Generating 1000%+ ROI with a chatbot

Outsourcing AI to SaaS Providers

Klarna made headlines last year by claiming its AI assistant eliminated the need for 800 workers. Employee numbers at the company are down 38% since 2022, in part due to AI adoption.

In its IPO filing last week, the company revealed savings of around $49m from its AI assistant, marketing and automations. This is less than 2% of annual costs. Klarna did reduce sales and marketing costs by $203 million, in part because of AI, and said that 96% of employees use generative AI in their daily work.

Yet if the poster child for AI adoption is saving less than 10% of annual costs, the AI revolution has a way to go. A Federal Reserve survey of surveys shows employees use AI three times more than companies, suggesting that businesses are yet to find an essential role for AI. One reason for that is outsourcing AI to existing SaaS suppliers.

While AI can make software more efficient, it’s an important marketing message that applications have “AI inside”. As this is a defensive measure on the part of suppliers, it won’t necessarily provide a productivity boost for users.

The Evolution of AI Products

Azeem Azhar has a four quadrant model of AI ventures. I’ve adapted this, putting return on investment on the vertical axis and defensible moat, or how long an advantage lasts, on the horizontal.

The ideal situation is in the top right, where AI enables an efficient and lasting product. The opposite outcome is the AI graveyard in the bottom left. The other two boxes represent expensive moats to defend market share, and effective but short-lived AI successes.

Azhar speculates that as ideas are easily copied, many AI companies will fall into the top left category of fleeting success. This creates challenges for companies that buy AI services. If your supplier is going to be superseded by a superior product, then you will need to update your tech stack on a regular basis. Even if every upgrade integrates seamlessly and is a direct replacement for what came before, the friction from frequent change may erode any cost savings.

How to Build A Chatbot

The biggest challenge with in-house AI projects is the time to completion. Even leveraging existing models from established providers such as OpenAI, Meta and NVIDIA, you are looking at around a year to deliver results. At least 60% of this time will be preparing data, as discussed last week. Nonetheless, the returns can be worth it.

I am working with a client to develop a chatbot for its website. This is a consumer facing company, selling small ticket educational services. The current site is available in 15 languages and receives around three quarters of a million visitors a year. The services retail at around £20 a time.

For around £150,000 development costs over six months, and £50,000 annual operating costs, the client will have a chatbot that is an expert in its products and services. The bot will guide visitors’ journey through the site, make recommendations and answer questions. Increased visitor traction will boost sales. There will also be a reduction in support staff handling common queries, while freeing others to handle more complex enquiries.

Assuming staff savings do no more than offset development costs, the project pays for itself if only 0.3% more visitors purchase a product. The client is seeking a 15% increase in sales. Assuming 10% of current visitors make a purchase, this uplift would generate over 100% return in year one, and 450% return on the increase in annual costs.

Furthermore, with a scalable site, the client can expand its range of languages. It currently does nothing in Indian or Chinese dialects. Chatbot producer Tidio estimates return on investment of 1,275% from staff savings alone. Our client has a clear path to making these type of returns from increased revenue.

Revenue Growth rather than Cost Saving

To date, the savings from AI adoption have been muted. In part this is because a lot of AI adoption is to justify or increase the price paid for existing software services. Even companies running their own projects focus on sales, marketing and coding costs. The real benefits of AI will come from reimagining service provision in a way that makes it much more scalable. AI will then be about higher revenue rather than just cost savings.

Questions to Ask and Answer

1. What are the major blockers to growing revenue in my business?

2. Would sales be scalable if some processes were automated?

3. Would we sell more from our website by guiding visitors?

Here are 3 ways I can help:

1. Book a consultation to talk about all things AI.

2. Read our deep dive into data science.

3. Explore our use cases using accelerated computing.