the mechanics of business and real-world applications of machine intelligence

Key Takeaways

Ben had two successful exits, both enabled by taking data-driven sales team & process optimization to the extreme. New technologies in the world of AI have moved the max optimization limit much further forward, and it's now possible to take Ben's ideas to their full potential.

Topics Covered

• Background on Yodle, Ben's first business
  • Ben's businesses have involved building large scalable sales and support teams, scaling them very quickly, and typically selling to small businesses across multiple verticals 01:23
  • Founded Yodle in 2005 around thesis that historically small businesses advertised in the yellow pages, and in the future, their lives will be online; ultimately Yodle supported these small businesses in all aspects of building their online presence 02:02
• The insights from optimizing Yodle's GTM function 03:19
  • Original go-to-market was trying to hire Yellowpages salespeople, which didn't work because they couldn't sell online marketing 03:55
  • College grads were also tough because they didn't want to put in the sales hours; they thought they were too good for the job 04:06
  • Ultimately realized we needed to build a scalable scripted process with super hardworking and coachable people 04:22
  • The company ultimately scaled to 1500 people, and over 1000 were on the phone; we made 80 million calls a year, calling all 20 million small businesses on average four times 04:49
  • We had a tremendous amount of data on every one of those calls 05:07
  • To start it was all US based, but we found through time the process was so dialed in and scripted that we had a team in St. Lucia that worked well; we were the only business at the time that did remote outbound sales & credit card collection 05:14
  • I would never outsource sales until you figure out your sales process; sales is too critical to do anything that could hurt conversion 06:01
    • To scale a sales team, you first need high IQ people to figure out how to pitch the product, then you can scale 06:13
    • What does a scripted playbook look like? 06:59
      • It was dialed in, meaning words being read from a script, not to say that you couldn't deviate if someone had a question, but especially in the demo process where it was much more us talking to them, we wanted everyone to say the exact same thing 07:31
      • Process was in four main piece: 1) cold call / peaking interest, 2) fact-facting / needs-analysis, 3) the demo, and 4) the negotation 07:51
      • All of our objections were scripted as well; including e.g. "I need to talk to my wife/husband" 08:37
        • We had transcripts of every call, and would look through to see where wife/husband/partner was mentioned, see which became a sale and which didn't 09:00
        • If someone said something off-script that scalable and worked well, and then take a separate team called the sales lab to test it 09:14
        • Then if it worked, we could take the new rebuttal and deploy it to the sales team 09:37
  • How did the analysis work from a technology perspective? 09:45
    • Every call was transcribed and stored in a database; then you can query for specific words/objections 09:50
    • They heyday of this optimization was between 2011-2016; the transcription wasn't perfect, but it was way better than anyone else was doing 10:31
    • The tech available today would be made us much more efficient and be able to iterate faster 12:00
      • There are also parts of the demo we could have automated too 12:15
    • Back then, you're leaving a lot of voicemails, so we found very early on that the voicemail you left drove a huge differenace in callbacks... it varied by reps, time of day, etc. 12:26
      • Eventually nobody had to leave a voicemail ever again; we just took those with the highest callback rates and played them back 12:58
      • With the tech today you could do that in not just voicemails; you could do it for a lot of pieces of the demo and cold call as well 13:14
    • The other big opportunity today is in rep training and feedback; back then the quality team did their work manually, and it in the form of random audits... today you can analyze every word 13:35
      • You could be much more proactive too; previously we were just reactive to customer complaints in many cases 14:55
    • Then in hiring, we wanted to test for 3 things: 1) coachability 2) work ethic, and 3) attitude 15:33
      • Re attitude, people are making 200 calls a day and the best reps had 1-2 sales; you need to handle rejection well 16:10
      • We did a fairly good job of screening for those, but today there are definitely better ways to screen for these traits 16:35
      • The people who were most successful were best at making the script come to life; many of our best people were actors, not salespeople 17:14
      • Today your interview could be a lot more efficient; you just automate the process of having people read the sales script, and you can now make it more interactive 17:41
    • Post hiring, better training grounds are helpful because reps can get practice without burning leads; that's always valuable 18:18
• Background on Ben's next company, Opcity 20:37
  • We built a network of 150k real estate agents to whom we transferred live consumers who were looking to buy properties; we bought these leads from online portals and took a fee when an agent successfully closed a sale with one of our leads 20:58
  • When you put your contact info into e.g. a Realtor.com portal, we could call you in 4 seconds and match you with someone in our agent network 21:18
  • We would call the agent, explain the situation, then bridge the call, introduce the two parties, and then help the agent and consumer actually finish that transaction 21:33
  • Three major elements from a data science perspective 22:03
    • The first was matching agents and consumers, based on attributes of consumer and attributes in agent network 22:09
      • The data science problem was not just matching, but how to optimize the waterfall... if agents 1-2 were very similar and speed is critical, you want to quickly flip to the 2nd agent if the 1st isn't available 22:52
    • The second problem was we had a database of millions of consumers that the CSR team could call; how do you know which consumer to call any any moment? 23:13
    • The 3rd and final piece was once you have the consumer/agent match, what can you do to help the deal close; what actions should we take; is it emailing, texting, calling? 23:33
  • Biggest learning from Yodle was just because you sold a lead doesn't it wasn't a good experience for consumer in the lead, or the plumber who bought it; that was the core insight for Opcity 24:38
• Key learnings from Opcity 25:18
  • Timeliness matters a lot; if a consumer is still on the portal, you had a much higher chance of converting 25:27
  • Second most important was that the agent you were connecting a consumer with had actually closed a deal in the zip code at a similar price point 27:15
  • Another interesting one was languages spoken -- a lot of times people said a lead was no good and it was because they didn't speak Spanish 27:44
  • The core value in Opcity was because since we had such a big network, we could always make sure we got the most value out of any given lead; if you're just a single buyer, you may get a good lead, but you're just not the right person to receive it 27:55
  • Related to language, we also got creative with "these people speak slower and have southern accents", but historical conversion and close rate was a more significant factor 28:44
• What is the unifying strategy/theme in how you think about running a business? 34:12
  • Build the infrastructure and culture within your company to do nonstop testing and make it really easy 34:37
  • Sacred cows make the best burgers; nothing should be untouchable; we should always be challenging the status quo 34:52
  • Much of this data may not be quantitative / in spreadsheets; Sam Walton famously spent more time in his competitors' stores than the owners of those stores 37:47
• Will the ability to leverage operational data make it easier or harder to build an early-stage company? 40:24
  • **Startups have a bigger advantage now than they've ever had"; you now have so much more leverage 40:59
  • Yes big companies have access to data, but they move slowly because they're bureaucratic, they don't know how to use their data properly, often a lot of the best people don't want to work in these larger places because they can't make as big of an impact 41:37
• Do you believe in the efficient markets hypothesis in the world of finding applications for new innovation? 42:21
  • This is the early days of the internet. There were lots that didn't make it, but a few who made it really big and who are central to our lives. 42:41
  • Compare to Microsoft Word/Excel/PPT -- people use it put it on their resume; now it's a given. What are those tools today that will follow a similar path? 43:00
• Ben's overview of his latest business, SetPoint: They help intermediate the relationships between 1) tech-enabled players (e.g. lenders, ibuyers, etc.) that need credit / capital to fund the products they offer and 2) the large banks and originators of this capital 43:55
  • Today it's all in Excel 44:20
  • If what you describe today is just "act 1", what is the long-term vision? 45:47
  • This exchange of assets/risk and capital happens at multiple points down the chain of the life of a given asset; along that chain, there's a lot of owners/buyers that need to do the same verification work; doing this in a single platform makes a lot of sense 46:18
  • Another we've launched and is called "portfolio manager"; helps originators optimize across the facilities, so you're using the optimal facility to get the lowest cost of capital 46:55
  • By reducing the cost of capital for an originator, you can pass this savings on to the consumer 47:18
    • Like with real roads, when you pave and make highways, people get things much faster 47:32
    • "If credit had "super highways", credit could flow much more easily and cheaply" 47:43
• Where's the "through line" in your experiences? 47:55
  • I saw this problem first-hand at my angel investments in prop tech; the biggest risk was humans making errors as part of the asset origination 48:14
  • You need a first customer you intimately know that can be a sandbox; I had this at each of my businesses 49:13
  • With Opcity, I walked in with 15 years worth of data with an actual brokerage, so I could test things before going to market 49:47
• What gets your most excited about applying LLMs / AI generally at SetPoint? 50:56
  • Being able to ask questions of these 400 page loan agreements; whover negotiated the facility you may not be able to reach anymore, so being able to pull out key details is critical 51:13
  • Not just pull info out, but also automate more of the process with that information; if you double fund a property because a human made a mistake, it's a big deal 51:38