From Wait to Great: Leveraging LLMs to Boost Aleph’s Organizational Velocity | by Eden Shochat | Aleph

Shout out to fellow productivity geeks that commented on early drafts of this: Eytan Levit (introduced the notion of strategic vs. tactical wait states), Omer Perchik (pushing me on value vs. privacy costs), Jacob Bank (need for culture shift to prefer decision velocity even at some cost to quality).

Organizational velocity is determined by two key factors:

1. Speed of decision-making: How quickly decisions are made directly affects the ability to seize opportunities.
2. Quality of decisions: The better the decisions, the more effective the outcomes.

However, several obstacles often impede these factors:

• Information bottlenecks: Crucial information gets trapped in individual inboxes, making it the wrong information in the wrong place at the wrong time.
• Fear of mistakes: Individuals afraid of making errors due to fear of not having the information required may delay decisions, further slowing down processes.

In this blog post, I’ll focus on speed of decision making. Decision quality is heavily influenced by the availability of necessary information, but deserves its own series of posts.

How do you make decisions faster?

1. Get the decision to the person who needs to make it immediately.
2. Give them all the surrounding context to mitigate their uncertainty.
3. Create a culture that values speed of decision making, accepting some decisions must be undone.

As a founder, you can probably relate to these scenarios:

• Delayed talent referral: Someone seeking a new role emails me. It takes a week for me to forward the email to our talent team, even though there’s an immediate and high priority need in a portfolio company I’m unaware of. By the time I do, that person has already signed up with a high-flying company and is no longer available.
• Missed investor connections: A portfolio CEO asks if we know a particular investor that reached out to them. Not only is one of my partners familiar, they have also helped the specific investor in question and thinks there is a great investor/founder fit. The connection exists, but the information isn’t promptly shared. The founder deferred the meeting to their CFO. No second chance to make a first impression.

These are examples from day-to-day life of a venture capital firm. One might ask whether it makes sense to give up on privacy for such “banal” cases. My answer is absolutely yes. Getting the right talent to the right company at the right time is probably the most important thing a VC can do to influence the success of a portfolio company. Helping make sure the right investors join the cap table is probably the second most important action we could take.

What’s common in these cases?

Someone else at Aleph could have handled the message more efficiently than I did. Either I didn’t get to it in time, or I wasn’t aware that someone else was better suited to handle it. Routing alerts to our talent team and deal flow pods allows them to support talent recruiting and investor referencing without me needing to actively ask them, or even be aware (!).

To eliminate these wait states, we needed a solution that ensured information would be routed to the right people without unnecessary delays. Our answer was to leverage LLMs to democratize access to my inbox.

Everyone at Aleph can now prompt my email. They get alerted when their custom prompt matches an email in my inbox.

Let’s say I get an email from a friend recommending top talent that’s just about to leave Meta and would like to jump on a call to discuss which portfolio companies I’d recommend they check out.

• Any team member at Aleph can set up prompts against my email using a Zapier interface-based editor. We use Zapier quite extensively within Aleph, and Zapier Interfaces provide for quick-to-build, simple user interfaces such as this one. This system is entirely under their control, so I don’t need to know what they’re interested in. In this case, Bar of the Talent Pod has set the system to alert her if a talent-related email arrives in my inbox.

• Positive & negative indicators: Subscribers set positive and negative indications to refine which emails are of interest. For example, the Talent Team may choose positive indicators like “hire,” “referral,” etc. Negative indicators are useful to filter out emails that match the positive indicators but that wouldn’t be relevant, like a newsletter or a talent-related conference.
• AI processing: Each incoming email is sent to GPT-4 alongside a prompt created from these indicators. We use gpt-4o-2024–08–06 as it supports Structured Outputs, which makes the JSON output far less error-prone.
• Approval process: A configured alert goes through an approval process, where the initial matching emails are manually reviewed before they are automatically forwarded. This ensures privacy and security are maintained. More on this below.
• Relevance filtering: If the AI deems the email relevant, it gets forwarded to the subscriber over Slack. One cool recent addition was that following a thumbs-down emoji, we forward subsequent prompts to the email router. Through this, subscribers can have the system adjust the positive and negative indicators, or include different elements in the forwarded summary.

• Public collaboration: The forwarded messages go to a shared Slack channel, allowing us to continue discussions publicly and collaboratively.

One significant hurdle in many organizations is the political power that comes from keeping knowledge to oneself. Information hoarding can create silos, slow down processes, and breed mistrust.

At Aleph, we operate on complete trust and an extreme transparency policy. By making information accessible and allowing everyone to participate, we dismantle traditional power structures that rely on information gatekeeping.

We do understand the sensitivity of accessing emails and the importance of confidentiality. That’s why any configured alert undergoes an approval process. Initial matching emails are manually reviewed before automatic forwarding is enabled. This safeguards sensitive information while still allowing for efficient information flow.

One more subtle issue in implementing such a system is the culture component of velocity, and that leaders have to make it clear how much they value velocity by accepting that quality may go down when they democratize decision-making within the organization. Otherwise if a leader just says, “Let’s make better decisions, faster,” it’s as if they’re saying nothing. I love the story about Steven Sinofsky instituting a “same-day decision” policy at Microsoft after the Vista debacle, and the resulting snail pace of making decisions. (Side note: I couldn’t find a reference to it when researching. If you have one, please comment!)

This system doesn’t make decisions for us, but it ensures that the right information reaches the right people promptly. By eliminating wait states, we’re able to:

• Accelerate decision-making: Decisions are made faster because the information is readily available to those who need it.
• Improve decision quality: With better access to relevant information, the quality of decisions improves.
• Foster a collaborative culture: Public discussions in shared Slack channels encourage transparency and collective problem-solving.

Eliminating wait states isn’t just about implementing new technology; it’s about shifting culture. By trusting our team and embracing transparency, we’ve created an environment where information flows freely, decisions are swift, and everyone is empowered to act. The political power of keeping knowledge to oneself diminishes, and the organization moves forward with greater agility and cohesion.

• Empower your team: Allow team members to set up custom prompts and alerts based on their needs.
• Leverage AI thoughtfully: Use AI to facilitate information flow, not replace human decision-making.
• Maintain privacy and security: Implement approval processes to safeguard sensitive information.
• Cultivate transparency: Open communication channels break down silos and reduce wait times.
• Foster collaboration: Shared platforms encourage collective problem-solving and innovation.

If you’re interested in eliminating wait states in your organization, think about leveraging AI to improve information flow. I’m considering open-sourcing our prompts and workflows in the future. If you’re interested in getting a copy or learning more about how we’ve implemented this system, feel free to reach out to me.