Where IBISWorld Fits into Your Workflow
Whether you’re rolling out AI platforms for the first time or already using agentic workflows, integrating IBISWorld via MCP will likely be new territory.
Start by identifying where IBISWorld should support the work you already do. Ask yourself:
Which tasks require industry context?
What information should the AI retrieve from IBISWorld?
Where should that information appear in the final deliverable?
You may want IBISWorld to serve as a preferred source across an entire agent, support a specific step within a broader workflow or provide context only when certain types of questions arise. The right setup depends on how consistently the data is needed, how repeatable the task is and how much control you need over the final output.
Three Ways to Shape the Output
There are three main layers where customization can influence your output: prompts, instructions and skills. Deciding where to integrate IBISWorld will largely depend on the nature of your workflow and how often you plan to run similar tasks using IBISWorld data.
The diagram below can help you decide where to start customizing. Just remember: these functions are not mutually exclusive; depending on the use case, you may rely on one, two or all three.
Prompts
Definition: the natural-language question you (or your agent) ask in the moment. This is the fastest lever to pull, and the right place to start.
“Give me an industry snapshot for specialty coffee roasters, including market size, growth drivers, key companies and top risks. Cite the report title and year.”
Test it by: running a handful of prompts that reflect your most common asks and checking the output against the checklist below. If problems arise, try small wording changes first — before touching instructions or skills.
Instructions
Definition: the standing rules that sit behind the agent and apply to every response, regardless of what's asked (e.g., “always cite the report title and year”). These typically live at the account, project or agent level and may be owned by an administrator.
“For any question about industry size, growth drivers, competitive landscape or risk, always call the IBISWorld MCP server before using the web. Name the specific industry in your response, and cite the underlying report title and year.”
Test it by: running the same prompt across a few sessions or team members and confirming the rule holds consistently. Instructions are a good lever when the same rules will apply across many different prompts.
Skills
Definition: a more involved, reusable capability that goes beyond a single instruction. Skills can be quite long, and can call on other skills or tools in sequence (for example, a skill that pulls IBISWorld data and then hands off to an Excel skill to parse the data).
If your team already has skills built for a workflow (market sizing, credit memos, benchmarking), the goal isn't to write a brand new skill from scratch. Instead, you can update the skill you already have so it knows to reference IBISWorld, and knows what to do with that data once it's fetched. This can include the specific metrics required, the order in which they are presented, the level of detail, the format and any design or template conventions the output should follow.
Test it by: running your skill end-to-end on a real scenario and confirming IBISWorld shows up correctly at the point it's called. Because skills are more involved and often nested, this can take longer to validate than a prompt or instruction change.
Test & Evaluate: The Checklist
Whichever lever you're using, test against the same core criteria every time. These are the things that come up most often, but you should add your own if your use case has other requirements.
A general rule of thumb: if an issue shows up on one prompt, tweak the prompt. If it shows up across every prompt in a use case, tweak the instructions or the skill. If tweaking the wording doesn't move the needle at all, it may be a model or platform limitation worth raising with your IBISWorld Client Relationship Manager.
Criterion | What to look for | How to troubleshoot |
Right report | The industry matches what was asked, not relying on broader sector data or adjacent industries. | Tune your prompt specificity; if matching issues persist, add an instruction to confirm industry match. |
Citations | Report title, geography, year, and (if you need it) chapter are present and consistent. | Improve your instructions. Make the citation requirement explicit. |
Formatting & style | Structure (bullets vs. paragraphs, tables, headers) matches how your team actually wants to consume the content. | Write instructions for repeat needs; adjust prompt wording for one-off asks. |
Length & depth of analysis | The level of detail matches the decision it's supporting (not too shallow, not padded). | Adjust instructions or prompt for scope. Choose a higher-capability model if the source has the depth but the output doesn't. |
Accuracy | The data and claims match what's actually in the underlying report. | Spot-check against the source report directly; flag persistent mismatches to your IBISWorld Client Relationship Manager. |
Best Practice: Instructing Agents, Skills, Projects & More
These instructions are intended for inclusion in your organization's deployment of the IBISWorld MCP Server. They cover the required data integrity behaviors your AI platforms must follow when using IBISWorld data.
To preserve the accuracy and trustworthiness of our data, please add the following instructions to any skills, agents, projects or other customization layers where IBISWorld sits:
Data Integrity Requirements (Required Instructions for All Deployments)
Never fabricate or estimate a figure when a real one is available. If IBISWorld has published the number, use it. Do not approximate, round beyond what's in the source, or "fill in" a plausible-sounding figure.
Never derive or calculate a number that isn't explicitly reported. If the assistant needs a figure not directly published (e.g., a custom ratio or a multi-year CAGR), it must show its work and clearly label the result as a derived calculation, not a reported IBISWorld figure.
Always cite the source report and publication date. Every data point pulled from IBISWorld must be traceable back to the specific report and date it came from, with a source link where available.
Label historical vs. forecast data explicitly. Never present a forecast figure as if it were an actual historical result, or vice versa.
Respect geographic scope. Don't apply national data to a regional or local question (or vice versa) without flagging the mismatch.
Never silently blend data across reports. If pulling from multiple reports to answer one question, state which report each figure came from — don't merge them into a single unattributed number.
If no relevant data exists, say so. The assistant should state that IBISWorld does not cover the specific industry/region/metric requested, rather than substituting a plausible-sounding but unsourced answer.
Appendix: Definitions
Prompt: the natural-language question a user types in.
Instructions: the “always-on” rules that sit behind the agent or project and apply to every response. These sometimes live in an agent's configuration and thus might be owned by an administrator.
Skill: a more involved, reusable capability that can be nested, calling other skills or tools. Skills are built for a specific workflow rather than a single question.
Model: the underlying language model (e.g., GPT-5.5, Opus 4.7, Sonnet 5). Determines analytical depth and reasoning quality.
Platform: the AI application the user works in (e.g., Microsoft 365 Copilot, Claude, ChatGPT) and the mechanism that connects the data source to it (e.g., an MCP server). The same model can render the same content very differently depending on the platform.
Source: the data the agent draws from (e.g., the IBISWorld content set).
MCP Server: The system that connects an AI application to an external data source through the Model Context Protocol. It receives requests from the AI, retrieves the relevant information or action and returns it in a structured format that the AI can use in its response or workflow.