You Can’t Buy One M&M

It just doesn’t make sense to package, ship and stock a single M&M, sheet of paper, day of insurance coverage, etc. The constraints of efficient batch sizes shape many of our commercial experiences. It’s why we see volume discounts, value-packs, buy-one-get-one and on, and on. Experts call the net result of all those intersecting optimization functions the Economic Order Quantity (EOQ).

Technology and the internet have brought many economic benefits, but some big ones come down to filling in gaps in the EOQ function. When it comes to expertise, Upwork and other online marketplaces let you buy design, marketing, or coding not by the FTE, but by the batch of hours or project. Google provides bites of expert information—if you can find, ingest and quality-control articles, videos, and more. And, only if the answers are publicly available.

Subject-Matter Expert Bots

With the rise of AI Subject-Matter Experts (AI SMEs), everyone will have access to small bites of knowhow. There are already many emerging services that fit that bill, from light GPT wrappers that help you write (e.g. Writesonic) to deeper solutions that review new software code (e.g. CodeRabbit) or analyze legal documents (e.g. Pincities).

Disruption is coming to enterprise functions built on the fractional expertise model to say nothing of the $7T global professional services sector. But rather than focusing on displacement, we believe pay-by-the-drink, trustworthy AI SMEs will result in a massive increase in the demand for expert advice and services.

When phones with capable cameras landed in everyone’s hands, we took 4x more pictures. It will be similar with SME AIs. Here are some reasons why:

Massively lower evaluation costs

Most of the challenge in hiring an expert as an FTE, contractor, service provider, or from a marketplace like Upwork is that it takes a lot of time to assess quality. Obviously, that’s because even professional service firms with good reputations employ individuals with a wide range of skill-levels. To say nothing of the cost of determining the quality of a so-called expert on a marketplace or a pile of resumes for a full-time hire. Even if there are a handful of SME AIs in a given area, they will have trial packages and reviews. And their performance will be consistent every time.

No ego risk, no dumb questions

There are many times we all have avoided asking for an expert opinion in part because to do so might expose an area of ignorance. This is especially true in areas of expertise we are supposed to possess. An AI is just not going to tell our bosses that we asked a dumb question.

Anyone can buy

Pricing for SME AIs will vary widely, but the mere fact that engaging an expert won’t require a purchase order will mean that more junior people will have access. All levels of the organization will benefit from the ability to engage a specialist.Subject-Matter Expert Bots

Herding Risk

In a world where there are reviews for AI SME services, we can anticipate that quality and consistency of the value they offer will be at a premium. No AI founder will want screenshots of dubious interactions with their SME bot to show up in a G2 review or social media post. For this reason, we can expect that the answers one receives from AI SMEs will have less variability and be more risk-averse than answers from humans might be. This effect is called herding—when experts give a narrow range of answers because to do otherwise would carry reputational risks. This same mechanism will ensure consistent quality but will tend to underperform on creativity and innovation in some predictable ways..

Who Wins?

We continue to firmly believe that winners in AI will have access to unique training datasets (check out our Intelligence Islands thesis). There are many areas where SME AIs have massive market opportunities but the scope for unique data inputs may be relatively-limited. A proxy for this is already seen when you use ChatGPT. If you ask it to do things like write a chunk of code, or a blog post, it can produce quality outputs in part because of the prevalence of robust public training data. For this reason, the number of companies trying to provide expertise in those areas will be enormous. There will still be large winners there, but it will be hard to pick them.

One clear way to build defensibility and increase the expertise of a given AI solution is to gather unique, non-public data in the natural course of delivering product value. That’s one reason why we think first-mover advantage may be more important in this space. Start-ups that can accumulate more users and more relevant information will have a bigger data moat.

Many areas of expertise require deep knowledge of a customer’s context. This can be easy or hard to provide. Giving an AI SME access to internal data assets can be both critical to delivering the service and a source of competitive advantage for the provider; however, if doing so requires the client to do involved integrations, data processing, or other hurdles, that will slow down the purchase process for an AI SME offering.

Conclusion:

The rise of SME AIs opens up vast new markets for expertise in addition to the ocean of economic opportunity already exploited by human experts. Quality issues must be resolved and these robots may provide overly-cautious or banal answers. Even so, we predict that massive companies will be created by unlocking the insights trapped in unique training datasets. Early entrants with abundant data will have a competitive edge. Employees at every level will have the expert help and guidance needed to deliver new levels of productivity and output.