StarZero All shows Try the podcast clipper →

Nov 29, 2023 · 24m · mad

How Glean AI Slashes Vendor Costs: CEO Howard Katzenberg on AI Accounting

Howard Katzenberg · 16m spoken Matt Turck · 3m spoken Tanya Dua · 32s spoken Ed Manzi · 19s spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this episode of The MAD Podcast, host Matt Turck interviews Glean AI CEO Howard Katzenberg about how his platform uses AI and LLMs to automate accounts payable and vendor spend management. Katzenberg shares his journey from CFO to founder, explaining how Glean AI drives cost savings, streamlines invoice extraction, and shapes company spending culture.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 14.1% of the talking time here. How this is scored →

Matt as informed peer 3.2 Guest teaching 2.5 Guest disagreement 0.5 Matt pushing back 0.7
05100:0010:0020:000:10–4:33 · Matt as informed peer 2/10 Howard Katzenberg's Path from CFO to Founder Matt opens with a warm background question regarding Howard's transition from CFO to founder. Howard provides a detailed breakdown of manual CFO cost-cutting processes and why legacy approval workflows fail to catch vendor spend leakage.4:33–9:35 · Matt as informed peer 6/10 Glean AI: Bill.com with a Brain Matt demonstrates clear market expertise by citing Bill.com's revenue, market cap fluctuations, and the overall scale of the AP automation sector. Howard explains Glean's product architecture and transition from NLP/OCR models to LLMs.9:35–12:18 · Matt as informed peer 4/10 Human-in-the-Loop Validation and AI Defensibility Matt asks about human validation in finance workflows and challenges Howard on business defensibility when building on commercial LLMs. Howard details how proxy confidence scores and historical data models create competitive defensibility.12:18–14:43 · Matt as informed peer 3/10 Customer Impact, ROI, and Culture Shift Matt asks Howard to articulate the ROI framing between reduced headcount and faster processing. Howard clarifies that the value stems from spend culture changes yielding cash savings alongside moderate FTE time savings.14:43–18:18 · Matt as informed peer 3/10 Future Vision and Leading Technical AI Teams Matt poses a thoughtful question on how non-technical founders manage and interface with technical AI engineering teams. Howard describes his humble, curiosity-led approach to learning from internal experts.18:18–24:29 · Matt as informed peer 1/10 Audience Q&A: Market Trends, PE Channels, and Fine-Tuning Audience members lead the Q&A on macro spend trends, private equity channels, in-house fine-tuning, and vendor privacy. Howard shows mild contrarian pushback when explaining why he rejected his internal data team's static cost assumptions regarding third-party LLMs.0:10–4:33 · Guest teaching 3/10 Howard Katzenberg's Path from CFO to Founder Matt opens with a warm background question regarding Howard's transition from CFO to founder. Howard provides a detailed breakdown of manual CFO cost-cutting processes and why legacy approval workflows fail to catch vendor spend leakage.4:33–9:35 · Guest teaching 2/10 Glean AI: Bill.com with a Brain Matt demonstrates clear market expertise by citing Bill.com's revenue, market cap fluctuations, and the overall scale of the AP automation sector. Howard explains Glean's product architecture and transition from NLP/OCR models to LLMs.9:35–12:18 · Guest teaching 3/10 Human-in-the-Loop Validation and AI Defensibility Matt asks about human validation in finance workflows and challenges Howard on business defensibility when building on commercial LLMs. Howard details how proxy confidence scores and historical data models create competitive defensibility.12:18–14:43 · Guest teaching 2/10 Customer Impact, ROI, and Culture Shift Matt asks Howard to articulate the ROI framing between reduced headcount and faster processing. Howard clarifies that the value stems from spend culture changes yielding cash savings alongside moderate FTE time savings.14:43–18:18 · Guest teaching 1/10 Future Vision and Leading Technical AI Teams Matt poses a thoughtful question on how non-technical founders manage and interface with technical AI engineering teams. Howard describes his humble, curiosity-led approach to learning from internal experts.18:18–24:29 · Guest teaching 4/10 Audience Q&A: Market Trends, PE Channels, and Fine-Tuning Audience members lead the Q&A on macro spend trends, private equity channels, in-house fine-tuning, and vendor privacy. Howard shows mild contrarian pushback when explaining why he rejected his internal data team's static cost assumptions regarding third-party LLMs.0:10–4:33 · Guest disagreement 0/10 Howard Katzenberg's Path from CFO to Founder Matt opens with a warm background question regarding Howard's transition from CFO to founder. Howard provides a detailed breakdown of manual CFO cost-cutting processes and why legacy approval workflows fail to catch vendor spend leakage.4:33–9:35 · Guest disagreement 0/10 Glean AI: Bill.com with a Brain Matt demonstrates clear market expertise by citing Bill.com's revenue, market cap fluctuations, and the overall scale of the AP automation sector. Howard explains Glean's product architecture and transition from NLP/OCR models to LLMs.9:35–12:18 · Guest disagreement 0/10 Human-in-the-Loop Validation and AI Defensibility Matt asks about human validation in finance workflows and challenges Howard on business defensibility when building on commercial LLMs. Howard details how proxy confidence scores and historical data models create competitive defensibility.12:18–14:43 · Guest disagreement 0/10 Customer Impact, ROI, and Culture Shift Matt asks Howard to articulate the ROI framing between reduced headcount and faster processing. Howard clarifies that the value stems from spend culture changes yielding cash savings alongside moderate FTE time savings.14:43–18:18 · Guest disagreement 0/10 Future Vision and Leading Technical AI Teams Matt poses a thoughtful question on how non-technical founders manage and interface with technical AI engineering teams. Howard describes his humble, curiosity-led approach to learning from internal experts.18:18–24:29 · Guest disagreement 3/10 Audience Q&A: Market Trends, PE Channels, and Fine-Tuning Audience members lead the Q&A on macro spend trends, private equity channels, in-house fine-tuning, and vendor privacy. Howard shows mild contrarian pushback when explaining why he rejected his internal data team's static cost assumptions regarding third-party LLMs.0:10–4:33 · Matt pushing back 0/10 Howard Katzenberg's Path from CFO to Founder Matt opens with a warm background question regarding Howard's transition from CFO to founder. Howard provides a detailed breakdown of manual CFO cost-cutting processes and why legacy approval workflows fail to catch vendor spend leakage.4:33–9:35 · Matt pushing back 1/10 Glean AI: Bill.com with a Brain Matt demonstrates clear market expertise by citing Bill.com's revenue, market cap fluctuations, and the overall scale of the AP automation sector. Howard explains Glean's product architecture and transition from NLP/OCR models to LLMs.9:35–12:18 · Matt pushing back 2/10 Human-in-the-Loop Validation and AI Defensibility Matt asks about human validation in finance workflows and challenges Howard on business defensibility when building on commercial LLMs. Howard details how proxy confidence scores and historical data models create competitive defensibility.12:18–14:43 · Matt pushing back 1/10 Customer Impact, ROI, and Culture Shift Matt asks Howard to articulate the ROI framing between reduced headcount and faster processing. Howard clarifies that the value stems from spend culture changes yielding cash savings alongside moderate FTE time savings.14:43–18:18 · Matt pushing back 0/10 Future Vision and Leading Technical AI Teams Matt poses a thoughtful question on how non-technical founders manage and interface with technical AI engineering teams. Howard describes his humble, curiosity-led approach to learning from internal experts.18:18–24:29 · Matt pushing back 0/10 Audience Q&A: Market Trends, PE Channels, and Fine-Tuning Audience members lead the Q&A on macro spend trends, private equity channels, in-house fine-tuning, and vendor privacy. Howard shows mild contrarian pushback when explaining why he rejected his internal data team's static cost assumptions regarding third-party LLMs.

speaking balance: gold is Matt, purple is the guest (3 minute bins)

0:00 · Matt 18% · guest 82%0:00 · Matt 18% · guest 82%3:00 · Matt 17.5% · guest 82.5%3:00 · Matt 17.5% · guest 82.5%6:00 · Matt 22.7% · guest 77.3%6:00 · Matt 22.7% · guest 77.3%9:00 · Matt 16.6% · guest 83.4%9:00 · Matt 16.6% · guest 83.4%12:00 · Matt 11.7% · guest 88.3%12:00 · Matt 11.7% · guest 88.3%15:00 · Matt 24.7% · guest 75.3%15:00 · Matt 24.7% · guest 75.3%18:00 · Matt 1.9% · guest 98.1%18:00 · Matt 1.9% · guest 98.1%21:00 · Matt 0% · guest 100%21:00 · Matt 0% · guest 100%24:00 · Matt 7.1% · guest 92.9%24:00 · Matt 7.1% · guest 92.9%
Sharpest disagreement ▶ 21:05 Rejecting static third-party API cost assumptions

Howard directly rejects the premise proposed by his internal data scientist regarding in-house model savings, arguing that commercial LLM compute and API costs will decline rapidly over time due to competition.

 Hardest push from Matt ▶ 10:31 Challenging AI wrapper defensibility

Matt pushes Howard to explain how Glean builds actual enterprise defensibility rather than serving as a thin wrapper on commercial LLM vendors.

 Biggest teaching moment ▶ 11:04 Explaining context-aware LLM prompts for multi-invoice trend analysis

Howard educates the audience on why standard LLM extractions fail without feeding pre-calculated historical invoice CSV data, showing how data modeling enables unique contextual summaries.

 Matt holds his own ▶ 5:43 Demonstrating market domain expertise on AP automation valuation

Matt displays deep knowledge of the financial software ecosystem by detailing Bill.com's revenue figures, market cap fluctuations, and explaining why accounts payable is a massive market.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
Howard Katzenberg's Path from CFO to Founder 2300 Matt opens with a warm background question regarding Howard's transition from CFO to founder. Howard provides a detailed breakdown of manual CFO cost-cutting processes and why legacy approval workflows fail to catch vendor spend leakage.
Glean AI: Bill.com with a Brain 6201 Matt demonstrates clear market expertise by citing Bill.com's revenue, market cap fluctuations, and the overall scale of the AP automation sector. Howard explains Glean's product architecture and transition from NLP/OCR models to LLMs.
Human-in-the-Loop Validation and AI Defensibility 4302 Matt asks about human validation in finance workflows and challenges Howard on business defensibility when building on commercial LLMs. Howard details how proxy confidence scores and historical data models create competitive defensibility.
Customer Impact, ROI, and Culture Shift 3201 Matt asks Howard to articulate the ROI framing between reduced headcount and faster processing. Howard clarifies that the value stems from spend culture changes yielding cash savings alongside moderate FTE time savings.
Future Vision and Leading Technical AI Teams 3100 Matt poses a thoughtful question on how non-technical founders manage and interface with technical AI engineering teams. Howard describes his humble, curiosity-led approach to learning from internal experts.
Audience Q&A: Market Trends, PE Channels, and Fine-Tuning 1430 Audience members lead the Q&A on macro spend trends, private equity channels, in-house fine-tuning, and vendor privacy. Howard shows mild contrarian pushback when explaining why he rejected his internal data team's static cost assumptions regarding third-party LLMs.

Statements from this episode (11)

Assertion Not checkable as stated
Katzenberg: Manual vendor spend audits consistently yielded about 10% savings
“Inevitably, we'd find errors, we'd find consolidation opportunities, we'd play bad guy negotiations, but we'd find about 10% of savings opportunities each time we conducted this exercise.”
Howard Katzenberg Nov 29, 2023 ▶ 2:12
Assertion Not checkable as stated
Better identified $1.5M in savings on $15M of annual vendor spend
“The last time we did it was in 2019 and, at Better, the run rate on vendor spend at the time was about fifteen million dollars annually. And we've identified 1.5 million of savings.”
Howard Katzenberg Nov 29, 2023 ▶ 2:27
Assertion Not checkable as stated
Katzenberg: 99 out of 100 vendor bills were approved automatically at Better
“And when I actually analyzed what our approval rate was at Better, 99 out of a hundred bills were getting approved.”
Howard Katzenberg Nov 29, 2023 ▶ 3:56
Disclosure
Glean AI shifted 100% of its machine learning stack to LLMs
“Over the course of the last 12 to 18 months, we've done a hundred percent shift to LLM modeling for the complete stack. And we're using Vertex and OpenAI.”
Howard Katzenberg Nov 29, 2023 ▶ 8:57
Assertion Not checkable as stated
LLM migration eliminated annotation teams and increased Glean AI's gross margins
“The accuracy rates are better. Our cost to compute has gone down and like, it is the data scientist. We don't need a team to do annotations anymore for the models. And it's just like, it's like our gross margins. It's done like two step functions over the cour…”
Howard Katzenberg Nov 29, 2023 ▶ 9:16
Assertion Not checkable as stated
Katzenberg: Commercial LLMs do not provide extraction confidence scores
“One of the drawbacks of using LLMs is like, we're not receiving confidence scores. On the extractions.”
Howard Katzenberg Nov 29, 2023 ▶ 9:55
Assertion Not checkable as stated
Katzenberg: LLMs are very poor at performing calculations
“We're going to do the calculations because the LLMs are very poor at that.”
Howard Katzenberg Nov 29, 2023 ▶ 11:41
Assertion Not checkable as stated
League Apps saved $20,000 on Salesforce contract using Glean AI benchmarking data
“So we gave him benchmarking data and he reported back that they saved 20 K versus their prior contract.”
Howard Katzenberg Nov 29, 2023 ▶ 13:08
Assertion Not checkable as stated
League Apps estimates Glean AI saves 2-3% of non-payroll spend
“And then he also said like, he estimates that Glean, like the value is like two to three percent of non-payroll spend.”
Howard Katzenberg Nov 29, 2023 ▶ 13:15
Assertion Not checkable as stated
Katzenberg: Glean AI business grew over 3x in 2023
“Our business is up like three X, over three X so far this year.”
Howard Katzenberg Nov 29, 2023 ▶ 18:56
Prediction Held up
Katzenberg: External AI model pricing will fall over 12-18 months
“I think their pricing will continue to improve. So we can't just look at like where their pricing is today and assume it's going to be static over the next 12 months or 18 months. Cause like as they compete, their compute costs will come down. Like they'll, I …”
Howard Katzenberg Nov 29, 2023 ▶ 21:20
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more