We have two strategic and fast-growing uses of AI at RunSignup.
1. Building AI into our products for customers to use
Expected cost: $10,000+ per month by 2027
2. Using AI within our development team to build our product
Expected cost: $20,000–$40,000 per month by 2027
We already know that building AI into our products can be meaningful. Our AI Chat Agent for customer websites has answered tens of thousands of questions that did not require the attention of event directors. More meaningful technology is coming this year as well, as described in our AI Plans for 2026. These are the kinds of features that help keep customers happy, differentiate us in the market, and attract new customers.
Our thesis on AI use by our development team is that it will enable us to roughly double our productivity over the next 18 months. In practical terms, that means releasing about twice as many features for customers with the same size team. We believe this is necessary to stay competitive. We also need to learn how to use AI better than those who may try to build similar capabilities on their own with AI DIY.
The costs are relatively low today, but they will become more significant as we move through this year and into 2027. We have been thinking carefully about this and working to understand the potential costs and benefits of each.The purpose of this blog is to share our thoughts on the substantial costs of AI, and why we think the investment is worth the cost.
Cost of Building AI Into Our Products
The total operating cost (not including our development and maintenance time) to date is only around $500 per month. The primary variable cost is the use of the LLM’s we use on the back end – Claude, ChatGPT and Gemini. As we build more of our own AI technology on top of AWS Bedrock and Bedrock AgentCore, those costs will build but will be a small fraction of the LLM token cost. We expect the total cost to be $2-4,000 per month by the end of the year, and hit $10,000 per month in 2027.
This estimate of $10,000 per month is comparable with our monthly costs for the free Email we provide customers. We think AI features we build into our product will have similar advantages to our business and be well worth the cost. As rough math, this is only 1-2% of our net revenue and we feel having features like free AI, Websites, EMail, and now free AI built into our products helps us keep and win business. Even better, they help our customers grow, which is mutually beneficial.
AI in our Products Cost Details
Currently, there are two uses of AI that are in our product. The first is our AI Chatbot for Website V2. As of the end of February we have 1,139 websites using the AI Chatbot, and a total of 24,414 conversations. We currently use ChatGPT for this application and have used over 227 Million tokens. We believe we will have around 10,000 websites in 2027 that use the service. We also believe that participants and ticket buyers will become more used to interacting with AI Chatbots over the next 2 years and prefer it to calling a human or sending an email and waiting for a response to their questions. These trends are confirmed in our monthly stats.
So as we look to 2027, we would expect a 20X growth in our current costs of about $200 per month. That is $4,000 per month. This will be our highest use application.
Our second currently deployed application is actually not visible to customers. We are using AWS Bedrock and Claude to try to prevent spam on our website and when customers send email. It has proven to be very useful as we get more AI driven spam and attacks. We are basically fighting AI with AI. The cost is relatively inexpensive – only $200 per month – and not likely to grow much.
As we look to the future, we expect most dashboard pages to have AI Chat capabilities built in. We will also expand the event website chat capability to increase with things like changing events from the marathon to the half marathon. Most of these are fairly low in volume of usage. However, there will be several high usage applications:
- AI Results within Race Websites. We have a new Results page coming in the next couple of months that will also include AI chat. With millions and millions of results posted every year, we expect this to be very popular and likely our #1 token generation application. It is hard to anticipate how much usage it will get, but probably at least the volume of the chatbot, and likely more. Again, results are the heart of races with participants and spectators wanting information like how someone’s times over the years have compared. Since customers want this, we will deliver it – again, for free even if our costs are $5-10,000 per month.
- AI Data Exploration. Events have a massive amount of data stored in RunSignup and TicketSignup. We offer lots of reports that event directors use to understand and visualize their data. AI brings a new level of flexibility to evaluating and understanding data. As we begin to build AI into our reporting dashboard pages, we expect high costs. However, there are only a limited number of event directors compared with people who signup. And the data exploration will be periodic in nature. So an estimate of 2027 costs might be in the $1,000 – 2,000 range.
Development Team AI Costs
We estimate that our AI token costs (in addition to licenses to tools like Cursor and Github) for our own development team to rise from about $2,500 per month now to $20,000 – $40,000 per month in 2027. To put this in context, our development team budget for 2026 is about $8 Million per year ($670K per month). So that cost is roughly 3-6% increase in our costs. Taken from another viewpoint, we believe that each developer will use about $1,000 per month in Tokens.
We are in the midst of testing out two agent-led projects. The first is trying to have an agent do code review (in addition to our human code review). Given the number of pull requests, the size of our codebase, and the increased volume we expect to get from AI, this could “auto-use” tons of tokens and might explode the costs even higher than our estimates. Here is an example of our current agent in test…
Example Comment looks exactly like a human review:
The second prototype we tried this week was taking a Github issue that one of our account managers submitted and just letting Claude see if it could come up with a plan. We actually ran it against Claude 4.6 and ChatGPT 5.4 and liked the Claude one a lot more. We then had it write the code. We manually reviewed the code, made some edits and it is in pull request now.
Remember, most of this was generated by AI after the initial Issue. So lots of agent generated tokens.
AI Writes 90% of the Code by Mid to Late 2026
By mid to late 2026, all of our developers will be doing AI first development. This means asking AI to write plans and to write the code. Our job will be to teach the AI how to stay within guardrails, learn what is important to our customers, and to assure the same type of quality that has led to only 6 minutes of downtime since 2015.
Summary
By 2027 we expect to be spending $30-50,000 a month on AI tokens – $350,000 – $600,000 over the course of the year. This is a large ramp in our spend on computers – last year we only spent $400,000 on AWS to run our entire infrastructure.
Our rapid adoption of AI fits our “Aggressive Patience” culture. We will move quickly, but safely and always assuring quality. I also think being an employee owned company is important because employees know that productivity gains and market share gains we make will be shared among all of us. And we have all bought into the concept that spending a lot of money on AI will benefit us and our customers.
From an economic perspective, this is a significant investment. Our bet on AI is simple: if we use it to produce results that customers genuinely appreciate, they’ll keep coming back—and that’s what ultimately drives growth. We also anticipate that we will not grow the size of the team as quickly as we have in the past. That said, I do not foresee a situation where we would need to lay off employees to afford more AI agents. We have an exceptionally strong product. We have no debt or outside investors. We are nicely profitable and cash-flow positive, with good reserves in the bank and a strong current ratio.