CAPMF - EPAM Systems: AI Could Have A Negative Long-Term Impact On IT Services

2023-07-17 10:20:01 ET

Summary

• As with any new digital technology, generative AI will create new incremental demand for IT services companies.
• In some cases, it will be the catalyst for wider digital transformation demand, but this will be the exception.
• For IT service companies, generative AI is also a business risk – productivity increases could reduce the amount of billable work.
• Companies that rely on providing billable engineering services, have no products to sell and no internal IP, and also do not engage in consulting at scale, are most at risk in my view.
• EPAM Systems, as an example, is at a high risk of seeing revenues stagnate or even decline.

Investment Thesis

The IT Services industry is currently going through a phase of stagnating growth. The general thinking seems to be that this is temporary. The expectation is that over time the industry will again see the previous tailwinds and customer demand for digital engineering services will increase again - as the ever-growing need for companies to create digital business models and to transform and improve operational capabilities does not go away.

As with practically any new digital technology (think about mobile apps, analytics, IoT, etc.), generative AI surely will create additional demand for IT Services companies. It is not difficult to imagine use cases like building AI-driven chatbots for internal or external use to access corporate data.

I suspect though that the new, incremental demand here will not be large enough to have a significant impact on most IT Services companies, which are usually quite large. I am using EPAM Systems ( EPAM ) as an example in this article because I have already covered the company on Seeking Alpha previously, and because it is the type of IT Services provider that will be most affected, in my view, by generative AI. While EPAM Systems is a relatively small player, it still has almost 58,000 employees . Accenture ( ACN ), the world's largest IT Services company, has 732,000 employees and is a corporate giant in comparison. But in both cases, it takes a lot to move the needle.

At the same time, I do not think that generative AI will create substantial new demand for digital engineering services. To make meaningful use of AI, organizations need modern and flexible information and data architectures. But they need this anyway for effective automation of business processes and digital customer journeys. I do not see a lot of organizations that have not done this before embarking on a digital transformation journey because they want to take advantage of generative AI. There are probably better and more urgent reasons for that.

It is a dirty secret of the technology services industry that engineering services can be inefficient, especially when practiced at scale, and there is a contradiction inherent in the business model. Less productivity can mean more revenue and profit. It is no coincidence, in my view, that companies like EPAM and Accenture gravitate towards large clients with hundreds and even thousands of people working for one customer. Only large companies can come up with a working business case to create their own technology specific to their business.

At the same time, the engineering work is very often highly repetitive and what is created is not especially exceptional or unique in my opinion. In my experience, at least 80 percent of the work goes into technical plumbing which is not much different in company A than it is in company B. IT services companies sell by convincing prospective customers that they have the capabilities, people, and best practices processes and technology skills to get the job done. This is the nature of the business.

On the other hand, it does not take a lot of creativity to conclude that large language models are a nice solution for the problem to create functionally correct software code (which is a formal language) based on sound architectural patterns, technologies in use, and organization-specific standards.

McKinsey estimates the economic impact of generative AI to be between $2.6 and $4.4 trillion per year. 75 percent of that additional value is in only four business areas - customer operations, marketing and sales, research and development, and software engineering (!).

Those are almost perfect conditions for disruption. It would be naïve to assume that IT Services providers will be the main or only beneficiaries of the additional value and productivity improvements. Clients will either consume even more engineering services as they can get more for the same cost (I do not think so, but it is possible that the law of demand applies here) or they will pay less, meaning less revenue for the provider. But even in the best of cases, the increased productivity should reduce the growth potential.

This certainly does not mean that the outcome is inevitable. The effects will not be immediate and will take time to play out, so there are mitigation steps providers can take (more on that later). But investors should be aware of the risks and include the impact of generative AI in their due diligence.

Due to the large managed services and especially consulting and project business, companies like Accenture are, despite their size, to a certain degree insulated from what is coming. Companies like EPAM will be much more affected, in my view. EPAM is mostly about digital engineering and its design and consulting businesses are relatively small. Furthermore, EPAM has practically no products or intellectual property to sell, and almost 90 percent of revenue is from time & material-based engagements .

The Impact of AI on Software Engineering

According to McKinsey , software engineering will be significantly impacted by generative AI, both from the size of the economic impact and the percentage of functional spending that will be affected. The only other business function that comes close is Customer Operations.

There are already tools available like GitHub Copilot and Amazon CodeWhisperer which help software engineers to code quicker and with better quality. As far as the art of the possible goes, those are still simple tools, in my view. They are advertised as AI pair programmer (Copilot) or AI coding companion (CodeWhisperer), which basically says clearly enough what the objective is – to assist a developer in his coding tasks. Despite the limited scope, they are already highly effective. A Microsoft study together with Cornell University, The Impact of AI on Developer Productivity: Evidence from GitHub Copilot, found that developers were on average 55% quicker when coding a specific task assisted by Copilot (Note - Microsoft owns GitHub, so there could be some bias here).

McKinsey expects the direct impact of AI on the productivity of software engineering to range from 20 to 45 percent of the current annual spending on the function. I actually think that this could be on the low side for enterprise-scale software development, where IT services companies like EPAM come into play.

Here, less than half of the cost (often much less) is usually spent writing code. There are a plethora of other functions and people which are needed and employed. Sometimes the names of the roles/functions do not make even sense to somebody who has not been involved in software engineering: business analysts, quality engineers (which come in two forms – for manual and automated testing), solution architects, data architects, UX designers and UI designers (which is not the same thing), product owners, scrum masters, infrastructure architects, DevOps engineers, etc. Their output ranges from structured text (like user stories ) and a variety of documentation and diagrams, to test or build scripts. The increasing complexity of software engineering and the proliferation of different technologies creates more and more of those specialized roles. I think it is very much possible that there are already AI prompt engineers.

There is a dirty secret in the industry and, at least I always thought, an inherent contradiction in the business model of technology services: you can make more money when you are less productive and need more people, as long as you can convince a customer to pay for it.

In short: technology services could be an industry ready for disruption, and AI could be the instrument for it.

The potential disruptive effect on IT Services

A lot of the excitement regarding AI on the executive level seems to come from an angle where AI is seen as a cost-cutting tool that allows capital to win the next battle in the perennial fight with its old adversary, labor.

It took not long after the launch of ChatGPT for IBM ( IBM ) CEO Arvind Krishna to declare that within IBM hiring in back-office functions will be suspended or slowed as a significant part of non-customer facing roles will now be replaced by AI and automation. Personally, I think he has not thought this through to the end. If you read the full Bloomberg article, IBM to Pause Hiring for Jobs That AI Could Do , you will notice that Kirshna sees AI reducing his cost, not his revenue. I think this is a one-sided and incomplete view.

To a very large part, IT Services companies charge customers for work done based on the number of people that were used and the time they spent working. Even if client engagements are outcome-based, the charging mechanism underneath is usually based on people, skills, rates, and time. The economics of such a technology services business in principle come down to selling working time.

Who will be affected most?

To be fair to IBM, the company has significant revenue outside of this model. But others, like EPAM, are almost 100% within the model.

If investors think my thesis is correct or want to investigate the risk, there is an easy shortcut to check whether and to what extent revenue is at risk.

There are only few criteria to check:

1. Revenue is driven by charging working time. When this is the case, companies usually publish utilization as a critical business metric in their reporting. Utilization is not always calculated in the same way, but in essence, it means the percentage employees are charged to clients. You can see this metric in the reporting from companies like EPAM Systems, Cognizant ( CTSH ), or Capgemini ( CGEMY ). Companies might use different terminology. Accenture, for example, often uses chargeability instead of utilization, but it means the same thing.
2. Revenue is driven by engineering services, versus design, consulting, and outsourced managed services.
3. Revenue is predominantly time & material, versus outcome and project (fixed price) driven. Unfortunately, not all companies publish this information and you might need to look at earnings call transcripts or other material to discover it.

EPAM scores high on all three criteria, so – assuming I am right – it is very much at risk. EPAM does very little consulting. It has tried to move upwards from engineering to consulting and design with its EPAM Continuum brand, but when it does those things it is usually tied to an engineering engagement. Contracts are overwhelmingly time & material. Time & material made up 88.7% of revenue in Q1 2023 . It had been only 85.1% two years back in Q1 2021 .

When could the disruption happen and what can be done about it?

When OpenAI released ChatGPT in November last year, it seemed like something new and exciting had suddenly appeared. As a matter of fact, this had been in the making for years. We just did not notice it because it happened - as the saying goes - gradually and then suddenly.

It is next to impossible to predict when the next steps will come. My guess is sooner than we think. This is based on the significant investment that is going into all things AI currently. Still, the transformation will take a few years, so in theory, organizations have time to react.

An obvious choice is to move from engineering to more design and consulting activities. But this can only help to a degree. Ultimately IT services companies need to augment their focus on technology and invest in industry-specific capabilities. Such capabilities are hard to build internally, at least not in a reasonable timeframe, so acquisitions are probably the way to go. A good example is a recent acquisition by HCL Technologies , an Indian technology services company with more than 200,000 people. HCL recently acquired ASAP Group, which specializes in autonomous driving and e-mobility.

Investors should watch out for the actions companies are taking to see whether they stay within the technology space, or are broadening their capabilities.

Conclusion

Generative AI will have a tremendous impact on software engineering by making it significantly more efficient.

This could actually reduce demand for engineering services over time and be very disruptive for IT Services companies – the extent depends on how much the business model relies on the provision of engineering services, and not on design, business consulting, and managed services. Significant productivity gains will be achieved, but it is unlikely that providers will be the sole beneficiaries. The productivity gains will need to be shared with customers, possibly reducing the demand for those services.

Investors will need to consider this in their due diligence as not all IT Services companies will be equally affected and there are ways to mitigate the coming headwinds.

For further details see: