Think Small to Develop AI Strategy
In this session, Mark Jamieson (President and CEO of 2W Tech) and Tom Geairn (CTO of 2W Tech) discuss the current state of AI in business, practical applications, and how to implement AI effectively within your organization.
+ Full Transcript
How to Develop an AI Roadmap and Strategy for Your Business
Moderator: Welcome and thank you for joining us today. We’re discussing how to develop an AI roadmap and the strategy that goes with it.
Today’s presenters are Mark Jamieson, President and CEO of 2W Tech, and Tom Garren, CTO.
For those of you who aren’t familiar with 2W Tech, we’re a technology solutions and managed services provider specializing in manufacturing and distribution, with expertise across Epicor ERP, Microsoft technologies, cloud, cybersecurity, data, and artificial intelligence.
Where Should Businesses Start with Artificial Intelligence?
Mark: AI is everywhere right now. We’re seeing huge investments and announcements from technology companies, software vendors, ERP providers, and cloud platforms.
That creates an obvious question for business leaders: How do we actually take advantage of it?
The answer doesn’t necessarily begin with the biggest AI project you can imagine.
Sometimes, the best use of AI is a relatively small, clearly defined problem where you can create measurable value.
Before we get there, let’s define some of the technology we’re talking about.
What Is Natural Language Processing (NLP)?
Tom: Natural language processing, or NLP, refers to technologies that allow computers to work with human language.
That can include interpreting text, categorizing messages, extracting meaning, converting speech to text, generating text, and allowing people to interact with software using normal language.
For businesses, that’s important because a significant amount of company information exists in language: emails, documents, customer communications, service tickets, meeting transcripts, SOPs, product documentation, and other unstructured content.
What Is a Large Language Model (LLM)?
Tom: Large language models, or LLMs, are AI models trained on large amounts of information so they can understand and generate language.
General-purpose models can work across a broad range of topics, while business AI applications can add organizational context or ground those models in approved company information.
That’s where AI starts becoming significantly more useful to a business.
Instead of only asking a general AI question, you can begin asking questions that depend on your own documents, products, processes, customers, and business data.
What Types of AI Tools Are Available to Businesses?
Mark: There are several categories of AI tools businesses should understand.
General-purpose AI assistants: These are conversational AI experiences that can help with research, brainstorming, writing, explanation, summarization, analysis, and other broad tasks.
Application-specific AI: Technologies such as Microsoft Copilot, GitHub Copilot, and AI capabilities embedded inside ERP and business applications are designed to assist with particular workflows and applications.
Business-specific AI: Organizations can also create AI experiences that are grounded in their approved business information and designed to support specific workflows, processes, or audiences.
The important point is that there isn’t one AI tool that’s automatically best for every business problem.
Choose the tool based on the use case.
How Should Businesses Protect Sensitive Data When Using AI?
Tom: Before putting business information into an AI service, understand how that service handles your data.
Ask where information is processed and stored, what identity and access controls apply, whether prompts or outputs are retained, whether organizational data is used to train models, and what compliance and governance controls are available.
Sensitive financial information, customer data, intellectual property, employee records, and regulated business information require additional care.
Enterprise platforms can provide stronger organizational security and governance controls, but those controls still need to be configured and managed properly.
What Is the First Step in Creating an AI Roadmap?
Mark: Start with the use case.
Don’t begin by asking, “What can AI do?”
Start by asking:
Where is the friction in our business? What is too manual? Where are employees spending too much time looking for information? Which repetitive tasks slow people down? Where do we have bottlenecks, delays, or inconsistent processes?
Then ask where AI could help.
That’s a much easier and more productive question than trying to build an enormous AI program before you’ve identified the business problem.
Tom: Starting with “What could AI do with all of our company data?” can become overwhelming very quickly.
Even relatively small organizations have years or decades of documents, emails, spreadsheets, policies, product information, and other data.
Start with a specific workflow and a specific outcome.
Why Should an AI Strategy Start Small?
Mark: Think small and look for quick wins.
AI tends to work particularly well when you’re summarizing large amounts of information, finding information across documents, automating repeatable work, categorizing information, or combining those capabilities into a workflow.
Instead of trying to make AI run your entire company, find something specific that wastes time today.
Solve it. Measure the result. Then find the next opportunity.
AI Use Case: Analyzing Complex Excel Spreadsheets
Tom: Almost every business has what we jokingly call a “spreadsheet from hell.”
It’s the giant Excel workbook with multiple tabs, formulas, data sources, historical information, and logic that only one or two people really understand.
AI can help employees interact with those complicated datasets using natural language.
Instead of understanding every formula and worksheet, an employee may be able to ask a specific question about the information and receive a summarized answer or identify relevant trends.
This can be an excellent starting point because the data already exists and the business problem is easy to understand.
AI Use Case: Searching Product Documentation and PDFs
Mark: Another common opportunity is product documentation.
Maybe your customer service team has hundreds of product sheets, specification documents, manuals, or historical PDFs.
Instead of manually searching through all of those documents, employees could ask a question in natural language and retrieve information from the relevant approved documentation.
The same idea can help with replacement parts, product compatibility, specifications, troubleshooting information, sales questions, and customer support.
AI Use Case: Risk Assessment and Data Protection
Tom: AI can also support security, compliance, and data governance.
Organizations can monitor business information for sensitive data and look for situations where that information may be stored, shared, or transmitted inappropriately.
That might include payment information, credentials, access keys, regulated data, confidential business records, or other sensitive content.
Technologies such as Microsoft Purview can help organizations classify, protect, and govern information across their Microsoft environment.
Automated classification and monitoring can also help identify sensitive documents that have accidentally been stored in overly permissive locations.
AI Use Case: Cybersecurity and Vulnerability Management
Tom: Security teams can use AI and automation to process large volumes of cybersecurity information and prioritize potential problems for human investigation.
Instead of expecting people to manually monitor every security signal, automated systems can surface the issues that appear to require attention.
That’s especially valuable as environments become larger and more complex.
AI Use Case: Billing and Invoice Validation
Mark: Here’s an example from our own business.
Billing involves several sources of information: client agreements, consultant time entries, invoicing rules, and the resulting invoice.
AI-assisted automation can compare those inputs and identify situations where the information doesn’t align.
Instead of manually reviewing every invoice, employees can focus their attention on exceptions requiring human judgment.
That’s an important AI pattern: automate the normal and send the exceptions to people.
AI Use Case: Recruiting and Resume Processing
Tom: Recruiting produces a lot of unstructured information too.
Resumes, questionnaires, job descriptions, and related information can arrive in many different formats.
AI can help organize and summarize that information into a more consistent format for human review.
It can also help generate relevant follow-up questions or identify information that reviewers may want to investigate further.
AI shouldn’t make consequential employment decisions on its own, but it can reduce administrative work around organizing information for the people responsible for those decisions.
AI Use Case: Meeting Recaps and Follow-Up
Mark: Meeting follow-up is another practical use case.
A meeting can be transcribed, and AI can help turn that transcript into a summary, action items, decisions, outstanding questions, and a draft follow-up message.
A person then reviews the output before anything is distributed.
That can reduce a significant amount of administrative work while creating more consistent follow-through after meetings.
AI Use Case: Service Ticket Routing and Customer Support
Tom: Service requests are another strong use case because they’re repeatable and often involve a lot of manual classification.
AI can help summarize a request, identify the likely issue, route the ticket to the appropriate team, recommend a priority, identify missing information, and prepare an initial response.
It can also use screenshots, error messages, documentation, and previous service information to give technicians additional context.
Almost every business has some variation of this workflow, even if it doesn’t formally call the work a “service ticket.”
AI Use Case: Standard Operating Procedures and Tribal Knowledge
Organizations can also ground AI experiences in approved SOPs, work instructions, policies, and other internal documentation.
Instead of asking the employee sitting next to them, a user could ask, “How do we handle a nonconforming part here?” or “What’s our procedure for entering this type of order?”
The important distinction is that the answer should come from the organization’s approved process documentation rather than a generic internet answer.
This can help make institutional knowledge easier to find and reduce dependence on tribal knowledge.
Can AI Translate Internal Business Knowledge?
AI can also make organizational information more accessible across languages.
An organization’s approved documentation may exist in English while an employee prefers to ask questions or receive explanations in another language.
With appropriate human review for important instructions, AI can help bridge that language gap and make approved business knowledge more accessible to employees.
How Can AI Help Create Presentations?
Mark: Even this webinar is an example.
We started with a conversation about what we wanted the webinar to cover.
We recorded and transcribed that conversation, turned the transcript into structured content, and then used AI to help create an initial presentation.
From there, humans refined it, researched additional information, reorganized the narrative, and created the finished webinar.
That’s another useful AI pattern: don’t necessarily ask AI to create the final product.
Use AI to dramatically accelerate the first draft, then apply human expertise and judgment.
How Do You Prepare Business Data for AI?
Mark: Once you’ve identified your use cases, the next step is preparing the data those use cases require.
The good news is that when you start with small use cases, you don’t need to clean every piece of data your company has before you begin.
Identify the specific data required for the use case.
Which documents are authoritative? Are there duplicates? Is anything obsolete? Is important information missing? Who should have access?
Clean and organize the information that matters for this problem first.
How Can Microsoft Purview Help Prepare Data for AI?
Tom: Microsoft Purview can help with information protection, data governance, classification, and identifying sensitive information.
Organizations can use built-in and custom classifications to help identify different types of documents and data.
That can help determine which information is confidential, regulated, outdated, irrelevant, or appropriate for a particular AI use case.
Data governance becomes increasingly important as AI makes information easier for employees to discover.
Does Your AI Strategy Require Cloud Infrastructure?
Mark: Cloud platforms can make AI adoption significantly easier because many modern AI, data, security, collaboration, and automation services are designed around cloud architectures.
For organizations heavily invested in Microsoft technology, services such as Microsoft 365, SharePoint, Azure, and other Microsoft cloud platforms can provide a foundation for AI initiatives.
That doesn’t mean every workload has to move to the cloud immediately.
Hybrid and specialized on-premises scenarios still exist.
But the organization’s data architecture should make the required information securely accessible to the AI services supporting each approved use case.
What Should an AI Roadmap Include?
A practical AI roadmap can follow a straightforward process:
1. Assess.
Understand your business goals, current technology, data environment, security posture, and AI adoption that’s already happening.
2. Identify Use Cases.
Look for specific business problems where AI could reduce friction, improve access to information, automate repetitive work, or accelerate decisions.
3. Prioritize Quick Wins.
Start with use cases that are achievable, measurable, and meaningful enough to prove value.
4. Prepare the Data.
Identify the relevant information, clean it, classify it, establish appropriate permissions, and determine the authoritative source.
5. Choose the Right AI Tool.
Determine whether the use case belongs in Microsoft Copilot, an ERP-focused AI tool, an enterprise AI platform, a custom agent, business intelligence, automation, or another solution.
6. Build Security and Governance Around It.
Define approved tools, access, sensitive-data rules, monitoring, ownership, and human-review requirements.
7. Pilot and Measure.
Establish a baseline and determine whether the AI workflow actually improves time, cost, quality, accuracy, employee experience, or another business outcome.
8. Scale What Works.
Once you’ve proven one use case, apply the lessons to the next one.
9. Continuously Improve.
AI tools change. Your business changes. Your data changes. Treat the roadmap as an ongoing program rather than a one-time project.
How Do You Get Leadership and Employees to Buy Into AI?
Mark: A lot of AI conversations are starting at the CEO or board level.
Leadership says, “We need to do AI.”
The real job is translating that direction into actual business goals and practical use cases.
Tom: And adoption can come from the other direction too.
Employees may already be using AI for writing, spreadsheets, programming, purchasing, content creation, research, or other everyday tasks.
That creates a different set of leadership questions.
How do we measure ROI? How much is this costing? Is company data safe? Are employees using approved AI tools? Which workflows should be standardized? What skills and training do employees need?
A successful AI strategy has to address both sides: leadership strategy and employee adoption.
AI Roadmap and Strategy: Webinar Q&A
How Much Does an AI Project Cost?
Mark: It depends heavily on the scope.
A focused use case involving a small set of documents or a specific workflow has a completely different cost profile than a large enterprise AI platform spanning millions of documents and multiple business systems.
Costs can include design and implementation, data preparation, storage, indexing or ingestion, AI model consumption, integrations, licensing, governance, and ongoing support.
That’s another reason to start small.
A narrowly defined pilot makes it much easier to compare the implementation and operating costs with a measurable business benefit.
Is AI Usage Pay-As-You-Go?
Tom: Some cloud AI services use consumption-based pricing while others may use per-user licensing, capacity pricing, subscriptions, or combinations of those models.
The amount of information being processed, how frequently the AI is used, which model is selected, the amount of data indexed, and the architecture of the solution can all affect cost.
Don’t assume AI consumption is automatically expensive, but don’t deploy it without a method for monitoring costs either.
Do You Need Epicor Prism to Use AI with Epicor Data?
Mark: The answer depends on the use case.
Epicor AI capabilities can provide valuable experiences inside the ERP, but they aren’t necessarily the only way to analyze Epicor data.
Epicor information can also be exposed through supported APIs, reporting structures, business intelligence platforms, data integrations, or other approved architectures.
Tom: Use the best tool for the job.
A broad ERP AI experience and a narrowly defined integration can solve very different problems.
Which AI Applications Should Businesses Use?
There’s no single answer.
Microsoft Copilot might be ideal for productivity work inside Microsoft 365. An ERP-specific AI capability might be better for ERP workflows. A research-oriented AI tool might be better for external research. A business-specific AI platform might make sense for a secure, specialized internal workflow.
Don’t develop tunnel vision around one product.
Start with the business problem and then choose the technology.
How Can AI Affect Workplace Culture?
Mark: One of the best uses of AI is reducing the amount of mundane administrative work employees have to do.
That can give people more time for customer relationships, problem-solving, analysis, innovation, and other work where human expertise matters more.
But organizations also need to recognize that employees may have legitimate concerns about how AI affects their roles.
Be transparent about why AI is being introduced, what workflows are changing, where human judgment remains essential, and how employees will be trained to use the technology.
Change management should be part of your AI roadmap.
Is AI Helping or Hurting Cybersecurity?
Tom: Both.
From a defensive perspective, AI can help security teams analyze large volumes of security information, identify suspicious patterns, prioritize incidents, isolate compromised systems, and accelerate investigation and response.
Attackers can use AI too.
AI can make social engineering more convincing, accelerate reconnaissance, help generate phishing content, and automate parts of an attack.
That means cybersecurity awareness, vulnerability management, identity protection, endpoint security, monitoring, and incident response become more important, not less.
Final Takeaway: An AI Roadmap Starts with Business Problems, Not AI Products
The goal of an AI strategy isn’t to use as much AI as possible.
It’s to identify where AI can create meaningful business value.
Start with the use case. Think small. Prepare the right data. Protect it. Choose the right tool. Measure the outcome.
Then stack the wins.
One successful workflow becomes proof for the next one. Over time, those small improvements can grow into a much broader AI strategy without requiring the organization to bet everything on a massive transformation project on day one.
Moderator: Thanks, everybody, for joining us today.