The 30-Day AI ROI Build Playbook

Your CFO cannot see what the AI spend did, and the vendor dashboard will not answer them in a review. The 30-day AI ROI build is the four weeks you use to put a measurement system in place they can read and challenge: which leads the tools touched, what those leads cost once the tool invoice is in the math, and how that compares with leads the tools never touched. The four inbound lead ROI metrics are the broader model. This post is the week-by-week build of them. PwC's 2026 AI Performance Study (n=1,217 senior executives, 25 sectors) found that 74% of AI's economic gains flow to 20% of organizations. The organizations capturing most of those gains had connected tool activity to business outcomes before they increased spend.

A 30-day AI ROI build produces three artifacts a CFO can read.

By day 30 you should be able to hand over three things. The first is a tagged pipeline in the CRM that separates leads an AI workflow touched from leads it did not. The second is a cost-per-outcome figure for each AI tool that was live in the period, with the tool's cost sitting in the numerator. The third is a four-row summary a CFO can read without you in the room to translate it. Those three outputs are what let you show how the tools are working against a baseline you wrote down before the sprint, instead of asserting that AI is working because the team feels busier.

The sprint does not produce a final ROI number. Thirty days of data is too short for B2B and SMB sales cycles, which typically run 60 to 180 days. What you have at day 30 is the instrumentation that can produce a defensible number at 90 days, at six months, and at the annual review. If you skip the instrumentation and wait for a closed-won miracle, the next review arrives with the same gap you have now.

The three artifacts are a tag, a baseline, and a first comparison.

Artifact one is the CRM tag structure. You add two custom properties on the Contact or Lead object so you can see which leads AI workflows touched and which tools were involved. Artifact two is the baseline you captured before day 1: pipeline for the last completed quarter, cost per qualified lead, sales cycle length, and AI tool cost per month. Artifact three is the first cohort comparison, a side-by-side of AI-touched and non-AI-touched leads on conversion rate, pipeline value, and cost per outcome.

Each week ends with something you can point at.

Week 1 ends with the two CRM properties created and every active AI touchpoint written on one page. Week 2 ends with the first pipeline query run and both cohorts exported and compared. Week 3 ends with cost per outcome calculated for the AI-touched cohort, tool cost included. Week 4 ends with the four-row summary written and read by someone who was not in the build.

Most teams never measure AI ROI because nobody owned the question.

McKinsey's March 2025 report, The State of AI: How Organizations Are Rewiring to Capture Value (n=1,363), found that fewer than 1 in 5 companies track well-defined KPIs for their gen AI deployments, and that the companies which do see the greatest bottom-line impact. The CRM already has pipeline data. The tool invoices are already in the budget. What is missing is a person who owned the question "how do we know if this tool is working?" before the tools went live, and a baseline written down before the first invoice.

Without that baseline, every result after launch can be explained three different ways. A rise in marketing-qualified leads after you turn on a scoring tool might be the tool, a seasonal bump, a new campaign, or a rep who started sourcing their own meetings. The baseline is what makes those explanations separable. Until it exists, the CFO is right to treat the spend as unproven.

PwC found that the gains sit with the teams that measure outcomes.

PwC's 2026 AI Performance Study analyzed 60 AI management and investment practices across 1,217 senior executives in 25 sectors. The study found that 74% of AI economic gains are captured by 20% of organizations. The separating practice is that the leaders measure growth outcomes, while the rest measure what the tools do: tasks processed, emails sent, leads scored. The outcome list is shorter and harder: pipeline added, cost per lead, sales cycle length. The 30-day build is a way to get onto that shorter list before the next budget conversation.

Workflow redesign is the McKinsey factor that tracks with EBIT.

Across 25 organizational attributes tested, McKinsey found that workflow redesign is the single highest-correlation factor for EBIT impact from gen AI. Only 21% of organizations using gen AI have fundamentally redesigned at least some workflows. The other 79% added tools to the process they already had and measured speed. The 30-day build forces the redesign in a narrow way: a tool stays in the measurement system only when you can name the outcome it is supposed to move.

You need four baseline numbers before day 1.

The pre-build work takes one afternoon, and it has to happen before you instrument anything. Write down four numbers: pipeline per quarter from the last completed quarter, cost per qualified lead from the last 90 days, median days to close on closed-won deals from the last 90 days, and total AI tool cost per month. Every claim you make at day 30 and after compares against those four. If the tools are already live and you never captured them, you cannot tell whether results improved, stayed flat, or got worse. You can only tell a story.

Each number comes from a report you can rerun.

Pipeline per quarter is the CRM report for the last completed quarter, summing closed-won and open pipeline. Cost per qualified lead is the marketing budget for the period divided by the leads sales accepted, which means the ones sales marked as qualified, not every marketing-qualified lead the system created. Median days to close is the closed-won report for the last 90 days, sorted by the days between lead creation and close, taking the midpoint. AI tool cost is the sum of monthly subscription fees plus any per-seat or per-use charges from the last billing cycle. Write the date range next to each number so the next person can pull the same window.

You can still build a baseline if the tools are already live.

If the AI tools went live before this sprint, use a matched set of leads and closed-won deals from the 90 days before the go-live date as the comparison group, and label that set as the baseline cohort. It is less precise than a true before-and-after, because other things changed in the business during those months. It is still enough for a first measurement cycle, and it is the alternative to claiming a return with nothing to compare against.

Week 1 of the build is tags and an inventory.

Week 1 is instrumentation only. You do not analyze, and you do not claim a return. The goal is to confirm that the data your Week 2 query needs will exist and will be filled in correctly. That is two tasks. Create two custom CRM properties on the Contact or Lead object, and write every active AI touchpoint in the marketing motion on one page. For each tool, the page names what it does in the workflow, which CRM object it touches, and what event or field update it produces. That page is what you pull out when the CFO asks which tool contributed to which result.

A touchpoint that never writes to the CRM cannot be measured yet.

Give each tool one row: the name, the function (email sequencing, lead scoring, content generation, or ad optimization), the CRM object it affects (Lead, Contact, Deal, or Company), and the specific field or event it creates or updates. If a tool does not write to the CRM, ask whether the vendor provides an API that can. If neither exists, leave that tool out of the measurement system until one of those paths exists. Measuring it from a vendor screenshot will not survive the review.

Two properties make the cohort query a single filter.

Create ai_touched (a boolean that defaults to false) and ai_tools_active (text or a multi-select that lists the tool names). Both update whenever an AI workflow fires against that lead. Week 2 then becomes one CRM filter, rather than a join across the tool's dashboard and the CRM that nobody has time to rebuild before the meeting.

Week 2 of the build is the first cohort comparison.

Week 2 is the first pipeline query. Pull two lists from the same 30-day window: leads where ai_touched is true, and leads where it is false. Compare pipeline stage distribution, the rate at which marketing-qualified leads become sales-qualified, and average deal size. You are not testing statistical significance yet. You are checking that the comparison can be produced at all. IDC's Business Value of AI study (Microsoft-sponsored, January 2025) found that gen AI delivers an average of $3.70 return per dollar invested across organizations with measurement infrastructure. Week 2 is the first output of that infrastructure, which is the comparison itself, not the return figure.

The query is two exports and four calculations.

In Salesforce, filter Contacts or Leads by AI_Touched__c = true for the period and export them. Run the same export with AI_Touched__c = false. For each cohort, calculate the count of leads, the pipeline value from associated opportunities, the marketing-qualified to sales-qualified rate (sales-qualified count divided by the cohort count), and the average deal size. That comparison is the first concrete measurement the sprint produces. If the export is empty, the Week 1 properties are not being written, and you fix that before you interpret anything.

The four-row slide shows the comparison and stops there.

Row 1 is pipeline value for the AI-touched cohort against pipeline value for the other cohort, same period. Row 2 is the marketing-qualified to sales-qualified rate for each. Row 3 is average deal size for each. Row 4 is AI tool cost for the period. Leave the implied return for later. Let the reader do the math from numbers they can see. The payback section of our agent cost guide shows how to extend this into a payback reading once you have 90 days of data.

Week 3 of the build puts tool cost into the outcome figure.

Week 3 answers a narrower question than "is AI working?" Take the AI-touched cohort from Week 2, add the AI tool cost for the period to the marketing budget in the numerator, and recalculate cost per qualified lead for that cohort. You are replacing hours saved as the primary metric. The hours saved is not ROI post is the longer argument for that swap. The question you can answer this week is whether, after the tool's own cost is included, AI-assisted leads cost less or more per outcome than leads the tools did not touch.

The formula is the budget plus the tool, divided by accepted leads.

AI-inclusive cost per qualified lead equals marketing budget for the period plus AI tool cost for the period, divided by marketing-qualified leads that sales accepted from the AI-touched cohort. The comparison figure is marketing budget for the period divided by marketing-qualified leads that sales accepted from the other cohort. If the AI-inclusive figure is lower, the tool is reducing cost per outcome, and that is a signal a CFO can read. If it is higher at 30 days, write it down as early-cycle data and put a 90-day recheck on the calendar before anyone treats it as a verdict.

A flat result at 30 days tells you the cost baseline.

No improvement at 30 days does not prove the tool is failing. Sales cycles often run longer than the sprint, and closed-won data this month still reflects deals that were already moving before the tool went live. What a flat result does tell you is the current cost per outcome with the tool included in spend. You need that number for the 90-day comparison. Without it, the next review is another conversation about feelings.

Three checks come before you present the Week 3 numbers.

First, the AI-touched cohort should cover at least 15% of total leads in the period. Below that, the comparison is too thin to mean much. Second, the comparison period should be the same season as the current period, because a summer cohort against a January cohort will lie to you. Third, name any concurrent change that hit the period: a price change, a sales hire, a campaign launch. If one of those happened, the footnote says so. A clean slide with a hidden confound is how measurement loses the room.

Week 4 of the build is the page you hand the CFO.

Week 4 is a written summary a CFO can read in five minutes and argue with. It has four elements: the four baseline numbers from before day 1, the 30-day data from Week 2 and Week 3, the AI tool cost for the period, and the implied return with the period and the cohort sizes named. Say in the document that this is an initial measurement. Thirty days does not close the loop on a sales cycle of 60 to 180 days, and pretending otherwise is how the next review gets harder.

Show the formula, the counts, and the tool cost.

Use the numbers from your export, and show the denominator next to any rate. A sentence the CFO can check names the cost per qualified lead on AI-touched leads, the same figure on leads the tools did not touch, the dates, the count of leads in each cohort, and the tool cost included in the AI-touched numerator. A percentage on its own, with no baseline, no count, and no tool cost in the math, creates more questions than it closes. The point of the page is that they can recompute it.

One page, four rows, three footnotes.

Row 1 is pipeline value from the AI-touched cohort against pipeline value in the baseline period, at a comparable cohort size. Row 2 is cost per qualified lead, AI-inclusive, against the non-AI-touched cohort. Row 3 is a CAC payback estimate, using the AI system benchmark as the industry comparison point. Row 4 is total AI tool cost for the period. Footnote 1 is the data window, with the dates. Footnote 2 is the cohort sizes. Footnote 3 names concurrent changes that may affect the comparison. Under the table, one paragraph states the next review date and the specific metric that will trigger a decision to expand the tool or pause it.

Three blockers stop the 30-day AI ROI build before it finishes.

Three things stop the sprint before it produces the summary. The first is CRM access: marketing cannot create custom properties or run export reports without revenue-operations approval, and that cycle runs two to three weeks if you ask on day 1 instead of the week before. The second is tool data: the AI tool pushes summary stats to its own dashboard and never sends event-level data to the CRM. The third is attribution design: nobody has written down whether pipeline credit goes to the last AI touch, the first, or is split across the touches in the buying cycle.

Each blocker has a fix you do before day 1.

For CRM access, ask revenue operations to create the two properties before the sprint starts, and describe it as two fields, not a systems project. For data availability, confirm with the vendor before day 1 that event-level activity can reach the CRM through an integration or an API. If neither exists, that tool sits out of the sprint. For attribution, decide in writing before day 1. Last-AI-touch is the cleanest starting model for most teams, and you can refine it at 90 days once you have seen how the touches actually fall.

One question tells you whether the sprint is 30 days or 60.

Ask revenue operations this, on the day you want the sprint to start: "Can you create two custom properties on the Contact object, and can our AI tool integrations write to them?" The answer, and how long it takes to get the answer, tells you whether you will finish in 30 days or 60. If the answer takes more than two business days, escalate. Everything else in the build waits on those two properties.

The methodology behind this 30-day AI ROI build.

Three published sources frame the sprint. McKinsey's March 2025 "The State of AI: How Organizations Are Rewiring to Capture Value" (n=1,363) is the source for the measurement gap (fewer than 1 in 5 companies track well-defined KPIs for gen AI) and for the workflow finding (21% of organizations have redesigned workflows, the highest-correlation attribute for EBIT impact across 25 tested). PwC's 2026 AI Performance Study (n=1,217 executives, 25 sectors, 60 practices analyzed) is the source for the concentration figure, 74% of AI gains going to 20% of organizations. IDC's Business Value of AI study (Microsoft-sponsored, January 2025) is the source for the $3.70 return per dollar invested among organizations that already have measurement infrastructure. The week-by-week artifact list is the operational version of the model in the inbound lead ROI post.