developers Conference (WWDC) kicks off, investors and developers are eager for clarity on Apple Intelligenceās future after a rocky first year marked by delayed features and lagging behind AI leaders like OpenAI and Google, according to Bloomberg. Despite its hardware edge and huge iPhone base, Appleās cautious AI rollout has underwhelmed, prompting calls for major movesāpossibly a blockbuster acquisition like Anthropicāto close the gap.
With competitors ramping up AI spending and rolling out advanced AI-powered products, WWDC is shaping up as a critical moment to see if Apple can shift gears in AI or risk falling further behind.
5. U.S.-UAE AI Data Hub Deal Hits Security Snags āļøāš„
A high-profile deal to build a massive AI data center campus in the UAE featuring top U.S. tech firms like NVIDIA and OpenAI is on hold due to unresolved security concerns. Despite President Trumpās recent visit and Abu Dhabiās pledge to align with U.S. regulations, Washington remains wary of the UAEās close ties with China, complicating chip export approvals.
The project, slated to go live in 2026 and powered by cutting-edge NVIDIA AI chips, promises to be a major player in global AI infrastructure but faces political hurdles that could delay its impact.
6. Meta Faces Backlash Over AI Deepfake Ads š
Metaās Oversight Board recently overturned Metaās decision to keep up a Facebook post featuring an AI-generated deepfake of Brazilian soccer legend Ronaldo NazĆ”rio in a gambling ad, which had racked up over 600,000 views.
According to the Board, this incident reveals a broader problem with Metaās platform, suggesting it may be hosting large amounts of scam content without empowering reviewers to enforce rules effectively.
š¦¾How You Can Leverage:
Most business leaders think 98% AI accuracy is pretty dang good.
But when it comes to finance or accounting?
98% can be pretty much catastrophic and that 2% gap is bigger than the moon.
We're talking about the hidden world of financial AI where one missing penny triggers week-long investigations and 99% accuracy still isn't remotely acceptable.
Thatās why in today's Everyday AI episode, we dive deep into why "good enough" AI becomes "absolutely catastrophic" the moment it touches your books.
We joined Aaron Harris at Sage Future to learn how Sage, a global leader in accounting, financial, HR, and business management software and services, is erasing that 2% gap.
How?
Aaron has spent 25 years learning what happens when technology meets money. The lessons apply way beyond accounting software.
Ready to understand why CFOs hunt missing pennies for days while rejecting your "pretty good" AI?
And see how Sageās new AI-powered finance assistant, Sage Copilot, is changing how companies can close their books?
Same.
1 ā The Five-Model Invoice Reality Check š²
Think one AI model reads invoices perfectly?
NOPE.
Aaron revealed something that'll blow your mind. Sage uses FIVE separate models for a single invoice. One hunts totals. Another double-checks that work. Others handle different sections.
Why the crazy overkill?
Off-the-shelf models hit 75-80% accuracy overall. Sounds decent, right?
Nah, shorties.
When tested specifically on finding invoice totals, they crashed harder than a Windows 95 computer. We're talking 30-40% accuracy.
They'd nail vendor names and dates like champions. Then completely butcher the ONE number that actually matters.
Aaron said something that'll stick with you forever. You don't use AI to do math.
You give AI a calculator to do the math.
This is why financial AI requires a completely different approach than content generation or customer service bots.
Try this:
Test your AI's accuracy on your three most critical data points separately.
Overall accuracy scores are LYING to you. If any critical point drops below 95%, you need specialized models or traditional programming for those specific tasks. Map these failure points before deployment, not after your first "oops we lost $50K" moment.
2 ā Server Webcams to AI Nutrition Labels š¹
Aaron told us a story that perfectly explains how trust works in new technology.
Twenty-five years ago, nobody trusted putting financial data in the cloud.
So his team did something brilliant (and slightly ridiculous). They installed a webcam pointing directly at their server. Customers could click "see my data" and literally WATCH their files through a live feed.
Worked like magic.
Until Murphy's Law struck.
A maintenance tech was doing server work, bending over, cables everywhere. Unfortunately, his belt had other plans. A sales rep chose that EXACT moment to demo the "see my data" feature to a hot prospect.
They got full transparency alright.
More than anyone bargained for.
That webcam disaster taught Aaron something golden. Transparency builds trust, but you need the RIGHT kind of transparency.
Fast forward to today's episode.
Aaron revealed that Sage just launched Trust Labels. Click any AI feature and BAM. You see exactly which models they use, training methods, safeguards, data handling. Everything.
Think nutrition facts, but for AI.
No regulations exist for AI transparency yet. So they're creating the playbook and pushing the entire industry to follow suit.
Try this:
Document EVERYTHING about your AI tools before stakeholders start asking uncomfortable questions. Which models? What training data? What happens when things go sideways? Who's actually reviewing the output?
Create this transparency document NOW. Trust problems are infinitely easier to prevent than fix after someone loses faith in your AI capabilities.
3 ā Why 7 Billion Beat 2 Trillion Parameters š§
Aaron dropped a strategy that totally breaks Silicon Valley's "bigger is better" obsession.
Instead of chasing GPT-4's massive parameter count, they grabbed a 7 billion parameter model.
Then turned it into an accounting GENIUS.
They fed it CPA exams. Accounting textbooks. Product documentation. API code. Every single piece of financial expertise they could get their hands on.
But here's the power move that nobody saw coming.
The American Institute of CPAs just partnered with Sage to contribute their professional content for model training.
Read that again.
The organization that literally CERTIFIES accountants is now helping build AI for accounting tasks.
They're not fighting the technology. They're making it exponentially smarter.
Aaron calls this approach "accept humility, embrace responsibility." Silicon Valley's beloved "move fast and break things" philosophy dies instantly when you're handling someone's money.
One bad AI experience with a CFO?
Game over. Forever. No do-overs.
This partnership signals something massive. We're moving past the hype phase into actual professional-grade AI integration where industry experts actively contribute to better models.
Try this:
Stop obsessing over the biggest, shiniest AI models.
If youāre trying to find the right AI solution for something domain-specific, you should look at the domain specialists.
In many cases, working with a smaller model thatās fine-tuned around institutional knowledge is going to be the winner.
Thatās exactly what Sage did. They built a smaller, specialized model and fine-tuned it with real accounting dataānot just generic internet text.
This delivered the precision and trust finance teams demand. For critical work, the smartest AI isnāt always the biggestāitās the one built for your field.
The one that can even get you past that 98%.
If youāre trying to tap AI to help with your financials, you should check out Sage. Theyāve been global pioneers in finance, accounting and technology for decades.
And if you want to see their unique vision on Agentic AI, we highly recommend you check it out here.
Reply