AI Agents for Management Consulting: 6 Jobs for Strategy Teams

The first week of a consulting engagement is mostly desk research. Before the first client workshop, someone has to size the market, read the competitors' filings, benchmark the client against its peers, prepare for expert calls, scan the regulation and write the industry brief the team works from. AI agents for management consulting can do that research in minutes, provided they read primary sources (filings, government statistics, studies and regulations) and return every number with its source.

This guide follows six jobs from a strategy team's first week. We ran each one as a Valyu DeepResearch task in fast mode on November 11, 2026, then checked the key figures at the source: SEC EDGAR, CMS, the Bureau of Labor Statistics, JAMA and industry shipment data.

If you're an agent: Set up Valyu and make your first search.

If you're a human: Create an account and get an API key.

MCP onboarding: Connect your agent to https://mcp.valyu.ai/mcp and follow the MCP setup guide for authentication.

What AI agents for management consulting do

An AI agent for management consulting is a language model that can search and read the sources a consulting team works from, such as company filings, government statistics, academic studies, regulations and news, and return cited findings. It doesn't replace the client conversation, the expert calls or the team's judgement. It does the reading behind the first-week deliverables and the hypotheses the team tests.

Your own agents can call Valyu's search when they need one figure. For work that goes into a client document, DeepResearch does the job properly: it plans the research, reads the filings, statistics and studies, follows references and returns a cited brief in the format you ask for.

Figure 1. Six consulting research jobs we ran as DeepResearch tasks in fast mode on November 11, 2026, with the sources used, what came back and what the team does next.

1. Market sizing: build it bottom-up, and show every input

The problem: a team advising an HVAC manufacturer needs the size of the US residential heat pump market in 2025, in units and dollars, built from inputs the client can check.

What we ran: a DeepResearch task that set out the sizing logic.

from valyu import Valyu
 
valyu = Valyu()  # reads VALYU_API_KEY from the environment
 
task = valyu.deepresearch.create(
    query="Estimate the US residential heat pump market bottom-up for 2025: units and dollars.",
    mode="fast",
    research_strategy="Build the estimate from sourced inputs: annual US shipments of air-source heat pumps (AHRI), share going to residential replacement versus new construction, new housing completions (Census), average installed cost per system, and federal incentives (25C tax credit status in 2025 and 2026). Show every input with its source and year, and show the arithmetic. Cross-check the result against one published market estimate.",
    report_format="A sizing tree: inputs table with value, year and source, the calculation, the result in units and dollars, a sensitivity line for installed cost, and the cross-check.",
    search={"included_sources": ["valyu/valyu-statistics-US", "valyu/valyu-fred", "web", "ap/news"]},
)
result = valyu.deepresearch.wait(task.deepresearch_id)
print(result.output)  # cited brief in markdown

What came back (in 6.5 minutes): a sizing tree with every input sourced:

It added a sensitivity line (each $1,000 of average installed cost moves the total by about $3.5 billion) and noted that the 25C federal tax credit covered 2025 installations but ended for equipment placed in service after December 31, 2025.

What we checked at the source: AHRI's 2025 shipment total of 3,644,596. It matches.

The catch: the cross-check looks like a contradiction. Published market reports put the same market at $5.8 billion to $10.5 billion. DeepResearch explained the gap: those reports measure equipment revenue at around $1,650 to $2,980 per unit, while the bottom-up estimate measures installed cost including labour. Neither is wrong, but a deck that puts them side by side without saying so will confuse the client.

The decision: state which basis the client needs (equipment or installed) before quoting a market size, and model 2026 without the 25C credit.

2. Competitor landscape: read the definitions, not just the numbers

The problem: a retailer considering an off-price format needs a side-by-side view of TJX, Ross Stores and Burlington: sales, comparable-store sales, stores, margins and growth plans.

What we ran: a DeepResearch task on each company's latest 10-K and 10-Q.

task = valyu.deepresearch.create(
    query="Competitive landscape of US off-price apparel and home retail: TJX, Ross Stores and Burlington.",
    mode="fast",
    research_strategy="Use each company's latest 10-K and 10-Q for revenue, comparable store sales growth, store count, new store plans, gross margin or merchandise margin, operating margin and inventory per store. State fiscal year ends, since they differ. Add each company's stated strategy from the 10-K business section.",
    report_format="A side-by-side table with every metric and period, a paragraph per company on strategy, then three implications for a client entering off-price.",
    search={"included_sources": ["valyu/valyu-sec-filings", "valyu/valyu-income-statement-US"]},
)
result = valyu.deepresearch.wait(task.deepresearch_id)

What came back (in 5.3 minutes): a side-by-side table from the filings, with each company's fiscal year labelled:

It also summarised each company's strategy from its 10-K: TJX's buying from more than 100 countries, Ross's packaway warehousing, and Burlington's plan to open about 100 stores a year.

What we checked at the source: Ross's and Burlington's 10-Qs for the quarter ended August 1, 2026. Ross's operating margin of 17.6% includes a benefit of about $253 million from refunds of IEEPA tariffs, and Burlington's gross margin of 46.2% includes a $55.5 million tariff refund benefit. Both match the filings.

The catch: the latest quarter flatters all three. Ross's cost of goods fell about 625 basis points as a share of sales, and 405 of those points came from the one-time tariff refund. And the three companies define comparable-store sales differently, from 14 months open at Ross to the third fiscal year at TJX, so the comps can't be compared directly. A table built from press releases would miss both.

The decision: strip the tariff refunds out before modelling off-price margins, and build the client's own comp definition rather than borrowing one.

3. Peer benchmarking: what explains the margin gap

The problem: a team working with an airline needs a 2025 benchmark against its peers: margins, unit revenue, unit cost and load factor, on each company's own definitions.

What we ran: a DeepResearch task on five airlines' 10-K filings.

task = valyu.deepresearch.create(
    query="Benchmark Delta, United, American, Southwest and Alaska Air Group on 2025 operating performance.",
    mode="fast",
    research_strategy="Use each airline's 10-K for fiscal 2025: operating revenue, operating margin, available seat miles, passenger revenue per ASM (PRASM), total revenue per ASM (TRASM), cost per ASM (CASM) and CASM excluding fuel, fuel cost per gallon and load factor. Use each company's own definitions and note where non-GAAP adjustments differ.",
    report_format="A benchmark table with each metric by airline, the peer median, then a paragraph on what explains the margin gap between the best and worst performer.",
    search={"included_sources": ["valyu/valyu-sec-filings", "valyu/valyu-income-statement-US"]},
)
result = valyu.deepresearch.wait(task.deepresearch_id)

What came back (in 2.7 minutes): a benchmark of all five carriers from their 10-Ks, with operating margins worked out from the reported figures:

Its explanation of the 7.7-point gap between Delta and Southwest was on the revenue side: Delta's unit revenue is about 25% higher even after excluding refinery sales, and its load factor is 6.6 points higher, while Southwest's total cost per seat mile is actually lower.

What we checked at the source: Delta's 10-K (operating revenue of $63,364 million and operating income of $5,822 million) and Southwest's 77.4% load factor. Both match.

The catch: the cost comparison doesn't hold together. United's 10-K reports no cost per seat mile excluding fuel, Alaska reports pro forma figures including Hawaiian, and each carrier excludes different items from its own measure. DeepResearch said the filings don't support a clean cost ranking, rather than building one from mismatched figures.

The decision: lead the diagnostic with revenue quality and load factor, and ask the client for cost data on a common basis before benchmarking cost.

4. Expert-call preparation: know the evidence before the call

The problem: a team assessing AI ambient scribes, the tools that draft clinical notes from doctor-patient conversations, has expert calls booked with a hospital CMIO, a practising physician and a vendor. It needs to know what's already public, so the calls test what isn't.

What we ran: a DeepResearch task across PubMed studies, news and company announcements.

task = valyu.deepresearch.create(
    query="Prepare for expert calls on hospital adoption of AI ambient clinical documentation (ambient scribes) in the US.",
    mode="fast",
    research_strategy="Summarise what is public: main vendors and their health system customers, published studies on time saved and clinician burnout (PubMed), reported pricing per clinician, and EHR integration. Turn the gaps into hypotheses and questions. Cite every fact.",
    report_format="An expert-call prep pack: what we know (cited), five hypotheses, 15 questions grouped by expert type (CMIO, practising physician, vendor executive), and the facts each call should confirm.",
    search={"included_sources": ["valyu/valyu-pubmed", "ap/news", "web"]},
)
result = valyu.deepresearch.wait(task.deepresearch_id)

What came back (in 5.1 minutes): a call pack with these parts:

• Studies. A table of eight time-saving studies, led by a 2026 JAMA study of 8,581 clinicians at five academic centres. That study found 13.4 fewer minutes of EHR time and 16.0 fewer minutes of documentation per 8 scheduled patient hours.
• Burnout evidence. A 2026 meta-analysis graded the burnout evidence as low certainty, with publication bias strongly suspected.
• Pricing. Most health systems pay $200 to $600 per clinician per month, according to a JAMA Network Open editorial. Most vendor-specific prices are third-party estimates.
• Vendors. A landscape including Epic's native AI Charting, released in February 2026.
• Questions. Five hypotheses, each with a test that would disprove it, and 15 questions grouped by expert type.

What we checked at the source: the JAMA study's clinician count and its 13.4-minute EHR time result. Both match.

The catch: the best-designed study shows modest savings, and after-hours EHR time did not improve significantly. That isn't the picture in vendor material. The prep pack also flagged that the largest burnout study comes from a health system with a disclosed investment in the vendor it tested.

The decision: use the calls to test three numbers: the real per-seat price since Epic's launch, sustained utilisation rather than licences sold, and whether coding-driven revenue survives payer scrutiny.

5. Regulatory scan: check the date before you build the plan

The problem: a client that uses AI to screen job applicants in the EU needs to know which EU AI Act obligations apply to it, and from when.

What we ran: a DeepResearch task that asked for adopted law to be separated from proposals.

task = valyu.deepresearch.create(
    query="Which EU AI Act obligations apply to a company that uses AI to screen job applicants in the EU, and when do they apply?",
    mode="fast",
    research_strategy="Identify the relevant classification (Annex III employment use cases), the obligations for deployers versus providers, the application dates, and any changes proposed or adopted in 2025 and 2026 that move those dates, including the Digital Omnibus. Cite the regulation's articles and the official EU sources. Separate adopted law from proposals.",
    report_format="A regulatory scan: obligations table with obligation, applies to, article and date, a timeline, a section on adopted versus proposed changes, then five actions for the client.",
    search={"included_sources": ["web", "ap/news"]},
)
result = valyu.deepresearch.wait(task.deepresearch_id)

What came back (in 9.2 minutes): an obligations table and timeline. Recruitment screening is a high-risk use under Annex III, point 4(a). The scan sorted the obligations by when they apply:

• Already in force. The ban on inferring emotions in the workplace, which covers candidates, and the AI literacy duty have applied since February 2, 2025. Transparency duties under Article 50 have applied since August 2, 2026.
• From December 2, 2027. The deployer duties under Article 26: human oversight, keeping logs for at least six months, informing workers' representatives and telling candidates when the system is used.
• Not required. A fundamental rights impact assessment under Article 27 doesn't apply to an ordinary private employer using recruitment AI.

What we checked at the source: the December 2, 2027 date for Annex III systems in published analyses of Regulation (EU) 2026/1744, the Digital Omnibus on AI. It matches.

The catch: many pages still give August 2, 2026 as the high-risk deadline. The Digital Omnibus, adopted in July 2026, moved it to December 2, 2027 for Annex III systems. A scan built from older pages would put the client on the wrong timeline. DeepResearch identified the omnibus as adopted law, not a proposal, and noted that the obligations that weren't deferred still apply.

The decision: switch off any emotion-inference feature now, document the AI literacy measures, and plan the Article 26 package for December 2027.

6. Industry brief: one document for day one

The problem: a team starting an engagement with a home health operator needs a first-day brief: market, economics, reimbursement, workforce, players and recent deals.

What we ran: a DeepResearch task across the CMS rule, BLS data and company filings.

task = valyu.deepresearch.create(
    query="First-day industry brief on US home health care for a consulting team starting a new engagement.",
    mode="fast",
    research_strategy="Cover market size and growth, Medicare share of revenue, the CMS Home Health Prospective Payment System final rule for calendar 2026 and its payment change, employment and wages for home health aides (BLS), the largest public operators with revenue from their 10-Ks, recent M&A, and the main issues (labour, reimbursement, Medicare Advantage). Cite every figure with its year.",
    report_format="A two-page brief: key numbers table, how the industry makes money, regulation and reimbursement, key players table, recent deals, five issues for the engagement and a glossary.",
    search={"included_sources": ["valyu/valyu-bls", "valyu/valyu-sec-filings", "valyu/valyu-fred", "web", "ap/news"]},
)
result = valyu.deepresearch.wait(task.deepresearch_id)

What came back (in 3.9 minutes): a brief with a key numbers table, an explanation of Medicare's 30-day payment model and a glossary. The central figures:

What we checked at the source: the CMS fact sheet for the calendar 2026 rule (a 1.3%, or $220 million, decrease) and the BLS pay and openings figures. Both match.

The catch: there's no single market size. Published estimates for 2025 range from about $107 billion to $142 billion depending on what each counts as home health. DeepResearch listed them with their sources and said so, rather than picking one.

The decision: agree the market definition with the client in week one, and model reimbursement scenarios around CMS's decision on the temporary adjustment.

What the six jobs cost

A fast task costs a fixed 25 credits, about $0.16 at pay-as-you-go rates of 1,600 credits for $10, whatever it searches and reads. The six jobs cost about $0.94 together. For deeper work on one question, standard mode uses 100 credits and heavy 500.

Go further: a client-ready document in one task

The six jobs each answer one question. A consulting team's actual output is a document: the industry primer, the market model, the steering committee readout. A single DeepResearch task in standard or heavy mode writes these, with files such as a docx brief, an xlsx model and a pptx readout.

task = valyu.deepresearch.create(
    query="Prepare a market entry assessment for a mid-size US retailer considering an off-price format.",
    mode="standard",
    research_strategy="Use TJX, Ross Stores and Burlington 10-K and 10-Q filings for sales, comparable sales, stores, margins and store targets. Strip one-time items such as tariff refunds out of margins, and note where comparable-sales definitions differ. Use Census retail data for category size. Cite every figure.",
    report_format="A market entry brief: Market, Competitors, Economics of an off-price store, Entry options, Risks, then 10 questions for the client. Every figure cited with its period and definition.",
    search={"included_sources": ["valyu/valyu-sec-filings", "valyu/valyu-income-statement-US", "valyu/valyu-statistics-US", "web"]},
    deliverables=["An xlsx competitor table with metric, company, period, definition and source", "A pptx readout of 8 slides"],
)
 
result = valyu.deepresearch.wait(task.deepresearch_id)
print(result.output)  # cited report in markdown
for file in result.deliverables:
    print(file.type, file.url)  # signed download link

The research strategy carries the lessons from the six jobs: strip out one-time items, state each definition and period, and separate adopted rules from proposals. To run the same brief for several clients or markets, the DeepResearch Batch API runs the tasks in parallel and calls one webhook when they finish.

Start from a ready-made workflow

Valyu's curated DeepResearch workflows include several that consulting teams use, each with set inputs, a recommended mode and file deliverables:

To run one from code, pass its slug as workflow_id instead of a query:

task = valyu.deepresearch.create(
    workflow_id="con-market-sizing",
    workflow_params={"market": "US residential heat pumps"},
)
result = valyu.deepresearch.wait(task.deepresearch_id)

To browse the full library, open Workflows in the Valyu dashboard, or see running a workflow in the docs.

Where AI agents stop in management consulting

The jobs above show what DeepResearch can do from public sources. Four limits shape how to use it:

• The client's data is the client's. Cost bases, customer data and internal plans come from the client. The agent prepares the outside-in view and the questions.
• Experts know what isn't written down. The call pack in job 4 is preparation for the expert calls, not a substitute. The gaps it found are the reason for the calls.
• Public sources can disagree. Market sizes, prices and definitions vary. The agent can list them with their sources, but choosing the right basis for the client is the team's call.
• Judgement stays with people. Recommendations depend on the client's position, appetite for risk and organisation, which no filing describes.

How to set up a consulting research agent

If you don't want to write code as shown above, there is another option to connect Valyu to Claude or ChatGPT as a custom connector at https://mcp.valyu.ai/mcp and ask for the same research in plain English. To build your own agent, follow how to build a financial research agent with Valyu. For related work, see our guides to comparing 10-K risk factors year over year, AI agents for economic research and AI for due diligence.

If you already run your own agents, they can call Valyu's search for single lookups, such as one company's store count. For anything that goes in front of a client, use DeepResearch.

Three habits made the jobs above reliable:

• Ask for the definition and period behind every number. Comparable-store sales, cost per seat mile and market size each meant different things in different sources.
• Ask for one-time items to be called out. The tariff refunds changed the margin picture for all three retailers.
• Ask for adopted law to be separated from proposals. In regulation, the date is often the whole answer.

FAQ

What are AI agents for management consulting?

AI agents for management consulting are language models that search and read the sources consultants work from, such as company filings, government statistics, academic studies, regulations and news, and return cited findings. Research agents such as Valyu DeepResearch plan the work, follow references and write a brief: a market sizing, a competitor landscape, an expert-call pack or an industry primer.

How do consulting firms use AI?

Consulting firms use AI to speed up desk research: sizing markets, reading competitors' filings, benchmarking clients against peers, preparing for expert calls, scanning regulation and writing industry briefs. In a November 2026 test, six Valyu DeepResearch tasks in fast mode did one of each for 150 credits, about $0.94, and all six finished within 10 minutes.

Can AI do market sizing?

Yes, if it shows its inputs. A DeepResearch task can build a bottom-up estimate from sourced figures, such as shipments, housing data and installed costs, show the arithmetic and cross-check it against published estimates. In our test, it sized the 2025 US residential heat pump market at about 3.5 million units and $21 billion to $29 billion installed, and explained why published reports give lower figures.

Can AI replace expert calls?

No. AI can prepare for them: summarise the published evidence, turn the gaps into hypotheses and write the questions for each type of expert. In our test on AI ambient scribes, the call pack included eight studies, five hypotheses with tests that would disprove them, and 15 questions. The calls themselves still produce the information that isn't public.

Is AI accurate enough for client work?

It depends on the sources and on checking. In our test, every key figure we checked matched the primary source: the AHRI shipment total, the airline and retailer filings, the JAMA study, the CMS rule and the BLS data. The care points were definitions and one-time items, which DeepResearch flagged. A consultant should still check every figure that goes into a client document.

What is the best AI tool for consulting research?

Consulting research tools fall into three groups. Firms' internal knowledge platforms search past engagement work. Research platforms, such as AlphaSense and Hebbia, search licensed documents inside their own products. Valyu DeepResearch researches primary sources, such as SEC filings, government statistics, studies and regulations, and returns cited briefs with xlsx, docx and pptx files through an API or inside Claude and ChatGPT.

How much does AI research for consulting cost?

With Valyu DeepResearch, a fast-mode task costs a fixed 25 credits, about $0.16, and a standard report 100 credits, about $0.63. In a November 2026 test, six fast-mode consulting research jobs cost 150 credits, about $0.94, in total. A heavy-mode report uses 500 credits, about $3.13. Free credits are included every month.

Start with one job

Pick the job that takes your team the most time at the start of every engagement, such as the competitor landscape or the industry brief, and run it as a DeepResearch task on a sector you already know. Compare the brief with your own work.

Create a Valyu account for free credits every month, up to 2,000 with a work email. Then connect it to Claude or ChatGPT at https://mcp.valyu.ai/mcp, or browse the data sources. For the investor side of the same research, see AI Agents for Financial Research.