Financial Analyst
Financial analysis requires making sense of vast amounts of data spread across earnings reports, financial statements, analyst reports and more. Manually combing through all this information to gain useful insights can be extremely tedious and time consuming. This is where AI techniques like document understanding and conversational interfaces can help. In this post, we'll explore how a financial analyst can integrate documentAI into their workflow to quickly build a chatbot assistant that can read and comprehend their collection of financial documents. By uploading reports and statements, an analyst can instantly start having natural conversations with the assistant to get answers, discover connections, and uncover insights faster.
Benefits of a chat interface
A chat interface for financial analysis has several advantages:
- Quick access to documents - Analysts can ask questions and rapidly retrieve information from financial documents without having to manually locate and search through them.
- Connecting insights - The assistant can identify relationships and connections between data points across multiple reports to uncover trends and patterns.
- Summarization - Provides quick summaries of reports or earnings calls to get the key facts and highlights.
- Forecasting assistance - Can help generate financial forecasts and projections based on historical data and trends identified across documents.
- Competitive benchmarking - Enable comparative analysis against financials of peer companies to put performance in context.
- Personalization - Can tailor examples and responses based on the analyst's company and sector context.
- Multilingual - Supports interacting in the analyst's preferred language through machine translation.
- 24/7 availability - Provides timely access to financial insights anytime without having to wait for human assistance.
- Revealing blindspots - Identifies overlooked risks, changes or inconsistencies by comprehensively connecting data points.
- Efficiency - Saves significant time by automating manual financial analysis tasks. Lets analysts focus on higher value work.
Get your API key
First get your API key from the console. Authenticate requests by passing this key in the X-API-KEY header. Learn more about authentication here.
Upload your reports
There are 2 options to ingest the reports, you can either download an external document or you can upload it directly.
Downloading
The easiest way to ingest reports is to download them. See documentation for details.
curl -X POST https://api.documentai.dev/v1/collections/[COLLECTION ID]/document \
-H 'X-API-KEY: [YOUR API KEY]' \
-d '{"url": "[REPORT URL]"}'Where:
- COLLECTION ID - ID of the collection, reuse existing or specify new.
- API KEY - the API key from the console.
- REPORT URL - Link to your report, e.g. https://www.apple.com/newsroom/pdfs/fy2023-q3/FY23_Q3_Consolidated_Financial_Statements.pdf.
Downloading is asynchronous, the response will contain a documentId, using which you can check status as per documentation.
{
"collectionId": "[COLLECTION ID]",
"documentId": "b0dd63b1-f78c-4989-94f9-f73768bb12dd",
"status": {
"date": "2023-09-03T12:22:36.291790Z",
"status": "QUEUED"
}
}Uploading files
If you have your reports stored locally you can also upload them directly. To do this you can use the upload API, full reference here.
curl -X PUT https://api.documentai.dev/v1/collections/[COLLECTION ID]/upload \
-H 'X-API-KEY: [YOUR API KEY]' \
--form file=may_report.pdf \
--form file=june_report.docxWhere:
- COLLECTION ID - ID of the collection, reuse existing or specify new.
- API KEY - the API key from the console.
You can upload multiple files if needed in a single request. The upload is asynchronous and you can monitor the status of your upload job using check upload API. The response will contain an uploadId.
{
"collectionId": "[COLLECTION ID]",
"uploadId": "6f207f16-c30b-47ef-9a58-efea9df9ae73"
}Integrate chatbot
After the download or upload is completed you can integrate with the chat API. Full reference is available here. The conversational interface provides a natural UX for analysts to ask questions about the content of their uploaded financial documents. The assistant can leverage its extensive business and finance knowledge to not just retrieve data but provide useful insights that connect the dots across reports. This helps the analyst quickly understand key points within documents, identify trends and relationships between them, and derive higher-level conclusions from their collection. The assistant augments the analyst's own expertise through the knowledge encoded in the language model to uncover non-obvious insights. Rather than simply searching for answers, it can provide strategic recommendations relevant to the analyst's specific context and objectives.
curl -X POST https://api.documentai.dev/v1/collections/[COLLECTION ID]/chat/[CHAT ID] \
-H 'X-API-KEY: [YOUR API KEY}' \
-d '{"message": [YOUR MESSAGE]}'Where:
- COLLECTION ID - ID of the collection, reuse existing or specify new.
- API KEY - the API key from the console.
- CHAT ID - the id of the chat. You decide the id of the chat and do not need to explicitly create it. All previous messages in a chat will be evaluated by the LLM.
- MESSAGE - the message you want to send.
{
"sender": "ASSISTANT",
"message": {
"id": "553beb15-9f2b-4c97-857f-94963bbce84f",
"date": "2023-09-03T12:24:09.026440Z",
"content": "According to the document, Apple Inc. sold 39,669 iPhones in the most recent reporting period.",
"context": [
{
"collectionId": "[COLLECTION ID]",
"documentId": "b0dd63b1-f78c-4989-94f9-f73768bb12dd",
"chunkId": "348a8973-4fc6-42ba-a557-0d89ea8aef54",
"content": "Rest of Asia Pacific 5,630 6,150 23,284 23,002 \nTotal net sales ! 81,797 ! 82,959 ! 293,787 ! 304,182 \n \n(1) Net sales by category: \niPhone ! 39,669 ! 40,665 ! 156,778 ! 162,863 \nMac 6,840 7,382 21,743 28,669 \niPad 5,791 7,224 21,857 22,118 \nWearables, Home and Accessories 8,284 8,084 30,523 31,591 \nServices 21,213 19,604 62,886 58,941 \nTotal net sales ! 81,797 ! 82,959 ! 293,787 ! 304,182 \n \n \n \n \nApple Inc. \nCONDENSED CONSOLIDATED BALANCE SHEETS (Unaudited) \n(In millions, except number of shares which are reflected in thousands and par value) \n \n July 1, \n2023 September 24, \n2022 \nASSETS: \nCurrent assets: \nCash and cash equivalents ! 28,408 ! 23,646 \nMarketable securities 34,074 24,658 \nAccounts receivable, net 19,549 28,184 \nInventories 7,351 4,946 \nVendor non-trade receivables 19,637 32,748 \nOther current assets 13,640 21,223 \nTotal current assets 122,659 135,405",
"metadata": {}
},
{
"collectionId": "[COLLECTION ID]",
"documentId": "b0dd63b1-f78c-4989-94f9-f73768bb12dd",
"chunkId": "bf3d056f-073b-46e2-b08b-7914fc511ba8",
"content": "Apple Inc. \nCONDENSED CONSOLIDATED STATEMENTS OF OPERATIONS (Unaudited) \n(In millions, except number of shares which are reflected in thousands and per share amounts) \n Three Months Ended Nine Months Ended \n July 1, \n2023 June 25, \n2022 July 1, \n2023 June 25, \n2022 \nNet sales: \n Products ! 60,584 ! 63,355 ! 230,901 ! 245,241 \n Services 21,213 19,604 62,886 58,941 \nTotal net sales (1) 81,797 82,959 293,787 304,182 \nCost of sales: \n Products 39,136 41,485 146,696 155,084 \n Services 6,248 5,589 18,370 16,411 \nTotal cost of sales 45,384 47,074 165,066 171,495 \nGross margin 36,413 35,885 128,721 132,687 \n \nOperating expenses: \nResearch and development 7,442 6,797 22,608 19,490 \nSelling, general and administrative 5,973 6,012 18,781 18,654 \nTotal operating expenses 13,415 12,809 41,389 38,144",
"metadata": {}
},
{
"collectionId": "[COLLECTION ID]",
"documentId": "b0dd63b1-f78c-4989-94f9-f73768bb12dd",
"chunkId": "858ec3d6-770f-40ac-a203-5dd219c1482d",
"content": "Common stock and additional paid-in capital, !0.00001 par value: 50,400,000 shares \nauthorized; 15,647,868 and 15,943,425 shares issued and outstanding, respectively 70,667 64,849 \nRetained earnings/(Accumulated deficit) 1,408 (3,068) \nAccumulated other comprehensive income/(loss) (11,801) (11,109) \nTotal shareholders’ equity 60,274 50,672 \nTotal liabilities and shareholders’ equity ! 335,038 ! 352,755 \n \n \n \n \nApple Inc. \nCONDENSED CONSOLIDATED STATEMENTS OF CASH FLOWS (Unaudited) \n(In millions) \n \n Nine Months Ended \n July 1, \n2023 June 25, \n2022 \nCash, cash equivalents and restricted cash, beginning balances ! 24,977 ! 35,929 \n \nOperating activities: \nNet income 74,039 79,082 \nAdjustments to reconcile net income to cash generated by operating activities: \nDepreciation and amortization 8,866 8,239 \nShare-based compensation expense 8,208 6,760 \nOther (1,651) 2,695 \nChanges in operating assets and liabilities:",
"metadata": {}
}
]
}
}You can also retrive the whole conversation using get chat API.
curl -X GET https://api.documentai.dev/v1/collections/[COLLECTION ID]/chat/[CHAT ID] \
-H 'X-API-KEY: [YOUR API KEY}'Where:
- COLLECTION ID - ID of the collection, reuse existing or specify new.
- API KEY - the API key from the console.
- CHAT ID - the id of the chat. You decide the id of the chat and do not need to explicitly create it. All previous messages in a chat will be evaluated by the LLM.
The response will include both your messages and assistant's messages with the relevant context so you can link back to the source.
{
"messages": [
{
"sender": "USER",
"id": "d595de0e-4664-43a6-b03b-396bb05933a4",
"date": "2023-09-03T11:24:09.026440Z",
"content": "How many iPhones did they sell?",
"context": []
},
{
"sender": "ASSISTANT",
"message": {
"id": "553beb15-9f2b-4c97-857f-94963bbce84f",
"date": "2023-09-03T12:24:09.026440Z",
"content": "According to the document, Apple Inc. sold 39,669 iPhones in the most recent reporting period.",
"context": [
{
"collectionId": "[COLLECTION ID]",
"documentId": "b0dd63b1-f78c-4989-94f9-f73768bb12dd",
"chunkId": "348a8973-4fc6-42ba-a557-0d89ea8aef54",
"content": "Rest of Asia Pacific 5,630 6,150 23,284 23,002 \nTotal net sales ! 81,797 ! 82,959 ! 293,787 ! 304,182 \n \n(1) Net sales by category: \niPhone ! 39,669 ! 40,665 ! 156,778 ! 162,863 \nMac 6,840 7,382 21,743 28,669 \niPad 5,791 7,224 21,857 22,118 \nWearables, Home and Accessories 8,284 8,084 30,523 31,591 \nServices 21,213 19,604 62,886 58,941 \nTotal net sales ! 81,797 ! 82,959 ! 293,787 ! 304,182 \n \n \n \n \nApple Inc. \nCONDENSED CONSOLIDATED BALANCE SHEETS (Unaudited) \n(In millions, except number of shares which are reflected in thousands and par value) \n \n July 1, \n2023 September 24, \n2022 \nASSETS: \nCurrent assets: \nCash and cash equivalents ! 28,408 ! 23,646 \nMarketable securities 34,074 24,658 \nAccounts receivable, net 19,549 28,184 \nInventories 7,351 4,946 \nVendor non-trade receivables 19,637 32,748 \nOther current assets 13,640 21,223 \nTotal current assets 122,659 135,405",
"metadata": {}
},
{
"collectionId": "[COLLECTION ID]",
"documentId": "b0dd63b1-f78c-4989-94f9-f73768bb12dd",
"chunkId": "bf3d056f-073b-46e2-b08b-7914fc511ba8",
"content": "Apple Inc. \nCONDENSED CONSOLIDATED STATEMENTS OF OPERATIONS (Unaudited) \n(In millions, except number of shares which are reflected in thousands and per share amounts) \n Three Months Ended Nine Months Ended \n July 1, \n2023 June 25, \n2022 July 1, \n2023 June 25, \n2022 \nNet sales: \n Products ! 60,584 ! 63,355 ! 230,901 ! 245,241 \n Services 21,213 19,604 62,886 58,941 \nTotal net sales (1) 81,797 82,959 293,787 304,182 \nCost of sales: \n Products 39,136 41,485 146,696 155,084 \n Services 6,248 5,589 18,370 16,411 \nTotal cost of sales 45,384 47,074 165,066 171,495 \nGross margin 36,413 35,885 128,721 132,687 \n \nOperating expenses: \nResearch and development 7,442 6,797 22,608 19,490 \nSelling, general and administrative 5,973 6,012 18,781 18,654 \nTotal operating expenses 13,415 12,809 41,389 38,144",
"metadata": {}
},
{
"collectionId": "[COLLECTION ID]",
"documentId": "b0dd63b1-f78c-4989-94f9-f73768bb12dd",
"chunkId": "858ec3d6-770f-40ac-a203-5dd219c1482d",
"content": "Common stock and additional paid-in capital, !0.00001 par value: 50,400,000 shares \nauthorized; 15,647,868 and 15,943,425 shares issued and outstanding, respectively 70,667 64,849 \nRetained earnings/(Accumulated deficit) 1,408 (3,068) \nAccumulated other comprehensive income/(loss) (11,801) (11,109) \nTotal shareholders’ equity 60,274 50,672 \nTotal liabilities and shareholders’ equity ! 335,038 ! 352,755 \n \n \n \n \nApple Inc. \nCONDENSED CONSOLIDATED STATEMENTS OF CASH FLOWS (Unaudited) \n(In millions) \n \n Nine Months Ended \n July 1, \n2023 June 25, \n2022 \nCash, cash equivalents and restricted cash, beginning balances ! 24,977 ! 35,929 \n \nOperating activities: \nNet income 74,039 79,082 \nAdjustments to reconcile net income to cash generated by operating activities: \nDepreciation and amortization 8,866 8,239 \nShare-based compensation expense 8,208 6,760 \nOther (1,651) 2,695 \nChanges in operating assets and liabilities:",
"metadata": {}
}
]
}
}
]
}Summary
Hopefully this post has demonstrated the benefits of using AI techniques like document understanding and conversational interfaces to create a knowledgeable assistant for financial analysis. By integrating these capabilities with your collection of financial documents, you can have an assistant ready to enhance your productivity and uncover valuable insights through natural conversational queries. With just a few API calls to documentAI, you can upload reports, statements and more to start conversing with your assistant. It can help you rapidly discover connections, derive strategic insights, and enhance your understanding of the financial landscape relevant to your specific context. Rather than hunting through spreadsheets, you can simply ask questions conversationally and accelerate your analysis.