Like most other tech organisations we’ve been considering our options with regard to AI, deciding how and where to use it. Over the last few months we’ve taken concrete steps to introduce AI into our working practices. We’d like to take this opportunity to let you know what we’ve done so far, and what our plans are. We also want to point out what we are not planning to do, as some of the hype around AI may suggest behaviour that we have no intention of adopting. There are some very good reasons to be cautious, but it is clear that the potential benefits cannot be ignored.

AI in Development

Our primary focus so far has been on integrating AI tooling into our development environment. We’ve engaged Dr George Bilchev, previously a Principal Architect at BT and an AI expert, to lead this effort. We are now able to use AI to assist with all parts of the development life cycle, from requirements gathering and design through to coding, testing and documentation. It is important to note that we are using AI as an assistive technology. No code produced with the help of AI will go out without being reviewed, understood and verified by an OrderFlow developer. We have no intention of using vibe coding, where many of these jobs are passed on to AI itself.

AI in Support

Using AI to develop and test OrderFlow code happens behind the scenes, so it’s not something that will be visible to our users. However the next step has been to see how we can improve our support processes, the areas where you interact with us on a daily basis. This work is in its very early stages but it’s already clear AI is going to be useful in many areas of support activity. These range from producing customer documentation, writing tests for customer specific configuration changes, and identifying anomalies and patterns in OrderFlow logs that require investigation. Talk of using AI in software support may raise concerns about replacing personal customer service with automated chat-bots and call handling systems. This is definitely not our goal. The aim is to give the support team improved tools for providing a better personalised service, rather than attempting to automate our interaction with you.

AI in OrderFlow

The final part of the exercise will be to determine how AI might be incorporated into the OrderFlow platform and service itself. Adding an AI agent into OrderFlow would potentially allow appropriate users to perform tasks which previously would have required direct involvement of the OrderFlow team. For example, as well as answering ad hoc questions about the customer’s environment and data, AI could be used to allow customers to more easily create new custom reports and set up basic new configurations, in way that is safe but still self service.

Part of the benefit we are also expecting is that the use of AI will significantly lower barriers to entry in developing new OrderFlow features that add intelligence to the platform.

Privacy

An early concern we had with AI is that we would be giving away important commercial information by interaction with agents. We now understand that the Large Language Models (LLMs) we use are pretrained on an enormous but fixed data set. They don’t ‘learn’ from our interactions with them. Our access to LLMs is through a commercial subscription to the Cursor tool, with all members restricted to having ‘Privacy Mode’ enabled, which prevents accidental sharing of information outside of our conversations with AI agents.

Safety

We are now able to connect our internal AI toolset with customer environments. This is done on a strictly read only basis, with no write privileges allowed. We are able to use AI to help develop scripts, but these will always be viewed and run manually.

Commercial Impact

The great promise of AI is its potential to dramatically improve workforce productivity. From OrderFlow’s perspective, this means being able to deliver enhancements and customer solutions more quickly, to fix bugs more quickly, all with less effort.

It allows for great qualitative improvements, such as being able to provide more comprehensive testing and better documentation.

While there is likely to be a direct cost benefit through a reduction in the time to carry out specific tasks, it is still too early to quantify this, so we intend to be cautious in reducing estimates until the overall impact is well understood. However, as we generally provide capped estimates on development and solution preparation work — where you only pay for the time actually spent — you already stand to benefit from lower charges on equivalent jobs compared to before.

It is also important to understand that the use of AI tools involves a direct cost to OrderFlow for accessing the Large Language Models (LLMs) that these tools rely on. At present, we are absorbing these costs as part of our investment in learning how to use AI effectively and responsibly.

Our current assessment is that this cost is not substantial compared to labour, but it is by no means trivial, and it is difficult to predict how LLM pricing will evolve over time. To ensure transparency and fairness, we have developed internal mechanisms to understand how AI usage relates to specific support and development activities.

Looking ahead, if and when the benefits of AI usage are clear and measurable, we may begin to factor a proportion of these costs into customer billing. Before introducing any such change, we will ensure that the additional cost is more than offset by a corresponding reduction in billable time for equivalent work.

Summary

We’re very excited about the potential of AI and how it will benefit OrderFlow and our customers, and we’ve already started to see some of this potential being realised. We’re well aware of the need to be cautious and thoughtful in its adoption. We see AI as an assistive technology, with the capacity to greatly enhance rather than replace human efforts and customer service interactions. We also recognise that, alongside its benefits, AI introduces new costs, and that any future pass-through of these costs to customers must be clearly justified by demonstrable net benefits.

Yours,

Phil Zoio
OrderFlow CTO