The Adoption Of AI And Its Effects On Businesses

A robot’s hand typing on a keyboard This is a really interesting question, and honestly I think that everyone is needing to update their predictions on a regular basis—for example, I would have answered this question somewhat differently a month ago.

At present, however, I have three main thoughts in response: (1) like AI adoption in the past, I expect the surrounding ChatGPT to far exceed the it is actually used in businesses; (2) while only time will tell, it does appear that AI use is recently trending upward at a much quicker rate than in recent years; and (3) contrary to what we have been finding in past years, the most recent measures of AI use seem to imply a relatively large usage rate among very small firms.

If I had only one sentence to respond to this prompt, I would likely say, “It will happen at lower levels than you’re probably expecting.”

This opinion comes largely from my own experience with nationally representative data on firm-level AI use in recent years.

The Census Bureau has begun administering surveys on AI and other advanced technology use among firms beginning in 2018.

I have coauthored multiple research papers using the Annual Business Survey (ABS), which contains both direct and indirect measures (depending on the year) of AI use among businesses.

The 2019 ABS asked firms directly whether they used any AI in their production processes.

The result: only 3.2% of firms reported using AI (see Table 1 ).

We used the 2018 ABS (which has an even larger representative sample) to generate an even more inclusive estimate of AI use and found a rate of 5.8% (see Figure 1 ).

It’s a more generous estimate because we could only AI use based on reported use of technologies that are generally AI intensive but don’t necessarily require AI.

Of course, it is also true that these rates were both measured well before the release of ChatGPT in 2022, but these are relevant because there was still a significant “hype” associated with AI at the time.

For example, surveys by and reported AI usage rates from 25% to 50%—orders of magnitude larger than the nationally representative numbers from the ABS.

Of course, surveys like these don’t have the random selection or sample sizes necessary to realistically estimate a national average like the ABS does.

So, what about more recent measures of AI since the release of ChatGPT?

I’ll re-emphasize my previous point by highlighting the gap between reality and the “hype.”

If you simply Google something like “how many businesses us ChatGPT?”, you’ll likely pull quick results like this that finds that 49%(!) of companies currently use ChatGPT (which would imply that an even higher share of businesses are using at least some form of AI).

In contrast, the Census Bureau recently introduced an AI question to its .

This survey is administered every two weeks and asks firms about activity over the previous two weeks (i.e., trends) and planned activity over the next six months (i.e., outlook).

BTOS doesn’t benefit in the same way as the ABS from extremely large sample sizes and response rates, but it is nonetheless representative of the universe of U.S. employer businesses across important characteristics like size, industry, and geography.

Up to this point in time, BTOS has received 6 rounds of responses asking about AI use in firms.

The first round, whose collection ended 9/24/23 (roughly 10 months since the release of ChatGPT), yielded an estimated usage rate of 3.7%—much closer to the ABS’s 3.2% than to ResumeBuilder’s 49%.

Estimates from the next three collection periods (6 weeks) were all between 3.7% and 3.9%.

At the same time, 6.3%–6.5% of businesses reported expectations to use AI during the following 6 months, which would be a sizeable jump in a short amount of time (i.e., around 70% growth in AI use over 6 months).

I initially chalked this up to BTOS’s projected rates simply picking up more of the hype.

However, the last two BTOS collections yielded current AI usage estimates of 4.4% and 4.6%, respectively.

To put that into context, those usage rates represent around 20%–25% growth in reported AI use over roughly 2 months (i.e., since the initial collection period).

That amount of growth—if sustained—could put the future AI usage rate on schedule to hit right around the original “outlook” rate of 6.3% after 6 months.

Because of these latest results, I’m somewhat more likely to believe that AI use (led by ChatGPT) may really begin picking up among firms.

However, it remains to be seen whether this trend will persist or if it’s mostly noise.

Finally, it appears that the most recent (post-ChatGPT) measures of AI use show a larger share of smaller firms reporting use in their business processes.

This is contrary to the robust patterns of AI use with respect to firm size measured in the ABS.

In both 2018 and 2019, firm size is a very strong predictor of whether a business uses AI, showing basically a monotonic relationship with size (i.e., larger firms are consistently more likely to use AI than smaller firms).

This figure shows the average AI use reported in BTOS by various firm size classes.

What I find to be the most interesting pattern in the BTOS data so far is the more U-shaped (rather than strictly increasing) pattern of AI use with respect to firm size.

That is, the largest firm size classes report the highest rates of using—and planning to use—AI.

Again, time alone will tell—and a look at the microdata will be necessary to verify that this size relationship holds within sector—but this pattern could very well be the new norm given the easy accessibility of ChatGPT-like tools to small businesses.

Provided by Zachary Kroff