How to Use Claude to Analyze HubSpot Campaign and Attribution Reports
Marketing ops can use Claude to analyze HubSpot campaign and attribution reports in two ways: use the HubSpot connector for Claude, which needs a paid Claude plan, or upload exported CSV reports. Then run four prompts each month: sanity check, performance summary, attribution model comparison and anomaly list. Claude drafts the analysis; a person checks the numbers in HubSpot before any budget moves.
Key takeaways
- The HubSpot connector for Claude requires a paid Claude plan (Pro, Max, Team or Enterprise), while uploaded CSV exports can be analyzed with file features available on every Claude plan.
- Turn on the interaction types for your assets in HubSpot's revenue attribution settings first, or the attribution reports will not populate.
- In the worked example, the same $90,000 of hypothetical revenue gives the webinar campaign $54,000 under first touch, $16,000 under last touch and $26,000 under linear.
- Claude may draft summaries, model comparisons and anomaly lists, but a person decides budget moves, the attribution model and data fixes.
- Reconcile every total Claude reports against the HubSpot report it came from, because HubSpot's own documentation says Claude can make mistakes.
- Fix tracking URLs, UTM conventions and campaign tagging before any analysis, because Claude can only read what HubSpot recorded.
How does Claude get access to HubSpot campaign and attribution data?
Claude reaches HubSpot data in two ways: the HubSpot connector for Claude, which answers questions within each user's HubSpot permissions, or a report you export from HubSpot and upload as a file. You can use both: the connector for questions, the export for arithmetic you need to check.
Digitalegy is a HubSpot partner.
The HubSpot knowledge base article on the connector for Claude describes the first route in one line: "Use the HubSpot connector for Claude to bring HubSpot context into Claude." It lets you ask plain-language questions about your HubSpot records. It needs a paid plan; the article states: "You must have a paid Anthropic subscription (Pro, Max, Team, or Enterprise)." It also limits what each person sees: "Users can access only the data allowed by their existing HubSpot user permissions." When a Super Admin changes the connector's data permissions, each user has to reconnect before the new data appears.
The second route is an export. HubSpot lets you export a selected campaign's performance details as a csv, xls, xlsx or image PDF file. The Claude Help Center confirms that you can upload CSV, TSV and other files for data analysis and visualization, with a maximum of 30MB per file, and that code execution and file creation are available to all Claude users, Free included. Choose CSV over image PDF: a CSV keeps every number as data that can be calculated.
| Route | What it allows, per public documentation | Claude plan | Best use in a monthly review |
|---|---|---|---|
| HubSpot connector | Questions about HubSpot data within each user's HubSpot permissions; page and post performance needs the campaign name and date range in the prompt | Pro, Max, Team or Enterprise | Exploration and follow-up questions |
| Report export and upload | CSV, TSV and other files up to 30MB each, for data analysis and visualization | Code execution and file creation on all plans | Model comparisons and arithmetic you must reproduce |
For a monthly review, a practical setup uses both routes: the connector for quick questions, and exported CSV files as the fixed record you check every number against.
Which HubSpot reports should you give Claude for a campaign review?
Give Claude three inputs for the same month: campaign performance with the attributed revenue vs spend report, one attribution report per conversion type you track, and new contact and closed won deal data. Prepare them with one date range, identical campaign names and a short note on what each column means.
Start with these inputs:
- Campaign performance. HubSpot's guide to analyzing performance across campaigns in Marketing Studio says this view helps you "measure the overall effectiveness of your campaigns." Include its attributed revenue vs spend report, which is grouped by campaign.
- Attribution reports. HubSpot's attribution reporting overview explains that you can create three types of attribution reports, each measuring a different type of conversion. Export one per conversion type you report on and put the model in the file name. Deal Revenue Attribution, which measures which sources, assets and interactions had the greatest impact on deal revenue, is for Marketing Hub Enterprise accounts only.
- Contact and deal data. HubSpot's campaigns FAQ defines first-touch contacts as "the number of newly created contacts, attributed to the campaign that directed them to your website for the first time." It also says a deal needs to have been associated with the influenced contact before it was set to closed won.
Then prepare the files so Claude reads them the way HubSpot means them:
- Use one date range in every file and repeat it in the prompt.
- Keep campaign names identical across files; a renamed campaign looks like two campaigns.
- Add a short data dictionary. For example, sessions are not visitors: HubSpot notes that "A single visitor can have multiple sessions and page views to your site."
- Record report settings: model, conversion type and sampling. HubSpot says attribution reports include up to 20 million interactions per report after sampling, and represent at least one year of interactions when available.
- Remove columns you do not need, such as personal contact details, before uploading.
What four prompts turn HubSpot data into a monthly campaign review?
Run four prompts in order: data pull and sanity check, performance summary by campaign and channel, attribution model comparison, and an anomaly and follow-up list. Each prompt builds on the last and asks Claude to show its sources and arithmetic.
HubSpot's connector article suggests a starting prompt: "Ask which campaign generated the most revenue in the last month." That is a useful question, but it is one question. The sequence below turns it into a review you can repeat. Two practical limits: the connector requires the campaign name and date range in the prompt to pull performance data for individual pages or posts, and uploads are capped at 30MB per file. Replace the bracketed text before you run each prompt.
Prompt 1: Data pull and sanity check
You are helping me run a monthly HubSpot campaign review for [month, year]. Using the HubSpot connector or the attached CSV files, list every campaign with activity between [start date] and [end date]. For each file or data pull, report the date range, number of rows, attribution model, conversion type, and totals for sessions, new contacts, deals and attributed revenue. Flag missing values, duplicate or near-duplicate campaign names, campaigns with revenue but no spend, and any total that does not match the sum of its rows. Do not analyze performance yet. Cite the report or record behind each total.
Prompt 2: Performance summary by campaign and channel
Using only the data you verified in the previous step, summarize performance by campaign and by channel for [month]. Build a table with campaign, channel, spend, sessions, first-touch new contacts, deals created, attributed revenue and revenue per dollar of spend. Show the formula for every calculated column. Then write five bullet points on what changed compared with [previous month], each with the numbers behind it. Treat sessions as sessions, not unique visitors. Label any explanation of why something happened as a hypothesis.
Prompt 3: Attribution model comparison
Compare attribution models for the same revenue and date range: first interaction, last interaction and linear [add any other model in the files]. Build a table with each campaign's credited revenue under each model and the gap between its highest and lowest credit. Show the arithmetic for the three campaigns with the largest gap. List which conclusions hold under every model and which depend on the model. Do not recommend a model; list the questions a person should answer to choose one.
Prompt 4: Anomaly and follow-up list
List anomalies for [month]: campaigns whose spend, sessions, contacts or revenue moved more than [X] percent from [previous month]; deals with revenue but no campaign interactions; campaigns or tracking URLs with no visits; anything that looks like a tracking or tagging problem rather than a performance change. For each item, give the evidence, the HubSpot report where I can confirm it, a possible data cause, a possible performance cause and a suggested owner. Sort by revenue at stake.
Run all four in one conversation so the later prompts can refer to the totals checked in prompt 1. If sales wants the same rhythm for pipeline, read How to Use Claude to Review HubSpot Pipeline Data and Flag Forecast Risk Every Week.
How do first-touch, last-touch and linear attribution change Claude's conclusions?
The model you choose decides the story: in the hypothetical example below, the webinar campaign earns 60 percent of revenue under first touch and 18 percent under last touch. Ask Claude to compare models side by side and to separate conclusions that hold under every model from those that depend on one.
An attribution model is the rule that splits credit for a conversion across the interactions that led to it. HubSpot's attribution reporting overview says the first interaction model "attributes 100% of the contact credits to the contact's first interaction in the conversion path," the last interaction model gives 100% to the last interaction, and the linear model "attributes the contact credits equally to each interaction in the conversion path." Linear is a multi-touch model: it credits more than one interaction.
The hypothetical dataset
These numbers are illustrative, not client data. Three campaigns (Webinar, Paid Search, Email Nurture) touched four closed won deals worth $90,000. Monthly spend: Webinar $12,000, Paid Search $15,000, Email Nurture $3,000.
| Deal | Revenue | Interaction path, first to last | Linear credit per interaction |
|---|---|---|---|
| D1 | $30,000 | Webinar, Email Nurture, Paid Search | $30,000 / 3 = $10,000 |
| D2 | $20,000 | Paid Search, Email Nurture | $20,000 / 2 = $10,000 |
| D3 | $24,000 | Webinar, Paid Search, Email Nurture | $24,000 / 3 = $8,000 |
| D4 | $16,000 | Email Nurture, Webinar | $16,000 / 2 = $8,000 |
Credit by model, with the arithmetic
| Campaign | First touch | Last touch | Linear | Revenue per $1 of spend (first / last / linear) |
|---|---|---|---|---|
| Webinar | D1 + D3 = $54,000 | D4 = $16,000 | $10,000 + $8,000 + $8,000 = $26,000 | 4.50 / 1.33 / 2.17 |
| Paid Search | D2 = $20,000 | D1 = $30,000 | $10,000 + $10,000 + $8,000 = $28,000 | 1.33 / 2.00 / 1.87 |
| Email Nurture | D4 = $16,000 | D2 + D3 = $44,000 | $10,000 + $10,000 + $8,000 + $8,000 = $36,000 | 5.33 / 14.67 / 12.00 |
| Total | $90,000 | $90,000 | $90,000 |
How the story changes
- First touch makes the webinar the hero: $54,000 of $90,000, or 60 percent. Email nurture gets the least credit ($16,000), and paid search returns the least per dollar of spend (1.33).
- Last touch flips it: email nurture earns $44,000 (49 percent) and the webinar $16,000 (18 percent). A reader might cut the webinar.
- Linear narrows the spread to $26,000, $28,000 and $36,000.
One conclusion holds under all three models: email nurture returns the most revenue per dollar of spend. The webinar's rank moves from first to last depending on the model. That is why prompt 3 asks Claude to separate model-proof conclusions from model-dependent ones. Keep HubSpot's caution about linear in mind:
"This model is not ideal at identifying which campaigns or assets are more effective, because all interactions get credit equally." (HubSpot, Understand attribution reporting)
Recalculate one campaign yourself from Claude's table before you share it.
Where must human judgment stay when Claude analyzes campaigns?
People keep four decisions: budget moves, the choice of attribution model, data quality sign-off, and anything HubSpot never recorded, such as offline conversations or dark-funnel influence. Claude can draft the analysis behind each, but a person checks it before money moves.
Claude reads only what HubSpot recorded and what you upload. It does not see a conference conversation, a peer recommendation or dark-funnel influence, meaning buying research in private channels your tracking never captures. A person has to add that context. Model choice is a policy decision too: a first-touch model hands 100% of the contact credits to the first interaction, so picking it changes what your team rewards. HubSpot's connector guidance is direct: "Never indiscriminately rely on any AI outputs to make important decisions."
Claude drafts, human decides
| Analysis task | May Claude draft it? | Human check before acting |
|---|---|---|
| Data sanity check (rows, totals, missing values) | Yes | Match totals to the HubSpot report export |
| Performance summary by campaign and channel | Yes | Spot-check three campaigns in campaign analytics |
| Revenue per dollar of spend | Yes, with formulas shown | Confirm spend figures with the campaign owner |
| Attribution model comparison | Yes | Confirm model and conversion type per file; recalculate one campaign |
| Choosing the reporting model | No, only the questions to answer | Marketing and sales leaders agree and document the choice |
| Anomaly and follow-up list | Yes | Owner confirms each item in HubSpot |
| Why a campaign worked | Hypotheses only | Campaign owner confirms or tests |
| Budget reallocation | No | Marketing director decides with offline and pipeline context |
| Tracking and settings fixes | Drafts a fix list | HubSpot admin makes and tests the change |
What are the risks, and how do you check Claude's numbers against HubSpot?
The main risks are data privacy, broad access, wrong calculations and overconfident causal claims. You reduce them with your plan's data settings, HubSpot user permissions, a reconciliation step and prompts that ask for evidence.
- Data privacy. Anthropic's privacy center states: "By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API, Claude Gov, etc.) to train our models." That default covers commercial products such as Claude for Work, so if your team uses an individual plan, check which terms apply before you upload campaign data. It adds that chats may be used for training if you explicitly report feedback, for example with the thumbs up or down button, or otherwise allow it. Set a team rule on feedback buttons, and confirm your privacy obligations in Canada and the United States before uploading contact-level data.
- Access. Connector access follows each user's HubSpot permissions, so review who holds which permissions before people connect.
- Wrong calculations. HubSpot's connector article warns: "Claude can make mistakes." A reconciliation step is how you catch them.
- Causal claims. Attribution shows which interactions came before revenue, not what caused it. Keep the hypothesis line in prompt 2.
Check Claude's numbers in five steps:
- Compare row counts and totals from prompt 1 with the HubSpot report you exported. Attribution reports include up to 20 million interactions per report after sampling, so compare against the same report, not a different one.
- Ask for citations. HubSpot advises: "Ask for citations and references for any important facts and assertions so that you can quickly and easily confirm them in your HubSpot account."
- Recalculate one campaign by hand from the arithmetic Claude showed.
- Confirm the date range and model in every table heading.
- Open the three largest anomalies in HubSpot before sharing them.
How long does a monthly review take, and which plans do you need?
Plan about two hours a month once tracking is clean; that is a planning estimate, not a measured benchmark. You need a paid Claude plan for the connector, file features for exports, and Marketing Hub Enterprise if you use Deal Revenue Attribution.
| Step | Estimated time | Owner |
|---|---|---|
| Export reports and run prompt 1 | 30 minutes | Marketing ops |
| Prompts 2 and 3, with spot checks | 45 minutes | Marketing ops |
| Prompt 4 and anomaly checks in HubSpot | 30 minutes | Marketing ops with campaign owners |
| Write-up and review | 15 minutes | Marketing director |
Requirements, citing only public documentation:
- Claude. The connector requires Pro, Max, Team or Enterprise. Anthropic's pricing page lists a plan at $20 if billed monthly, or $200 billed up front with the annual subscription discount, and notes: "Price and plans are subject to change at Anthropic's discretion." Check current pricing and billing currency before you budget, especially for Canadian teams.
- HubSpot. Deal Revenue Attribution is for Marketing Hub Enterprise accounts only. Check the availability of other campaign and attribution reports against your own subscription.
If you want help running campaign reporting as an ongoing practice, see the marketing operations service.
When should you fix tracking and campaign tagging before using Claude?
Fix tracking first if campaigns lack consistent tracking URLs, if interaction types are off in your revenue attribution settings, or if deals close without associated contacts. Claude can only analyze what HubSpot recorded, so weak tagging produces confident analysis of the wrong data.
HubSpot defines the basics in its guide to creating tracking URLs: "A tracking URL is a URL with parameters attached." Visits from a tracking URL are attributed to its campaign in the sources report and count toward campaign analytics. UTM conventions are your team's rules for naming the source, medium and campaign values in those parameters. If one person writes paid-search and another writes ppc, Claude sees two channels.
Run this checklist before your first review:
- Interaction types for your assets are turned on in revenue attribution settings; HubSpot warns that otherwise the attribution reports will not populate.
- Every paid and partner link uses a tracking URL tied to a HubSpot campaign. A tracking URL appears in campaign analytics only after at least one visit after the campaign was created.
- Deals are associated with contacts before they are set to closed won.
- Campaign names follow one written convention.
- Spend is entered for every campaign you plan to compare.
If several boxes stay unchecked, set up tracking and attribution properly first. If you need help getting campaign tracking and attribution set up correctly, look at the HubSpot implementation service. Teams with messy or migrating data should start with How to Use Claude to Map Fields and Plan Data Cleanup Before a HubSpot Migration, and leaders standardizing workflows can begin with How to Improve Marketing Operations in 2026.
Frequently asked questions
Can Claude build charts from my HubSpot exports?
Yes. The Claude Help Center says you can upload CSV, TSV and other files for data analysis and visualization. Ask for the chart and the table behind it so you can check each value.
What if my export is bigger than Claude's file limit?
Claude allows 30MB per file. Split the export by date range or campaign, or remove columns the review does not use, and run prompt 1 on each file.
Should I give Claude contact-level data or aggregated reports?
Start with reports aggregated by campaign; they answer most review questions. Pull contact or deal records only to investigate a specific anomaly, and only within your privacy policy.
How does HubSpot's Empirical model fit the comparison?
HubSpot says the Empirical model replaces the U-shaped, W-shaped, J-shaped and Inverse J-shaped models and uses historical interaction data to weigh interaction types. If your reports use it, add it as a column in prompt 3.
How should I keep a record of each month's review?
Save the exported files, the four prompts as run, Claude's output and your corrections in one dated folder. Next month, prompt 2 can compare against last month's verified totals instead of recalculating them.
Sources
- Set up and use the HubSpot connector for Claude, HubSpot (November 12, 2026)
- Analyze individual campaign performance, HubSpot (August 3, 2026)
- Create and edit files with Claude | Claude Help Center, Anthropic (Claude Help Center) (November 5, 2026)
- Analyze performance across campaigns in Marketing Studio, HubSpot (October 9, 2026)
- Understand attribution reporting, HubSpot (October 20, 2026)
- Create attribution reports, HubSpot (August 18, 2026)
- Campaigns | Frequently Asked Questions, HubSpot (August 3, 2026)
- Is my data used for model training?, Anthropic Privacy Center (August 18, 2026)
- Plans & pricing | Claude by Anthropic, Anthropic
- Create tracking URLs, HubSpot (October 20, 2026)
Want help putting this into practice?
