AI and IT Services (Part II)

You,•IT services
Back

This post will cover who the winners among IT Services are in this shift to AI, and what the opportunity for new entrepreneurs is. You can find my previous post on this topic here (opens in a new tab).

tl;dr - it's never been a hotter time for entrepreneurs to enter gen AI augmented services if the number of conversations with VCs I've had exploring this thesis is anything to judge by. Drop me a note if you're fundraising at anshulbhide@gmail.com!

Project Delivery

Feel free to skip to the next section if you know how IT Services companies deliver projects.

How do IT Services companies actually operate? It's not as easy as renting out developers, although there are some players that have built multi-billion dollar companies doing this e.g. Toptal.

There are two primary modes of engagement - Fixed Price Projects (FPP) and Time & Material (T&M) projects.

FPPs means that an IT Services company will take on a fixed scope of work for a lump sum amount of money. This is typically software development work that can be carved out with minimal customer involvement in actual software development.

T&M projects mean billing based on the number of people and time spent on a project. Therefore your revenue is a function of time spent, developers staffed and billing rate per developer.

Guess which model has more chance of productivity gains due to gen AI?

If you answered FPPs, you're correct. The only direct costs you incur while delivering IT services are your developers' salary costs. If you can now create a webapp for a client in one month with just one developer instead of two developers by using AI coding tools like Github Copilot or Replit (opens in a new tab) , then you have saved one developers' salary cost which goes straight to the bottom line.

However, there is scope for improvement in T&M projects as well. The little secret of successful IT services companies is that it's not uncommon for developers to be billed across multiple projects. So if a developer can complete the same amount of work (say 5 bug fixes) in 50% less time, then they can now work on more projects thereby boosting the company's top line and margins.

Vaibhav Domkundwar (opens in a new tab) from Better Capital (opens in a new tab) has spoken about how SaaS is going to move to "Outcome as a service (opens in a new tab)"; the same is going to happen to IT Services companies with gen AI.

So who are the winners here?

Large IT Services Companies

The large IT Services companies (Infosys, HCL, TCS etc.) will benefit from the proliferation of gen AI. The reason is that it is easier for them to move to an FPP model than the smaller players. Even pre-GPT, large IT Services companies had more than half of their revenue coming from FPPs.

FPP is risky for small companies due to scope creep, where clients request features beyond the original project, leading to cost overruns. Additionally, projects may not meet client expectations unless they are well-defined (e.g. a product integration). Thus, smaller IT services companies prefer T&M over FPPs as they dont have the resources or financial buffer to endure the few projects that do go over budget while larger companies do. And the shift to gen AI aids in completing these FPPs with higher margins.

What might stop the large IT Services companies from winning? One word: culture.

These companies have millions of developers on their payroll, and adopting AI in software delivery at that scale will be incredibly complex. When you read headline numbers of the likes of TCS training millions of developers in gen AI, it is little more than a fresher completing a mandated training by clicking the "Next" button on each screen repeatedly without actually learning anything.

Click, click, click

In addition, most middle managers prize their fiefdoms and armies of developers under them. As you can imagine, they will not take kindly to anything that might threaten this including gen AI and adopting this new technology will be slow.

Finally, these companies have large amounts of legal exposure, and will be careful when using AI coding tools like Github Copilot on client projects that may have been trained on open-source data. Copilot has reportedly indemnified IT Service Providers who face lawsuits for using Copilot but they will nevertheless be very wary, especially for large projects.

No "moving fast and breaking things" for these IT Services giants.

Perhaps, there could be another winner.

Agentic IT Services Companies

Agents

Here's another industry secret: IT Services companies have products. They develop IP that they either license to companies directly (similar to a SaaS model), or use themselves internally to accelerate product delivery (accelerators). For example, we at Calsoft AI have a tool called CalTIA that automatically identifies relevant test cases to run using ML after a code commit thereby increasing efficiency of testing cycles (shoutout to Shrish Ashtaputre (opens in a new tab) & team for building this).

With Gen AI, there is a new type of accelerator that will become predominant: Agents.

No, not those kind of agents

Agents has become a much overused term in gen AI, and is likely being overstated in the short term and underestimated in the long term. Simply put, an Agent is a fancy name for a function call that makes use of external data via APIs and internal data via RAG to execute enterprise workflows autonomously.

For example, you can build and deploy an AI agent to summarise your emails and generate automatic replies based on the content. This involves passing the email contents through a data pipeline that uses RAG to enhance a large language model (LLM) with relevant context about your job function, and then generating a summary and suggested responses. Superhuman recently added such a feature to their product, showcasing how AI agents can streamline communication tasks efficiently. Jerry Liu (opens in a new tab) from LlamaIndex (opens in a new tab) has an excellent video (opens in a new tab) of how to add agents to your RAG pipeline.

However, in the context of IT Services delivery, this is just another accelerator; albeit one that accelerates software delivery by 100x rather than 20%.

Here is what a typical IT Services team looks like to create a basic ecommerce webapp.

Now what if you had a combination of agents for each of these tasks?

ChatDev, an OSS project has a set of communicative agents for software creation

Suddenly you can remove most of the humans out of the equation. You will still need a human in-the-loop to manage the entire workflow but suddenly the gross margins become very attractive. Traditional IT Services companies typically have gross margins of 50-60% while SaaS has margins of ~80%. If you can replace humans via agents, your margins can go up to 70%.

Opportunity

So what does this mean?

Traditionally a lot of IT Services projects have revolved around customisation or implementation of software products. Take SAP. It's a massive product company with a market cap of ~$250Bn. But it also has a massive IT services ecosystem; the SAP IT services market is estimated to be ~$100Bn. These ecosystem IT Partners customise software as per the clients' requirements.

Let's say you are a small toothpaste carton manufacturing player who has home-grown software for your Manufacturing Execution System (MES) - which SAP doesn't support through its middleware. You will need to hire in-house developers to integrate your MES with SAP SCM if you want to do this. But your core business skill set is not software development - you just want to manufacture toothpaste cartons! So you'll turn to SAP IT Services partners who can help you with this.

In a world with agent-led software development, this suddenly becomes a lot easier, cheaper and faster. Every SaaS software can now be customized. This is the core idea behind Chamath's 8090 incubator (opens in a new tab) which will create SaaS software with a 90% discount using offshoring and AI (likely agents).

I can't stress enough how big an opportunity this is for new-age IT Services companies. Traditional IT Service implementers that have built multi-billion dollar businesses like Bristlecone (opens in a new tab) just don't have the incentive to move to gen AI as their entire business model is predicated on billing customers as much as possible for customising and implementing software at a huge markup. New-age companies using AI agents to implement software can split the cost savings with the customer for an entry into this sector.

Conclusion

If you are an entrepreneur building with AI in the services space, I would love to speak with you to see how I can help! Please write to me at anshul@calsoftinc.com!

In addition, we are actively hiring at Calsoft AI (calsoft.ai (opens in a new tab)) so if you are a young talented developer who's excellent at Python and enjoys tinkering with LLM workflows, please DM me on LinkedIn.

Special thanks to Sahil Aggarwal (opens in a new tab) for his thoughtful feedback on this article and Krishna Kulkarni (opens in a new tab) for reviewing!

© Anshul Bhide.RSS