AI agents don’t just help people do the work anymore. They do the work.

Everyone keeps asking the same question: will AI-native startups beat the incumbents that are adding AI? It’s the wrong frame. This isn’t a clean product versus product fight. What’s actually happening is a land grab for services revenue, and AI agents are the first thing that makes that land grab executable at scale.

Software took the clean work. Agents take the rest.

Software already took the clean parts of business. Anything structured, repeatable, and easy to systematize got turned into SaaS over the last two decades. What remained was everything that didn’t fit neatly into software: messy workflows, human judgment, document-heavy processes, compliance, and the kind of work that usually gets described as “it depends.” Entire industries exist because software couldn’t fully replace the human in the loop. Consulting, audit, agencies, outsourcing. That constraint is now gone. AI agents don’t just help people do the work anymore. They do the work.

This isn’t another wave of SaaS. SaaS scaled by selling tools to humans. Agents scale by removing the human from the loop entirely. That changes the economic model. Pricing shifts from seats to outcomes. Value shifts from having a good interface to actually executing the work. Defensibility isn’t in features, it’s in controlling the workflow and the data that comes with it. You’re not selling into a services company anymore. You’re competing with it.

AI-native versus incumbent is the wrong comparison

Incumbents do have real advantages: distribution, trust, regulatory positioning, and existing revenue. They’re also structurally constrained. AI-native companies don’t carry that baggage. They can rebuild workflows from scratch and aggressively remove labor because they don’t depend on it for margin.

This doesn’t play out as a feature comparison. It plays out as cost structure versus output. If a new entrant delivers most of the outcome at a fraction of the cost and does it faster, it doesn’t need to win the entire customer relationship. It only needs to take enough to break the incumbent’s margins. Once margins compress, the model becomes fragile.

Incumbents will respond by layering AI into what they already do. They’ll improve internal efficiency, quietly reduce headcount, and reposition their services as AI-enabled. What they won’t do, at least not fast enough, is fully cannibalize their own business. Their economics are still tied to billable hours and headcount. AI-native companies are built on the opposite incentive: compress labor, automate aggressively, and price on outcomes. That asymmetry matters more than any individual product decision.

Every AI-native startup is claiming a slice of the same market

Right now, nearly every AI-native startup is doing the same thing: picking a slice of services revenue and going after it. Audit, legal, accounting, agencies, recruiting. These are large, fragmented, inefficient, labor-heavy markets, and that combination makes them highly exposed.

The interesting part is that this won’t produce a single dominant winner in each category. The barrier to building these companies is lower than it was in previous cycles. They ship faster, require less capital early, and carry a very clear ROI narrative. So you don’t get one or two winners.

You get dozens, each taking a small piece of the same market.

That is going to change the investment landscape.

Fragmentation leads to aggregation

Customers don’t want a dozen narrow tools stitched together to run a single workflow. They want a system that handles the job end to end. So you end up with a layer of AI-native companies, each owning a slice, and then consolidation begins.

This won’t look like classic SaaS consolidation. It will look more like roll-ups. Private equity and later-stage platforms will stitch these pieces together, centralize distribution, and expand margins through integration. Value shifts away from any single tool and into the aggregated system.

What this means for investors

From an investment perspective, this is a very different shape than the last cycle. For venture, it means a high volume of opportunities with fast feedback loops and clear monetization paths. For private equity, it’s close to ideal: fragmented supply, proven revenue, and obvious paths to consolidation. But it likely produces fewer massive standalone public companies than people expect. Not because the market is smaller, but because value gets captured earlier and more incrementally through aggregation rather than through a single company scaling to dominance.

What wins in this environment isn’t the company with the best model or the most impressive demo. It’s the one that embeds itself in the workflow, controls the data loop, and consistently delivers a measurable outcome that replaces a real cost center. From there, the path is either scaling distribution aggressively or positioning as a critical piece of a larger platform that eventually gets rolled up.

So the real question isn’t whether AI-native startups will beat incumbents. It’s how quickly they can carve up services revenue before incumbents structurally adapt, and who ends up owning the layer that ties all of these capabilities together. That’s where the long-term value will sit.

Everyone has been chasing scale for the last few years: more data, more parameters, more compute. Something fundamental is changing. The real paradigm shift isn’t about building the biggest model possible anymore. It’s about building the right model for the job.

From “bigger is better” to “smarter is better”

For the past five years, the entire industry operated on a single narrative: bigger models mean better results. That was true for a while. GPT-4, Gemini, Claude, and LLaMA are all incredible feats of engineering and scale. They also brought trade-offs: astronomical compute costs, limited deployability, latency issues, and privacy concerns that made real-world enterprise adoption complicated.

Now the pendulum is swinging the other way. The next era of AI is being defined by right-sized intelligence: smaller, specialized models that are faster, cheaper, more efficient, and actually usable in production. These models don’t try to know everything. They’re built to do one thing really well.

Why small language models are suddenly the smartest move

1. Efficiency changes the economics.
Running a 175-billion-parameter model for every query isn’t just overkill, it’s bad business. SLMs operate at a fraction of the cost, with dramatically lower compute and memory requirements. That opens the door for startups, small teams, and even individuals to deploy serious AI capabilities without enterprise-scale budgets. The inference cost curve is flattening, and that’s what democratization actually looks like.

2. Domain fit beats general intelligence.
You don’t need a generalist model trained on all of Reddit to run an underwriting engine, a contract analyzer, or a support triage bot. You need a focused model, trained or fine-tuned on the data that matters. That’s where SLMs shine. They’re easier to specialize, faster to retrain, and far more adaptable to niche contexts. They also hallucinate less, because they’re not pretending to know everything. They’re scoped, efficient, and purpose-built.

3. Privacy and deployability matter more than scale.
We’ve all had the conversation with legal and compliance about where data lives, who sees it, and how to control it. The ability to self-host or run on-prem fundamentally changes that conversation. SLMs make it possible to deploy models behind a firewall, on devices, or at the edge, without sending data back to a black box API. That isn’t just good architecture. That’s trustable AI.

4. Sustainability is no longer optional.
Training massive LLMs consumes absurd amounts of energy, and the environmental and economic footprint is staggering. SLMs flip that narrative: less energy, less compute, lower cost per inference. They align technical innovation with environmental responsibility, and that alignment is going to matter more every year.

This isn’t a smaller model story, it’s a systems story

The real shift isn’t about model size at all. It’s about architecture. We’re moving from monolithic systems, one giant general model doing everything, to modular ecosystems of specialized models, each designed to excel at one domain and orchestrated together through smart routing, context sharing, and retrieval layers.

In other words, instead of “one big brain,” we’re heading toward networks of narrow intelligences.

That’s how humans work. That’s how organizations work. And now that’s how AI will work.

Why this is a paradigm shift, not a passing phase

Every major technology shift starts when the narrative around “more” flips to “enough.” We saw it in chips, moving from clock speed to efficiency, and in the web, moving from heavy pages to lightweight frameworks. Now we’re seeing it in AI. The move toward SLMs is that same inflection point, going from raw power to practical performance.

It’s a shift that lands on three dimensions:

• Accessibility: anyone can deploy or fine-tune an SLM. You don’t need a data center or a billion-dollar contract.
• Governance: data stays where it belongs. Enterprises regain control over how AI learns and operates.
• Strategy: AI becomes something you build into your stack, not something you rent from someone else’s cloud.

What SLMs mean for builders and investors

For product teams, this means rethinking architecture. Ask whether a feature really needs a massive foundation model, or whether a small, specialized model could deliver the same result faster and cheaper.

For founders and operators, it’s a distribution opportunity. You can now embed real intelligence into your product without killing your margins.

For investors, it’s signal detection. The next generation of breakout AI companies won’t just build LLM wrappers. They’ll own the SLM layer: fine-tuned vertical models, domain-specific inference stacks, or orchestration frameworks that blend models intelligently.

Small models don’t replace large ones entirely. There’s still a need for high-capacity reasoning, general world knowledge, and creativity, the heavy lifting that big models are built for. Even there, the future looks hybrid: SLMs for context and precision, LLMs for reasoning and synthesis.

Think of it like the evolution of computing itself. We didn’t stop building supercomputers, we just stopped expecting every problem to need one. Most of the world runs on smaller, distributed, purpose-built systems, and AI is finally catching up to that logic.

Structured data, meaning numerical tables, has traditionally been the focus of data governance. The rise of LLMs has pushed unstructured data to the front. Text documents, videos, images, and audio recordings are what power most AI applications worth building: chatbots, smart knowledge assistants, and content generation tools. Despite that, unstructured data has never received serious governance attention.

The typical scenario in most companies is an overwhelming amount of unstructured data scattered across platforms and systems with no systematic management behind it. That landscape creates three problems you have to solve before AI is reliable: data relevance, data quality, and data safety.

The three problems blocking reliable enterprise AI

Data relevance. LLMs are capable, and they still need extensive guidance to filter through large document sets and identify the most pertinent sources to extract information from. An insurance company running a knowledge assistant trained on a vast array of policies has to make sure every response is contextually accurate and relevant to the policy actually in question.

Data quality. Quality in the world of unstructured data is close to uncharted. Traditional methods of evaluating tabular data for outliers, freshness, and completeness don’t transfer. The problems that show up instead are inconsistent naming conventions, conflicting information, and data that is outdated or about to be.

Data safety. Protecting sensitive information, from personally identifiable information to proprietary data, is critical, and regulations like GDPR raise the stakes. Inadvertently including a customer’s personal data in a training set can mean deleting the entire model when that customer files a removal request. Internal access controls are their own challenge, particularly for knowledge assistants and chatbots that will happily surface whatever they can reach.

What businesses should be doing about it

Adopting LLMs well requires a strategic approach to managing unstructured data. Five things matter most.

Build a real data governance framework.
Set standards for data quality, security, and usability, with policies and procedures for how data is collected, stored, processed, and shared. Include guidelines for maintaining data integrity, accuracy, and relevance, especially across diverse and constantly changing unstructured sources.

Invest in data processing and curation.
Given how varied unstructured data is, you need real tooling: natural language processing to extract meaning from text, image and video analysis for visual content, and audio processing for sound. Curation is the part people skip. Feeding only relevant, high-quality data into your models is the cheapest available improvement to output accuracy.

Emphasize relevance and context.
Collecting and processing large volumes of unstructured data isn’t the goal. Aligning that data with the specific use case matters more. Tailoring data to the business problem improves model effectiveness and reduces irrelevant or inaccurate outputs.

Strengthen security and privacy.
Prioritize encryption, access controls, and compliance with data protection regulations like GDPR and CCPA. Anonymizing and pseudonymizing data where you can protects individual privacy while still letting models extract the insight you’re after.

Plan for continuous adaptation.
AI moves fast. Update your data management strategy regularly, stay current on developments in AI and machine learning, and adapt your processes and systems as the tooling changes.

Handle these and you create fertile ground for LLMs to work, which is what leads to AI applications that are genuinely innovative rather than demos. This approach to unstructured data isn’t only about harnessing what AI can do today. It’s about future-proofing the business against a technology landscape that keeps moving. I wrote more about where this is heading in the future of enterprise data governance in the age of AI and LLMs.

Below is how large language models are altering the dependence on experts, transforming governance itself, and changing what these projects cost to run.

Less reliance on experts and functional requirements

Enterprise data migration projects have traditionally leaned heavily on subject matter experts to interpret source data, define functional requirements, and lead the project. LLMs are changing that. These models can interpret source data on their own, which makes migration more streamlined and less expensive. Using an LLM to grasp the context and meaning of data cuts the time spent on manual analysis and the need for exhaustive functional requirements. Companies get to concentrate on strategic decisions while the model handles interpretation and migration.

Automated data transformation code

One of the clearest benefits of LLMs in enterprise data migration is their ability to write complex data transformation code. A model can analyze source data alongside target system requirements and produce the code needed to migrate and transform it, which removes most of the manual coding and testing. That makes development and testing faster, and it makes migration projects far more adaptable when requirements move.

Data governance shifts from rules to inference

Data governance has historically rested on data stewards who wrote definitions and set rules for how data gets used. LLMs shift governance from a manual, rule-based discipline to one driven by inference of process and data. Models analyze existing datasets to infer processes and relationships, which removes much of the manual work of creating definitions and policies.

That lightens the load on data stewards, and it also makes governance more consistent across an enterprise. With inference doing the heavy lifting, governance can be proactive and adaptive rather than a set of documents nobody updates.

What this does to the software and services market

Wider adoption of LLMs will reshape both the software and the services markets. Demand will grow for software built specifically for data migration and governance tasks. On the services side, businesses need new kinds of expertise to implement and manage AI-driven solutions, which means traditional service providers have to modify their offerings and invest in new skills to stay competitive.

The effect on project costs and ongoing expenses

Bringing LLMs into data migration and governance changes both project costs and run-rate expenses. Automating interpretation and code generation shortens timelines and reduces manual labor, which lowers overall project cost. More efficient governance processes then lower the ongoing expense of managing data.

Where this leaves data teams

LLMs are reshaping enterprise data migration and governance by reducing dependence on subject matter experts and formal requirements, automating code generation, and changing how governance itself works. The software and services markets have to adapt to that, which means adopting the technology and investing in the skills to run it. For the governance work that has to happen before any of this pays off, see how data governance shapes the future of AI applications.

Writing a good email about bad things is hard, so here’s the template I use for a complaint email that gets you the result you want while keeping your professionalism intact.

The five-part complaint email template

1. State the problem.
2. Explain how you feel about the problem.
3. Explain how you’ve tried to solve the problem.
4. Ask for help, and offer your own ideas for solving it.
5. State what you hope will happen, and what success looks like.

An example complaint email

Say you manage accounts for a company. Something changes and suddenly you have three times as many accounts. A job that was manageable becomes unbearable, you’re barely sleeping, and you’re on the verge of burnout.

State the problem.

Dear Manager,

In the latest restructure I ended up with three times the number of accounts to manage, and I’m working ten and twelve hour days and losing sleep. I’m extremely stressed and frustrated, and I’m falling behind to the point where tasks aren’t getting done and I’m letting down my co-workers.

How you feel about the problem.

Being unable to keep up with my work makes me feel awful and defeated, and I’m worried it will damage my relationships with my co-workers.

How you’ve tried to solve the problem.

I’ve tried posting my daily to-do list in the standup channel in Slack, but it hasn’t made a difference.

Ask for help, and offer ideas.

Could you help me find a solution? I have a few ideas: (1) moving some accounts off my desk, (2) building an interface so stakeholders can track statuses without emailing or Slacking me for updates, and (3) getting an assistant to help me process certain items.

What you want to happen, and what success looks like.

My goal is to make my job manageable so I don’t burn out, and to make sure I continue to be seen as reliable by my co-workers.

Why this structure works

This template scales to almost any level of severity, and a few things about its construction matter.

The biggest one is that you’re coming to the table with solutions, not just problems. Managers have a lot of people to manage, and their job is to solve problems and make their team successful. Employees who don’t participate in solving their own problems generally don’t get considered for a management track, and at the extreme end they can be treated as a nuisance when they surface too many problems, even warranted ones.

So what do you do if you don’t know how to solve the problem? Ask mentors or peers for advice first. If that isn’t available and you genuinely don’t know what to do, be candid: say you’ve tried to come up with solutions and you’re unsure. Telling your manager what you’ve already tried matters, and good managers read that vulnerability as a strength. It also builds trust and rapport, which is a good way to build a real relationship with the people who can affect your career.

What happens after you send it

Usually your manager will set up a meeting to discuss the complaint. Avoid blaming statements like “you” and “they” when you explain the situation, because that creates an us-versus-them dynamic. What you want is a collaborative space where it’s us versus the problem. Explain things in terms of how you experienced them. People are more sympathetic to feedback when they can see you’re not assigning blame. You attract more flies with honey than with vinegar.

One last thing about the closing section, the part where you say what you want to happen. It isn’t a list of demands. It helps your manager see what you think success looks like once the problem is solved, and it gives you both a measure for evaluating whatever solution you put in place.

Hopefully this gives you a clear path forward, and helps you turn a tough situation into a better one.

When it comes to modeling dilution and pro rata rights, a simple calculation lets you build out your current and future financing scenarios and see how ownership plays out over the long run.

You can access my cap table calculator here to download and follow along with this article, or use for your own purposes. If you have questions or need help, comment, email me, or find me on Twitter at @chloealpert.

Start at incorporation

That’s a standard incorporation table. Most companies incorporate a C corporation with 10M shares, though you don’t have to. You can incorporate with any number you want, and round numbers make the math easier. You also set a par value for your shares, which can be as low as you like, usually $0.0001 or some variation, so you can purchase your stock for next to nothing (because it’s currently worth nothing) and file your 83(b) election.

Modeling the seed round

Now assume you go out and raise $1M on a $5M pre-money valuation. Here’s the table.

The seed investor took 20% of the round, but their ownership on a post-money basis is actually 16% once you account for the total number of shares between the founders and the employee option pool.

Modeling the Series A

Now do a Series A where the same company raises $7M on a $25M pre-money valuation. The Series A investors take 28% of the round, and modeled out on a post-money basis, that investor ends up with 21.8% of the company.

How pro rata rights work

Seed and Series A investors generally have ownership targets to maintain. So let’s do a Series B where the Series A investor exercises their pro rata rights to hold their percentage.

Translated literally, pro rata means “according to the rate.” The way to think about it is the right to maintain a proportional ownership percentage of the company.

“You invest $50k in a seed round at a $5mm cap and own 1% of the company. The next round is a $3mm round at $9mm pre, $12mm post. If you don’t participate, you will be diluted 25% and will then own 0.75% of the company. On the other hand, if you buy 1% of the round, a $30k investment, you will continue to own 1% of the company. Your ‘pro-rata right’ in this situation is a $30k allocation in the next round.”
Via Jason Rowley

The math for a pro rata amount is simply (target ownership %) x (number of new shares being issued) x (share price at the new round). That’s reflected below. The original Series A investment gets diluted the same as everyone else, and the new money invested as pro rata brings the investor back to their ownership target.

Dilution versus the step up in share price

A few things to notice in this Series B. Even though the founders have been diluted below 50% ownership, the 4.6x step up in valuation means founder equity is now worth over $83M on the same number of shares. When you raise money, you have to understand the balance between dilution and step ups in share value. You can’t always optimize for dilution. If you truly care about maintaining ownership of your company, you probably shouldn’t take venture capital in the first place.

That aside, you can see the Series A investor put in additional capital to execute their pro rata right and hold close to a 20% ownership target. They most likely won’t participate beyond the B, and that decision comes down to their business model and fund mandate. I wrote about why fund mandates work that way in why venture capital firms with tons of money can’t write little checks.

When founders have to limit pro rata

Not every investor executes their pro rata rights, and if you let too many of them do it, there isn’t enough room to complete the round without losing your shirt. Investors understand that founders who are over-diluted early lack the incentive to keep operating. So even with math driving the valuation, founders sometimes have no choice but to block or limit the pro rata amounts of existing investors. That can produce hurt feelings and some angry words, and it’s one of the genuinely unpleasant parts of raising money, especially as you get bigger and start making real revenue.

Series C and the case for a secondary sale

To close it out, let’s take the company through a Series C and hit unicorn status on a post-money basis. The founders still own quite a lot, and what founders typically do at this round, and often at the B, is a secondary sale.

Investors push for founders to sell a small piece of their private stock to give them some initial liquidity for their time. The theory is that a founder with a little liquidity is less risk-averse about pushing the company to go big. It also frees up shares for existing investors who want to expand their ownership before an IPO.

All in, this is a quick simplification of a genuinely complex topic, but it should give first-time founders a sense of how to model ownership and investment in their companies and plan ahead.

At later stages, specifically once you’re doing a priced round, you’ll want legal counsel managing your data room, so this advice mostly applies to pre-seed through Series A. Anything after that takes a different form, and you also need a proper CFO with investment banking experience if you don’t have that background yourself.

There’s a camp that thinks this level of preparation at seed, when you’re raising on SAFEs or convertible notes, is excessive. They’re not wrong. You can close your round without this level of granularity. I subscribe to the view that showing this level of sophistication at the start of the financing journey signals to investors that you can execute the later rounds, where the proverbial shit gets real. Seed through A gets raised on vision, early traction, and team. At Series B and beyond, the rubber hits the road and it’s about metrics more than anything else.

The standard diligence checklist is below. Keep in mind that early-stage companies definitely won’t have everything on this list. If you’ve never issued stock to a non-founder or set up an employee equity incentive plan, you probably haven’t completed a 409a valuation and won’t have those documents. That’s fine.

Each heading below should be a folder in your data room, and the bullet points are the contents of that folder.

Corporate records and charter documents

• All minutes of directors’ and stockholders’ meetings, and all written consents of directors and stockholders.
• Certificate of Incorporation, Certificates of Designation, Rights, and Bylaws.
• A list of each state where the company does business, specifying whether it is qualified to do business in that state.

Business plan and financials

• Current business plan and any financial projections.
  • At an early stage this is a deck. You don’t need a formal written business plan.
  • In earlier stages, treat your financial projections as a use of proceeds and a budget, especially with no or limited revenue.
• Most recent financial statements.
  • The more the merrier here.

Security issuances and agreements concerning securities

• A list of the company’s stockholders, including issuance dates and original issuance price. This is your cap table.
• A chronological list of option grants including grant date and exercise price.
• Copies of agreements relating to outstanding options, warrants, and rights (including conversion or preemptive rights), agreements for the purchase or acquisition of any of the company’s securities, and agreements relating to past stock issuances.
• The company’s stock plan and all related form documents, plus any documents evidencing registration rights, agreements among stockholders, or agreements between the company and its stockholders.
• A summary of the vesting schedules of any stock or options subject to vesting, including any vesting acceleration.
• Agreements relating to voting of securities and restrictive share transfers.
• Evidence of qualification or exemption under applicable federal and state blue sky laws for issuance or transfer of the company’s securities.
• Copies of Internal Revenue Code Section 409A valuation reports.

Intellectual property

• A list of the company’s trademarks, patents, copyrights, and domain names, or any applications for them, including documentation of filing or registration with the appropriate government entities.
• Any documentation relating to the transfer of technology to the company or to any employee.
• Copies of the proprietary information and invention agreements signed by any service provider, including employees and consultants.
• A list of any employees or consultants who have not signed proprietary information and invention agreements, including any periods when they performed services for the company while not bound by such agreements.
• Any correspondence or documents relating to allegations that the company infringed the proprietary rights of others, or allegations by the company that its own rights were infringed.
• Copies of all material agreements licensing company technology to third parties, including cross licenses.
• Copies of all material agreements licensing technology from third parties.
• A list of all third-party software, including open source, and any derivatives used with or integrated into software the company distributes or hosts. List applicable licenses and licensors for each, and describe the communication and linking between that third-party software and the company’s proprietary software.

Material agreements

• Any agreements, understandings, instruments, contracts, or proposed transactions the company is party to or bound by that involve obligations of, or payments to, the company in excess of $25,000.
• Any personal property leases.
• Any agreements concerning the purchase, lease, or sublease of real property.
• Any documents evidencing indebtedness for money borrowed or other liabilities incurred by the company.
• Any documents evidencing mortgages, liens, loans, and encumbrances on company property or assets.
• Any documents evidencing loans or advances made by the company, including loans made to employees for any reason.
• Any agreements, understandings, or proposed transactions between the company and any of its officers, directors, or affiliates, including non-competition agreements, employment agreements, and non-form offer letters.
• Any licenses or agreements concerning the company’s or others’ patent, copyright, trade secret, or other proprietary rights, proprietary information, or technology, including employee confidentiality agreements.
• Any insurance policies held by the company or naming it as beneficiary, and a summary of those policies if available.
• Any judgment, order, writ, or decree binding the company or to which it is a party.
• Any standard forms of agreement used by the company.
• Any joint venture or partnership agreements.
• Any management, service, or marketing agreements.
• Any confidentiality or nondisclosure agreements.
• Any agreements requiring consents or approvals in connection with the financing.
• Any documents containing severance payments or acceleration of stock or option vesting.
• Any consulting contracts.
• Any other agreements material to the business, or outside the ordinary course of business.
• A list of officers and directors. If any officers are not devoting 100 percent of their business time to the company, note that here.
• A list of all acquisitions, dispositions, mergers, consolidations, and reorganizations, with all related documents, plus a description of any plans for events of this nature.
• Any product or service warranties or indemnities.

Disputes and potential litigation

• Any correspondence or documents relating to a pending or threatened action, suit, proceeding, or investigation, including those involving employees in connection with their prior or present employment or use of technology.
• Any correspondence or documents relating to allegations that the company infringed the proprietary rights of others.
• Any correspondence or documents relating to labor agreements or actions, union representation, strikes, or other labor disputes.
• A schedule of settled or concluded litigation, claims, suits, and proceedings, along with related consent decrees, judgments, orders, settlement agreements, and injunctions.
• A list of all claims made under any D&O policy or other insurance policy covering the company.
• Notices of breach or default under any material agreement.

Employees and employee benefits

• A list of employees and consultants, including title, base salary, target bonus if applicable, commission plan if applicable, classification (for employees, whether exempt or non-exempt), and state of residence.
• The company’s standard form of offer letter or employment agreement.
• Any plans, agreements, or arrangements providing benefits contingent on a change of control.
• Any severance or deferred compensation plans, including salary deferral agreements with employees or consultants, whether oral or written.
• Any employee benefit plan, including stock option plans, 401(k) plans, pension plans, and insurance plans.
• Any forms of agreement used with stock option plans, such as a form of option agreement, notice of exercise, and restricted stock purchase agreement.
• If the company sponsors a 401(k) plan, any determination or opinion letter and Form 5500 filings for the most recent year.
• All documents or information relating to loans made by the company to its employees, directors, or consultants.
• The company’s employee handbook.
• If the company has foreign employees, a list separated by country of all benefits provided to them.

Other information

• Any securities or ownership interests in other companies held by the company.
• All licenses, permits, or government authorizations held by the company or needed to conduct its business.
• A list of any current or past officers, directors, or key employees who hold a direct beneficial interest in any competitor, supplier, or customer, with a description of those interests.
• Any complaints regarding accounting, internal controls over financial reporting, and auditing matters received in the past three fiscal years.
• A description of all material off-balance sheet transactions or arrangements of the company or its subsidiaries, with copies of all relevant documentation.
• A company org chart, with future hires identified.
• Anything else you think is relevant.

Where to host your data room

I’ve used Carta’s data room feature and liked it, mostly for read-only watermarked documents and NDAs required to view. As of today I don’t recommend starting with Carta if you’re not already on it, because of their 2020 pricing structure.

I don’t recommend Google Docs, since I don’t think it’s secure enough. Dropbox is the most reasonable option for seed to Series A founders who aren’t running the data room through outside counsel. I also suggest DocSend for all deck exchanges. Understand that investors will screenshot everything and share it even when you disable downloads, so once one investor sees your deck, it will make the rounds. Just expect that.

What about NDAs?

Unless you have a truly groundbreaking deep tech innovation with unprotected IP, my take is that the vast majority of founders raising pre-seed to seed shouldn’t worry about NDAs for diligence. Check with your outside counsel on this. At seed, you want to be easy to work with, as much as you want your investor to be easy to work with. Series A is where companies do need to consider NDAs before sharing the data room, because by then you usually have confidential contracts you want to share safely.

Thank you to Perkins Coie LLP, who provided the bulk of the diligence checklist guidance.

No. But yes. But really, no.

via GIPHY

A few forces are at play in any venture investment, and they simplify down to (1) time and opportunity cost, (2) prospectuses, mandates, and business models, and (3) firm and investor type. Here’s how each one works.

Time and opportunity cost

This is the easiest one to explain, and I remember a lightbulb going off the first time someone walked me through it. I was 20 years old and genuinely frustrated that a fund with billions under management couldn’t write a $50k check and see what happens.

Some very large funds spin out a smaller vehicle for early-stage investment and even brand it, but then you’ve suddenly got a whole new venture fund, which brings its own constraints (more on that in the next section). The core problem is simpler than that. If you’re a giant fund writing a $50k to $500k check into a small company, you can’t justify spending time on that company when you have $30m in another one. Investors are people with a finite number of hours, so they have to be careful about where those hours go. You also have a fiduciary duty to your limited partners, the people and institutions whose money you’re investing. It isn’t financially responsible to spend time on an investment where so little is at stake when you have 600x that amount in another company.

Beyond partner time, there’s back-office work, meaning the legal work to get the investment done. Every investment carries closing costs, and it’s not unusual for the back-office spend to exceed the check itself. That’s exactly why convertible notes and the YC SAFE became popular. The benefit of the SAFE, or Simple Agreement for Future Equity, is that you skip engaging a legal team to execute a stock purchase agreement and run a full priced round. SAFEs carry other risks, though, because the investor doesn’t technically own the stock at signing.

Prospectuses, mandates, and business models

There are entire books on venture capital business models, so I won’t go deep here, just into the parts that matter for this question.

Venture funds have to raise money too, and to do that they need a business model with a mandate and a prospectus established for each fund. One firm can and usually does invest out of multiple funds, and the partners hold a financial interest in those funds. Generally a single fund, meaning a single bucket of money, runs a 60/40 split: roughly 60% of available capital goes to new investments and the remaining 40% goes to pro rata, meaning follow-on investments in later rounds of existing portfolio companies to maintain a target ownership percentage.

Side note: if you want to go deeper on pro rata rights, read my post on how to model ownership and dilution and how pro rata works.

When an investor raises a fund, they file a prospectus with the SEC detailing the investment security and the offering, meaning how the fund will function and how investors make money. Funds also establish a mandate, which determines how the money gets invested and drives the decisions of the investment committee.

Here’s a concrete example. In 2020, a Series A firm might have a business model that mandates a 15% to 20% ownership target with an average check size of $5m to $6m, investing out of a $300m fund, with up to $10m total into a company over the life of the investment. If you’re raising a Series A at a $60m pre-money, that investor is priced out of your round. They can barely get to 10% ownership, which is too little for their return math and outside their mandate.

Flip it around. If you’re raising $1m on a $5m valuation, they could easily hit a 20% ownership target, but now you’re back to the time and opportunity cost problem, plus the mandate that the LPs actually bought into. Investing outside your mandate creates legal exposure with your LPs, because partners have a fiduciary duty and that duty is about risk management and returns. A seed investment carries a different risk profile than a Series A.

Some funds have a single limited partner that is itself a business, investing off that company’s balance sheet. Amex Ventures is a good example. They’re less valuation sensitive and do carry ownership targets, but their mandate requires the company to hold specific strategic value to Amex. Other investors have multiple LPs, like private family offices, university endowments, or pension funds, and those funds tend to have less flexibility because it’s truly other people’s money. Mandates can also specify a vertical focus, like marketplaces or B2B SaaS.

One interesting exception: some funds write carve-outs into their prospectus or mandate for accelerator programs like YC or 500 Startups. Instead of the typical $5m to $6m check at a set ownership target, they set aside a percentage of the fund for $100k to $200k seed checks into YC or 500 companies only. That’s one reason to do one of those programs. Partners generally won’t manage those investments, and the relationships get handled by associates or principals.

The other exception is an established fund writing a small check into a company they intend to write a $5m to $6m check into later, usually because they have high conviction in a second-time founder they’ve worked with before, or they’re seeing exceptional early traction or strategic value. That situation often comes with term sheet language guaranteeing pro rata rights, or aggressive liquidation preferences to offset the extra risk the fund is taking.

Firm and investor type

The last thing to consider is the type of firm. An angel or super angel can do whatever they want with their own money, which is why founders get encouraged to go the angel route for their first tranche of capital, or get told they’re raising too little for an institutional investor to care. The challenge is that angels are active sporadically, so finding them or maintaining any meaningful database of them is hard.

Then there are firms and investment managers just starting out, building a track record to show they can get into meaningful deals and make LPs money so they can raise larger funds later. Picture a $10m fund getting into the Series B or C of a brand-name company (think Lyft or Stripe). At that point it isn’t about the dynamics of that specific investment. It’s about establishing a track record and showing you have access to good deals. Access is the biggest challenge in this business, and having the network to avoid getting iced out of growth rounds matters, because that’s where investors have more data to manage risk and make a lot of money.

The questions to ask an investor instead

Hopefully this explains some of what’s driving a large firm not to write small checks. More usefully, it tells you what to ask an investor when you meet them, so you can figure out whether they can invest in you at all:

1. What is your average check size, and what ownership target are you looking for?
2. What stages do you invest across, seed, A, B?
3. Who are your LPs, and is there anything we should know about your investment mandate?
4. What does your investment committee process look like?

The answers give you what you need to build a strategy for working with that investor and closing your round.

Here are the seven we run on.

1. We seek the truth

This stays our number one value because it covers everything we do, from finding the best candidate to deciding what feature to build. We focus on the truth so we can make the right calls instead of giving in to ego or agenda.

2. Be frugal, but not stingy

Spend what you need to do the job right and avoid extravagances. Whether it’s investor money or profit, we don’t believe in unnecessary excess.

3. When it seems impossible, find another way

Startups aren’t easy, and when you’re trailblazing a new path in an industry as difficult as healthcare, it’s easy to get discouraged. There is always another way to get something done, and giving up isn’t an option.

4. Know the mission, know your metric, prioritize

It’s easy to do work that feels productive but doesn’t move the needle. When every team member knows their mission, how it fits into the bigger picture, and how to quantify its impact, they can focus on the work that makes the biggest difference. Focus and execution are what matter.

5. We don’t make excuses

It’s easy to make excuses when you’ve been pushing a rock uphill all day. Success is doing the things you don’t want to do, and we’ve committed to holding ourselves fully accountable for the results we reap.

6. No job is beneath us

We want a service-based culture inside the company. Sometimes that means doing unpleasant tasks so everyone else stays unblocked. From writing code to taking out the trash, we do what it takes to keep moving forward.

7. Don’t be a jerk

You can be innovative, analytical, and efficient without being a jerk. We hire high performers who don’t need to make other people feel small to prove their prowess.

Cloud adoption came quickly to other enterprise industries, including banking and professional services, but healthcare has lagged. It trailed for two reasons: a lack of free cash flow, and the introduction of the Affordable Care Act in 2010, which pushed hospitals from a fee-for-service reimbursement model to a fee-for-value one. That foundational shift in reimbursement was so consuming that it slowed investment in IT innovation for nearly a decade. Nine years later, healthcare is finally catching up on cloud strategy. Three trends are driving it.

1. Millennial consumers are making their tech preferences known

92% of millennials have smartphones, versus just 57% of boomers, and only 58% of millennials say they trust doctors, versus 73% of previous generations. Millennials are changing the care delivery landscape because they prefer to interact with their caregivers through mobile and online interfaces rather than schedule an in-person visit.

The healthcare sector is responding. Companies building new care delivery models, including One Medical and Forward, are heavily focused on remote care and mobile technology. Kaiser Permanente now offers care by telephone and email, with scheduling through its mobile app. Other large healthcare organizations will follow, having finally recognized the need to invest in cloud-based healthcare.

2. Changing patient demographics are forcing hospitals to get efficient

By 2030 there will be 69.7 million boomers on Medicare, contributing to an estimated shortage of 120,000 doctors. Hospitals will have to get more efficient to serve their patients.

In 2018, the average hospital ran an operating profit of 1.7%, where 2.5% is a more sustainable number. With that little free cash flow, most hospitals can’t simply add staff to handle higher patient volume. So hospitals and healthcare practices are turning to technology to deliver care more efficiently, and to run the functions that support care delivery, like supply chain and materials management, which often account for 20% to 40% of a hospital’s capital budget.

3. A generational turnover in hospital leadership is driving modernization

Boomers started turning 65 in 2011, and eight years on we’re seeing boomer generation executives retire and make way for a younger set of millennial executives. Because the fee-to-value shift sparked by the ACA was such a radical, time-consuming change, outgoing executives never had the room to prioritize investment in technology and cloud-based systems. Their successors are digital natives with both the time and the inclination to jettison old legacy systems.

That generational shift is driving the same provider investment in cloud strategy we’ve already seen in other industries. Investor dollars support the trend too: 2018 brought an all-time high for diagnostics and tools M&A, plus more than $84.3 billion of venture capital investment across all healthcare sectors.

What the next decade of healthcare IT looks like

The next ten years will be about healthcare providers adopting cloud strategies wholesale. Driven by consumer tech demand, generational turnover in executive leadership, and the need to serve an ever-increasing number of patients efficiently, the days of limited, expensive legacy software are numbered. The move to cloud-based solutions will keep expanding and accelerating until it touches every part of healthcare.