Acorporateintelligenceweek,withanAIagentdoingthefirstpass
What an AI agent prepared across a corporate intelligence week: competitor briefs, earnings readouts, company profiles and a strategy red-team, with costs.
If you work in strategy, competitive intelligence, legal or on a commodity desk, a big part of your week goes on one job. You find out what changed at the companies you watch, then explain what it means to someone who has ten minutes before their next meeting.
Most of that time goes on the finding. Reading three earnings releases to locate the one paragraph about pricing. Checking whether a supplier's plant shutdown made the trade press in German before it reached English. Rebuilding a competitor profile from last year because half of it is out of date.
An AI agent connected to the right sources can take a good share of that work. Below is one working week, laid out job by job: what the agent prepared, what it cost in data, and where a person still had to decide. The agent works inside an assistant such as Claude or ChatGPT, connected to Bigdata.com's news, filings and transcripts, and it cites the passages it used.
One note before we start. The cost figures are Bigdata.com data charges. Bigdata.com prices by usage, metered in tokens, so you pay for the content your agents actually consume. The figures come from earlier test runs of the closest matching workflows, which were built for equity analysts. The retrieval underneath is the same kind of work, so they give you a fair sense of scale. Treat them as examples from specific runs, and expect your own numbers to move with what the agent reads.
Monday: the competitor brief you used to build by hand
Start with a list. Your eight closest competitors, your three biggest customers, the two suppliers you'd be in trouble without. The agent checks that list against overnight news, filings and company announcements, groups what it finds by company, and writes a short brief with a source link beside each point.
Give it the same three questions each time. What changed? Does any of it touch us: pricing, capacity, leadership, a lawsuit, a contract win? What deserves a closer look today? Asking the same questions every morning is what turns a pile of headlines into a brief you can forward.
Foreign coverage is where this pays off quickly. If one of your suppliers is a mid-sized Korean manufacturer, the first report of a fire at its plant may not be in English. Bigdata.com covers news in more than 35 languages, so the agent can pick up the story, summarise it in English and link the original.
Where you come in: the agent doesn't know that a competitor's new CFO matters because she used to run the division you're about to go up against. Relevance is your call. You can make the brief sharper by telling the agent why each company is on the list. "Watch Supplier B for anything about capacity in Poland" gets a better result than a bare company name.
Observed data cost: $1.66, from the equivalent morning-briefing run.
Tuesday: a competitor reports, and you need the parts that affect you
A strategy or sales team doesn't need earnings per share against consensus. You need what the competitor said about the things that change your plans: pricing, volumes in the segment you share, capex, new markets, hiring or cuts, and any comment about customers you both chase.
Keep a standing template and the agent fills it the same way each quarter. It pulls the results release and, once it's published, the call transcript. Then it answers your questions with the quoted passage behind each answer. If the CEO spent two minutes on the call talking about "rationalising" a product line you compete with, that ends up at the top of your readout instead of buried on page 14.
Check two things before anything goes into a slide. First, open the source behind any number you plan to repeat: wrong quarter, wrong currency and segment figures mistaken for group figures are the usual slips. Second, watch the verbs. Management language is hedged on purpose, and "we are evaluating options" should never become "they plan to exit" in your summary.
Observed data cost: $0.68, from the equivalent post-earnings flash.
Wednesday: a profile of a company nobody has looked at yet
Someone in the leadership meeting asks about a company that isn't on anyone's list. It might be a new entrant, an acquisition target, a possible partner, or the other side in a dispute. By Friday they want to know who it is.
The agent can put together the first version: what the company does, where it operates, its segments and reported figures if it's listed, recent news, management commentary, litigation and regulatory mentions, and a timeline of the last two years. Ask for that timeline. It's the part people skim first and the part that takes longest to build by hand.
Be clear with yourself about private companies. For a listed company, filings and transcripts give you hard numbers. For a private one, the agent works from news coverage: funding rounds, contract wins, plant openings, executive moves, and the occasional revenue figure a founder mentioned in an interview. That's often enough for a rough sense of size and direction. It isn't a set of accounts, so ask the agent to label every estimate with where it came from and to say plainly where coverage runs thin.
The framing stays with you. Who does this company really compete with? Which of its numbers matter for your decision? Is its own story about itself believable? Those questions are the research, and the agent's draft is the material you answer them with.
Observed data cost: $1.79, from the equivalent initiation-report run.
Thursday: the strategy memo, and the case against it
Thursday has two jobs, and the second is the one people skip.
The first is a memo that organises evidence around a view your team already holds. "Competitor X is moving down-market." "Our largest customer is consolidating its suppliers." The agent gathers the evidence behind the main drivers, the risks and the early signals, then drafts a structured memo with citations.
The second is to ask it to argue the other side. "Find the strongest evidence that Competitor X is staying where it is." A memo built from evidence picked to support a view will tend to support it. Running the opposite question makes the agent go looking for what you'd rather not find, and you find it before your VP does in the meeting.
In our runs the challenge cost almost as much as the memo, despite a shorter output. That's normal. Cost follows how much the agent reads, and a good counter-argument means reading widely.
Observed data cost: $1.93 for the memo and $1.85 for the challenge, from the equivalent long-thesis memo and short-thesis pressure test.
What the week cost
The five jobs came to $7.91 in data across the runs we measured. That's what Bigdata.com charged for the content the agent pulled in. Its pricing is token-based: you're billed for what your agents consume, so a quick readout costs less than a deep profile. Here's how the week splits.
Job | What the agent prepared | Closest test run | Observed data cost |
|---|---|---|---|
Competitor brief | Overnight changes across a watchlist, grouped by company, with sources | Morning briefing | $1.66 |
Competitor earnings readout | Pricing, segment, capex and customer comments from a results release and call | Post-earnings flash | $0.68 |
Company profile | First-draft profile with figures, news, litigation and a two-year timeline | Initiation report | $1.79 |
Strategy memo | Evidence for a view, organised by driver, risk and signal | Long thesis memo | $1.93 |
The case against | Strongest evidence against that view | Short thesis pressure test | $1.85 |
Three things about the bill are worth knowing if you've been burned by open-ended pricing before. You pay for the passages the agent retrieves, so it doesn't charge you for whole documents it never needed. You set a spend limit, and the agent works inside it. And new accounts start with $10 in free credits, which at these figures would cover a week like this one with some room to spare.
Your own costs will differ. A watchlist of 40 companies costs more to brief than one of 12, and a thinly covered private company may need more searching than a large listed one. Bigdata.com's pricing page carries newer example estimates than the runs above.
What the agent took off your plate, and what it didn't
The pattern held all week. The agent was good at work that repeats and needs a lot of reading: collecting, sorting and putting the evidence in one place with a link back to where it came from. That matters most for the jobs you do every week or every quarter, because once the questions are written down you can run them again on new evidence with one prompt.
In practice that meant:
Checking the same watchlist against the same questions each morning.
Filling a standard readout every time a competitor reports.
Refreshing a company profile when something new is published, instead of rebuilding it.
Keeping a strategy memo tied to the evidence behind each point.
Running a deliberate challenge against a view the team already likes.
What stayed with people was everything that moved from "what the sources say" to "what we should do about it." Choosing the companies and the questions. Deciding which sources count. Checking any figure that would end up in a board pack or a court filing. And bringing the context the agent can't see: your own pricing, your deal pipeline, the conversation a sales lead had last Tuesday.
If you're thinking about setting this up, start with one recurring job you already do by hand, write down the questions you ask each time, and run it for two weeks next to your usual process. You'll see quickly where the draft saves you time and where you still need to dig.
If you work in strategy, competitive intelligence, legal or on a commodity desk, a big part of your week goes on one job. You find out what changed at the companies you watch, then explain what it means to someone who has ten minutes before their next meeting.
Most of that time goes on the finding. Reading three earnings releases to locate the one paragraph about pricing. Checking whether a supplier's plant shutdown made the trade press in German before it reached English. Rebuilding a competitor profile from last year because half of it is out of date.
An AI agent connected to the right sources can take a good share of that work. Below is one working week, laid out job by job: what the agent prepared, what it cost in data, and where a person still had to decide. The agent works inside an assistant such as Claude or ChatGPT, connected to Bigdata.com's news, filings and transcripts, and it cites the passages it used.
One note before we start. The cost figures are Bigdata.com data charges. Bigdata.com prices by usage, metered in tokens, so you pay for the content your agents actually consume. The figures come from earlier test runs of the closest matching workflows, which were built for equity analysts. The retrieval underneath is the same kind of work, so they give you a fair sense of scale. Treat them as examples from specific runs, and expect your own numbers to move with what the agent reads.
Monday: the competitor brief you used to build by hand
Start with a list. Your eight closest competitors, your three biggest customers, the two suppliers you'd be in trouble without. The agent checks that list against overnight news, filings and company announcements, groups what it finds by company, and writes a short brief with a source link beside each point.
Give it the same three questions each time. What changed? Does any of it touch us: pricing, capacity, leadership, a lawsuit, a contract win? What deserves a closer look today? Asking the same questions every morning is what turns a pile of headlines into a brief you can forward.
Foreign coverage is where this pays off quickly. If one of your suppliers is a mid-sized Korean manufacturer, the first report of a fire at its plant may not be in English. Bigdata.com covers news in more than 35 languages, so the agent can pick up the story, summarise it in English and link the original.
Where you come in: the agent doesn't know that a competitor's new CFO matters because she used to run the division you're about to go up against. Relevance is your call. You can make the brief sharper by telling the agent why each company is on the list. "Watch Supplier B for anything about capacity in Poland" gets a better result than a bare company name.
Observed data cost: $1.66, from the equivalent morning-briefing run.
Tuesday: a competitor reports, and you need the parts that affect you
A strategy or sales team doesn't need earnings per share against consensus. You need what the competitor said about the things that change your plans: pricing, volumes in the segment you share, capex, new markets, hiring or cuts, and any comment about customers you both chase.
Keep a standing template and the agent fills it the same way each quarter. It pulls the results release and, once it's published, the call transcript. Then it answers your questions with the quoted passage behind each answer. If the CEO spent two minutes on the call talking about "rationalising" a product line you compete with, that ends up at the top of your readout instead of buried on page 14.
Check two things before anything goes into a slide. First, open the source behind any number you plan to repeat: wrong quarter, wrong currency and segment figures mistaken for group figures are the usual slips. Second, watch the verbs. Management language is hedged on purpose, and "we are evaluating options" should never become "they plan to exit" in your summary.
Observed data cost: $0.68, from the equivalent post-earnings flash.
Wednesday: a profile of a company nobody has looked at yet
Someone in the leadership meeting asks about a company that isn't on anyone's list. It might be a new entrant, an acquisition target, a possible partner, or the other side in a dispute. By Friday they want to know who it is.
The agent can put together the first version: what the company does, where it operates, its segments and reported figures if it's listed, recent news, management commentary, litigation and regulatory mentions, and a timeline of the last two years. Ask for that timeline. It's the part people skim first and the part that takes longest to build by hand.
Be clear with yourself about private companies. For a listed company, filings and transcripts give you hard numbers. For a private one, the agent works from news coverage: funding rounds, contract wins, plant openings, executive moves, and the occasional revenue figure a founder mentioned in an interview. That's often enough for a rough sense of size and direction. It isn't a set of accounts, so ask the agent to label every estimate with where it came from and to say plainly where coverage runs thin.
The framing stays with you. Who does this company really compete with? Which of its numbers matter for your decision? Is its own story about itself believable? Those questions are the research, and the agent's draft is the material you answer them with.
Observed data cost: $1.79, from the equivalent initiation-report run.
Thursday: the strategy memo, and the case against it
Thursday has two jobs, and the second is the one people skip.
The first is a memo that organises evidence around a view your team already holds. "Competitor X is moving down-market." "Our largest customer is consolidating its suppliers." The agent gathers the evidence behind the main drivers, the risks and the early signals, then drafts a structured memo with citations.
The second is to ask it to argue the other side. "Find the strongest evidence that Competitor X is staying where it is." A memo built from evidence picked to support a view will tend to support it. Running the opposite question makes the agent go looking for what you'd rather not find, and you find it before your VP does in the meeting.
In our runs the challenge cost almost as much as the memo, despite a shorter output. That's normal. Cost follows how much the agent reads, and a good counter-argument means reading widely.
Observed data cost: $1.93 for the memo and $1.85 for the challenge, from the equivalent long-thesis memo and short-thesis pressure test.
What the week cost
The five jobs came to $7.91 in data across the runs we measured. That's what Bigdata.com charged for the content the agent pulled in. Its pricing is token-based: you're billed for what your agents consume, so a quick readout costs less than a deep profile. Here's how the week splits.
Job | What the agent prepared | Closest test run | Observed data cost |
|---|---|---|---|
Competitor brief | Overnight changes across a watchlist, grouped by company, with sources | Morning briefing | $1.66 |
Competitor earnings readout | Pricing, segment, capex and customer comments from a results release and call | Post-earnings flash | $0.68 |
Company profile | First-draft profile with figures, news, litigation and a two-year timeline | Initiation report | $1.79 |
Strategy memo | Evidence for a view, organised by driver, risk and signal | Long thesis memo | $1.93 |
The case against | Strongest evidence against that view | Short thesis pressure test | $1.85 |
Three things about the bill are worth knowing if you've been burned by open-ended pricing before. You pay for the passages the agent retrieves, so it doesn't charge you for whole documents it never needed. You set a spend limit, and the agent works inside it. And new accounts start with $10 in free credits, which at these figures would cover a week like this one with some room to spare.
Your own costs will differ. A watchlist of 40 companies costs more to brief than one of 12, and a thinly covered private company may need more searching than a large listed one. Bigdata.com's pricing page carries newer example estimates than the runs above.
What the agent took off your plate, and what it didn't
The pattern held all week. The agent was good at work that repeats and needs a lot of reading: collecting, sorting and putting the evidence in one place with a link back to where it came from. That matters most for the jobs you do every week or every quarter, because once the questions are written down you can run them again on new evidence with one prompt.
In practice that meant:
Checking the same watchlist against the same questions each morning.
Filling a standard readout every time a competitor reports.
Refreshing a company profile when something new is published, instead of rebuilding it.
Keeping a strategy memo tied to the evidence behind each point.
Running a deliberate challenge against a view the team already likes.
What stayed with people was everything that moved from "what the sources say" to "what we should do about it." Choosing the companies and the questions. Deciding which sources count. Checking any figure that would end up in a board pack or a court filing. And bringing the context the agent can't see: your own pricing, your deal pipeline, the conversation a sales lead had last Tuesday.
If you're thinking about setting this up, start with one recurring job you already do by hand, write down the questions you ask each time, and run it for two weeks next to your usual process. You'll see quickly where the draft saves you time and where you still need to dig.
Frequently asked questions
No. It's a set of recurring corporate intelligence jobs arranged as a week. The cost figures come from test runs of the closest equivalent workflows, which were originally built for equity research.
An assistant that can connect to outside data, such as Claude or ChatGPT, a Bigdata.com account, and a written list of the companies you track and the questions you want answered each time. Bigdata.com connects through MCP, so you work in the assistant you already use.
Sometimes. When a private company's revenue has been reported in the news or stated by its executives, the agent can find it and cite it. When it hasn't, the agent should say so and show what it based any estimate on.
It can. News coverage spans more than 35 languages, and the agent summarises in English with the original source linked. How much it finds depends on how much has been published about the company.
Set a spend limit before you start. You pay for the passages the agent reads, and new accounts begin with $10 in free credits, so you can test a real job before committing a budget.
Developers
© 2026 Bigdata.com by RavenPack. All rights reserved.
Developers
© 2026 Bigdata.com by RavenPack. All rights reserved.