Software’s Next Meal Will Be Services: “Service as a Software”

This has been my Roman Empire for the past few months.

Deloitte, PwC, EY and KPMG, otherwise known as the ‘big four’, have been relatively unchallenged in their end-to-end delivery of professional services. In 2023 alone, they brought in around $200bn in revenue. Collectively, they employ one and half million people. 

This is about to change. Even though we’re still reasonably early in the AI revolution, anecdotal evidence points to the alchemical ability of large models to convert computing power into labor. If the first two decades of the 21st century have been all about “Software as a Service”, I believe the next two will be all about “Service as a Software”. 

Image was generated by Midjourney

The opportunity:

Historically, B2B SaaS companies have won market share by creating aesthetic digital simulations of pen and paper activities (think CRMs, spreadsheets, directories, presentation tool sets, etc). Flooded by ample venture capital dollars, the market for B2B SaaS has become oversaturated. This has turned the vast majority of business software applications into a red ocean. 

Services business might represent a bold new frontier for tech-challengers. Going by market size, they represent a massive opportunity. For example, the global accounting services market alone is set to be worth c. $676 billion in 2024. Precedent, however, has dictated that ‘services’ businesses are not ‘venture backable’. Given that these services tend to be performed by humans with scalability limits, they simply haven’t been attractive venture scale opportunities. Up until now, these businesses have also struggled to benefit from technology. SaaS has been unable to meaningfully lower business costs or reduce headcount, and so average operating margins have remained flat (c. 10%) since 1960. 

The one exception to the rule is Palantir, which is both technologically-enabled and services driven. Going by multiples, it’s also one of the most highly valued venture-backed companies. With the directional growth of AI, I would argue that it might not stay an outlier for much longer. I believe the next generation of SaaS will be “Service as a Software”, whereby hybrid companies - part human and part AI-enabled - will deliver end-to-end services. 

Why now? 

New AI models could enable a new category of service businesses, which outperform and outprice incumbents. Already, OpenAI’s GPT-4 scores at or above the 80th percentile of humans in most domains. Freelancer marketplaces - a good barometer of demand for knowledge work - are starting to reflect the capabilities of large AI models. New Upwork jobs show ChatGPT has had an impact on demand for writing, customer service and translation. 

Source dated February 16th 2024

At the bleeding edge, researchers are working on AI agents, autonomous intelligent systems performing specific tasks without human intervention. These AI agents are likely to be great at completing repetitive, arduous tasks, allowing human counterparts to focus on more mission-critical or creative activities. In the future, systems of AI agents might be perfect for solving pain points that businesses have historically turned to services for. For example, we are already witnessing how AI agents could augment or even replace software development work with tools like GPT-Engineer and Devin.  

It seems likely that the capabilities of AI models will continue to grow and the cost to use them will fall. If AI proves to be a form of labor replacement, there will be an opportunity for substantial margin enhancement. Unlike their incumbent software counterparts, legacy services businesses will struggle to properly enhance their product offering - humans. Instead, their business models will continue to rely on human labor and billable hours, which can be outmatched by AI-enabled vendors. For example, in response to growing demand for its services, Accenture recently announced that it plans to double its AI division to reach 80,000 employees. Imagine what a system of 80,000 AI agents could do, and at what cost?

Image generated by Midjourney

Before we get there though, there are substantial hurdles to overcome. Challenges such as potential inaccuracies or hallucinations can have significant consequences for any applied AI. Just recently, Air Canada was held liable for its own AI chatbot giving customers bad advice. While LLMs can be great, they’re not perfect yet. When it comes to AI agents autonomously doing tasks, they are even more at the mercy of these inaccuracies.

To be able to compete with traditional services businesses, new products using AI agents will have to provide robust, actionable solutions out of the box. What might this look like? 

1. Specific vs General: while foundational model capabilities continue to improve, AI agents will be more likely to have initial success by focusing on very specific tasks as opposed to general ones. AI-enabled services might prioritize starting with niche automation tasks, with a view to expand and serve a category more broadly. 

2. Human in the loop / ‘CoPilot’ Models: allowing for human feedback can be a stepping stone to producing effective agentic processes. Instead of letting error rates compound, the agent could intermittently ask a human for feedback to make sure it’s on the right track and then adjust if not. New services companies can excel in the way they design the relationship between their employees and AIs.

What services might be most ready for disruption?

Service firms are hired typically for one of three reasons: 

1. To do jobs that the customer either doesn’t want to do or doesn’t have resources to do.

2. To offer external expertise on particular decisions

3. To act as a layer that removes client liability. 

This first bucket typically deals with execution related projects, where the client just wants results. These tasks might include IT development and support, financial, accounting and auditing. Typically, this work is also repetitive and represents significant potential for automation. Some opportunities: 

• Software development 

  • Maintaining apps 

  • Pen testing 

  • Integrations 

  • Custom software implementations 

  • Cloud migration 

  • Quality assurance

• Customer Support 

• Data analytics 

• Finance

  • Auditing 

  • Bookkeeping 

  • Tax filing and planning

  • Financial reporting 

The second and third bucket are usually much more advisory-related work, which would include things like strategy, M&A, litigation and organization design. Here, the opportunities for automation are more limited - given clients are likely to be much more sensitive buyers and the tasks more complex / less repeatable. These dynamics also make it harder for them to trust AI in the loop. 

• Strategy and Management Consulting 

  • Organization design 

  • Market expansion / entry 

  • Cost optimisation 

• Finance

  • M&A advisory 

  • Capital raising

  • Financial due diligence 

• Legal 

  • Litigation 

  • Patent and IP Services 

  • Compliance

The higher the complexity, the more humans needed in the loop. The higher the level of trust, the more repeatable a task, the easier it will be for AI-enabled services to slot in. That said, there are still a lot of difficult questions. What is the right ratio of humans to AI? Which functions are businesses ready to trust to automated services? Will these automations work better than the status quo? How do liability and trust evolve? 

Here are a few ideas on how it might play out: 

1. The least complex, most repeatable will be the lowest hanging fruit: for example, going after certain finance and IT support functions might make more sense than the advisory work McKinsey is typically hired to do.  

2. Brand will matter: historically, services firms have been hired based on their brand. It will be just as important for new entrants to create strong, visible and trustworthy brands. Likewise, it will be important for product marketers to understand exactly what kind of messaging AI-enabled services can lean on to make themselves more appealing than their incumbent competitors. 

3. Consultancies and service organizations will still try to compete: nearly all big firms have launched comprehensive AI strategies, and several have partnerships with OpenAI / Microsoft. It is more likely that they will be able to own ‘Copilot’ products than autonomous workflows. It’s also true that some new entrants can also leverage these partnerships to create more disruption for themselves. For example, Harvey recently announced joint partnerships with OpenAI and PwC to train and deploy foundation models for tax, legal and human resources services. 

4. Liability will matter: some spaces will be extremely hard to innovate in simply because customers are looking to buy from a third party who they can blame if things go wrong. These will be unideal areas for AI-enabled services to compete in. Even still, it raises the important question of how liability is dealt with for AI systems. 

5. New business models will emerge: rather than usage-based pricing, AI-enabled companies could better align their business models with customers by billing for outputs. Legacy service providers will struggle to innovate their business models so long as their main product is humans, who work on billable hours. Incumbents will need to meaningfully alter not only their products, but also their GTMs to compete with this strategy. 

Who might your customers be?

Another way to look at things is to think about the sectors spending most on consulting. Services businesses typically work with  traditionally conservative industries. Some of their systems and ways of working haven’t changed for decades. That’s not to say that we can’t expect innovation from the industry, but rather that it’s going to take a very compelling go-to-market approach. But, as Palantir has shown through the public sector, the reward can be significant. For example, Palantir won a £480M NHS contract in 2023. The top five verticals are: 

1. Financial services 

   1. The Financial Services Industry is traditionally one of the largest users of Consulting. According to Gartner, banking and investment firms spent c. $270 billion in 2023 on IT consultants, an increase of around nine percent from 2022.

2. Public Sector 

   1. The public sector, namely government and administrations, is another very large segment. For example, federal spending on consulting services in the US rose by 16% from 2022 to 2023. In 2022, the UK public sector awarded £2.8bn of consulting contracts.

   2. Worth noting, governments are under pressure to reduce the cost of consulting engagements. Even the IMF recently recommended the reduction of reliance on blue chip consultancies.

3. Health and Life Sciences 

   1. Healthcare costs are growing in many countries. In the US, they are close to 20% of GDP. Healthcare payers are under pressure to reduce costs while providers need to increase their efficiency. Pharmaceutical companies manage vast R&D portfolios as their shareholders expect significant returns.

   2. For example, the UK’s 215 trusts in the National Health Service spend an average of £1.2M on external consultancy fees per year, with some spending up to £5.6M 

4. Manufacturing (including automotive, aerospace, consumer packaged goods, electronics, and process industries) 

   1. All these industries are facing growing headwinds, competition challenges, and an aging workforce. They are also challenged to embrace new technologies that can completely disrupt their value chains.

5. Energy and Environment

   1. Companies in energy and utilities face increasing customer demand, high capital investment, increased competition, and more stringent regulations. 

Some final thoughts…

With every major advancement, services have capitalized on assimilating new technologies and best practices to businesses as independent third parties. Consulting, as it’s now known today, was precipitated by the first industrial revolution. The digital revolution of the 1980s spawned IT consultants, because computer makers were barred from selling their expertise due to antitrust concerns. Fast forward to the present, Accenture’s generative AI revenue has already reached $1.1bn in 2024, surpassing most, if not all, venture-backed AI startups. That is to say, the incumbents absolutely should not be underestimated. 

AI-enabled services are novel, with a lot still left to be determined. Despite its best efforts, “Software as a Service” has had very little impact on the efficiency of much of the Fortune 500. It’s just never been able to get the job done from start to finish. It feels like we’re entering an era where this status quo might finally be challenged. And this represents an enormous opportunity to expand the TAM of B2B software. If you’re exploring or building one of these companies, I’d be very interested to hear from you (nikita@hummingbird.vc).