A.I. Tools - a guide to business use

Generative AI tools are now available in every major cloud application. The collaboration platforms all provide an AI add-on tool: Microsoft 365 CoPilot, Google Gemini, Zoom Workplace. And the major AI model developers all have standalone chat bots: Claude from Anthropic, ChatGPT from OpenAI, Perplexity, etc. And, on top of all of that, the major search engines have all integrated AI results into their interfaces.

So, what makes sense for your business?

3 things to consider

• How do you see AI tools being most useful to your company?
  • Is there specific information in your business that needs to be available to the AI tools?
  • Are there specific tasks that you can identify that could be improved with AI access?
• How much effort (time and money) do you want to spend on providing AI tools?
• What risks most concern you? Information leaks, errors in responses, time wasting?
  • All reputable AI services provide assurance that the data that you provide to the agent will not be used to train their models. When we consider the potential for information leakage, we are referring to changes you may make in your access and security policies to enable greater functionality from the AI agents. Whether or not the AI companies are hoovering up every piece of data that they can get access to, irrespective of their legal commitments, is an existential question beyond the scope of this article.

AI Decision Matrix

General AI use

If you can't think of specific information that your staff could leverage, or any specific tasks that you think could be improved with AI then there is little reason to spend much effort on deploying the tools, and your primary concern may be around the risks.

In this case, JustWorks can help you control the availability of the AI tools built into the platforms, and the deployment of any chat bots. 

Specific AI use cases

If there is proprietary business information that your staff could benefit from being able to interrogate using natural language, without having to escalate their query, and being able to find useful information themselves, then you have a case for targeted AI tools.

Low risk – Low effort

InfoWorks Assist

JustWorks can integrate your proprietary business information into our InfoWorks Assist agent so that your staff can get instant answers to their queries. InfoWorks Assist is tightly linked to your IT systems so that you can be sure that only the right people have access to the information they need.

The Assist agent is a zero effort deployment and requires no configuration by you. It is built in to JustWorks services and accessible from any web browser. It is available as a chat bot today and will get a voice interface in 2026.

InfoWorks Assist is platform independent so it doesn't matter whether you use Microsoft 365 or Google Workspace for your collaboration platform and is accessed from the same login page that your employees access all of the other applications your business uses.

Because InfoWorks Assist can only answer questions based on the knowledge provided to it, you retain maximum control, limit errors in responses, and minimise wasteful usage that might reduce productivity.

Depending on your service level with JustWorks, your proprietary business knowledge is refreshed every quarter, every month, or on a continual basis.

Medium risk – Medium effort

Microsoft Copilot and Google Gemini

The AI agents from the major collaboration platforms are built to leverage all of the data that you normally store on these platforms anyway. For instance, your email and files.

Microsoft Copilot has a broad range of third-party integrations allowing you to enable access to other data stored on other platforms such as Egnyte , Salesforce, et cetera.

Google Gemini is built on the basis that you will keep the data that you want it to include in its responses on the Google platform, either in Gmail or in Google Drive.

There are three major components to the effort required to make these platform AI tools work for your business:

1. Extra license costs
   • The basic versions of Copilot and Gemini are intended to be teaser products rather than functional AI tools for your business. To get specific business value out of these tools you will need to purchase additional licensing to access more functions.
2. Data management
   • Either getting your business proprietary data into one of the standard data stores, or setting up the connectors necessary to enable the AI agents to access data that you keep somewhere else, requires configuration and ongoing maintenance. Controlling access and securing your data requires careful planning and deployment.
3. Training
   • Infamously, generative AI tools can produce erroneous or even hallucinated responses. This means that there will be a training burden that your business will have to assume alongside the deployment of the AI tools.

The risks that you run by enabling access to these platform AI agents include:

• Productivity loss. Because these are generic agents, even though they have access to your specific business information, they are going to process requests using both your specific information and their generic capabilities. Staff may end up using the agents to perform all sorts of tasks or activities which you had not expected or intended.
• Data security. Getting your proprietary business information into a place where these platform AI agents can access it and use it is likely to require that you either move the data from what is currently a highly secure and controlled environment which you have been using for years, or enable a connection into the current storage location. In both cases, this increases the complexity and overhead of ensuring that access is controlled and secure.
• Errors. The more information in AI agent has access to, and the less instructions and direction that it is given, the more likely an AI agent is to produce erroneous or hallucinated responses. Because these platforms focus on broad use and general access to data sets, they offer less control over how that data gets used in the reasoning processes of the AI tools. 

JustWorks recommendation

Using the generic tools provided by the major collaboration platforms has the potential to be marginally useful across your business functions and staff. They can help with routine tasks such as formatting and spell checking emails, writing up notes from meetings, and providing guidance on the best ways to use standard productivity apps and features. In these ways they may be useful in improving general and administrative productivity across the business. Depending on your business these functions may justify any extra costs.

If you are looking to leverage specific and proprietary business information to generate value from specific use cases, then the standard versions of these tools are unlikely to provide you with useful functionality. You may be better off just enabling access to these tools without giving them access to your proprietary business information.

Low risk – High effort

Deploying dedicated chat bots from the major AI developers is a great way to leverage proprietary business information. And the ability to closely instruct AI agents means you can get advantage from using these powerful tools to achieve specific business objectives.

But, and it's a big but, to make these tools useful requires a considerable amount of effort. The effort falls into three categories:

1. Additional license costs, on a per user basis. 
   • You can expect to spend $300 - $500 a year per person.
2. A lot of time and effort spent curating the data and building the parameters for task specific agents.
   • Simple use case. In our experience, it takes an average of about three hours to set up a relatively simple project with custom instructions and a custom set of curated data. 
   • Advanced use case. This is a situation in which you have a more complex task that you feel can be beneficially automated and augmented with generative AI. In our experience, it takes an average of about twenty hours to set up a more complex project with fine tuned instructions and a custom connections to live or external data sources. 
3. Maintenance. Like any software project, custom AI agents require ongoing maintenance otherwise their functionality degrades over time. A rough guide here would be that you are going to spend the same amount of time that it took to set up a project, to maintain it each year.

Budget estimating

So as a rough estimate of the budget necessary to design and deploy 5 simple use cases and 1 advanced use case to 10 employees:

• $4,000 in licensing for 10 users
• $10,000 for 5 simple projects
• $12,000 for 1 complex project

So that would be $26,000 in the first year, with an ongoing maintenance cost of $15,000. That's going to be only a fraction of the cost of employing an additional person, but a careful analysis of your return on investment is always worth doing. You would have to increase the productivity of those users by more than a few percentage points to justify the expense.

You will also need to factor in the training costs for your employees as making effective use of these dedicated AI agents does require understanding and skills from the users.

InfoWorks Pro

Using dedicated agents from the major AI developers can have real business advantages. However, in our experience, this will mostly come from being able to add additional services to your business. Only looking to improve productivity is going to be more difficult business case to justify.

If you are looking to leverage specific and proprietary business information to generate value from specific use cases, then these tools are your best choice. You will need be ready to enable secured access to your proprietary business information and develop projects that direct agents to achieving specific objectives.

Reach out to us to find out how you might be able to use the JustWorks InfoWorks Pro service to help you get the best returns from your custom AI project.

Medium risk – High effort

Deploying dedicated chat bots from the major AI developers on your own is certainly possible and may even appeal due to the lower up front effort seemingly required.

You can sign up for a team account for Claude or ChatGPT and just give your users access and encourage them to experiment and develop novel use cases. If your staff are competent, trained, and you set appropriate guardrails, this can work.

But the risks are worth noting before you go down this road.

• Data security and compliance requirements. Without careful controls and a good understanding of the business' compliance obligations for maintaining the privacy and security of personal and business data, these tools make it easy to ingest and expose proprietary information. Users may set up connections to external data sources in order to access one particular data set but end up unintentionally exposing other data. There is vastly more effort being made today to enable agents to access data than there is attention being paid to control and audit the agents' use of the data.
• Errors. Instructing and directing agents is a skill that you will have to develop in the staff that have access to these agents. Your users will also need to employ suitable levels of skepticism about the outputs from the agents and be prepared to put the time into double checking results that they are given.
• Productivity loss. Staff may end up using the agents to develop projects and perform all sorts of tasks or activities which you had not expected or intended.

Unleashing AI agents across your organization could enable your business to move rapidly into new areas and to develop new product products or services. But before you do this, you should carefully consider both the training requirements and the controls and guardrails that are essential for minimising the risks.

If you do want to do this, you should definitely contact JustWorks and we can help advise on how to set up controls that will ensure that you maintain visibility and promote responsibility.