Business Intelligence KPIs: Which to Track in 2026

Two people present monthly recurring revenue at the same board meeting. Finance pulled it from the accounting ledger, the sales lead pulled it from the CRM, and the two numbers disagree. Nobody in the room can say which one is right, so the meeting spends its first twenty minutes on arithmetic instead of on the business. That is not a dashboard problem. It is a definition problem, and it is the most common way a business intelligence KPI program falls apart in its first year.

Almost every KPI listicle you will find is a menu of names: MRR, CAC, NPS, DSO, churn. Names are the easy part. The hard part, and the part that decides whether anyone trusts the number six months from now, is answering one question for each metric: which single system is allowed to produce it? This guide is a catalog of KPIs by function, and for every one of them it names the system of record, the definition traps that make two honest people disagree, and the review cadence that actually fits the metric.

Every business intelligence KPI needs one named source system

Source-of-truth discipline is four rules, and they are boring in the way that load-bearing things usually are.

One metric, one system. The system that creates the underlying events owns the metric. Stripe creates subscription charges, so Stripe owns MRR. The bank and the ledger create cash movements, so the accounting system owns runway. The CRM creates deal stage transitions, so the CRM owns pipeline. Anything downstream is a report, not a source.

Definitions are written before they are charted. A KPI definition is not "MRR." It is "the sum of active subscription items normalized to a monthly amount, excluding one-time charges, taxes, and shipping, net of active discounts, converted to USD at the charge date, measured on the last day of the month." If your definition does not name fields and edge cases, two teams will implement two different metrics with the same label.

Every KPI has a human owner. Not a team, a person. The owner decides what counts and answers when the number looks strange. Unowned metrics decay silently, and the first sign of decay is usually somebody quietly rebuilding it in a spreadsheet.

Wrong-source substitutions are the failure, not rounding. When revenue comes from the CRM, you are measuring what sales hoped would close, not what customers were charged. When traffic comes from the ad platform instead of your own analytics, you are measuring what the ad platform wants to take credit for. The gap between those numbers is not noise. It is a category error.

Here is the master catalog. Read the third column as the contract and the fourth as the argument you will otherwise have later.

Revenue KPIs: the biller owns the number, always

MRR is the most argued-over business intelligence KPI in software, and the argument is nearly always about normalization. An annual plan billed once is monthly revenue divided across twelve months for MRR purposes and a single cash event for the bank. Both are true. They are different metrics, and the moment somebody says "our revenue" without saying which one, you are back in the board meeting.

Track these five, in this order:

• MRR as of period end, normalized, excluding one-time charges, net of discounts, gross of taxes and processing fees.
• The four MRR movements: new, expansion, contraction, churned. The total is the least useful number on the list. The movements tell you what to do.
• ARPA, with a denominator of paying accounts. Rising ARPA with flat MRR means you are losing small customers, which is often good and sometimes the start of a concentration problem.
• Failed payment rate, because involuntary churn is a billing configuration problem wearing a retention costume. Dunning settings, card expiry handling, and retry schedules move this number more than any customer success effort.
• Refunds and credits, tracked separately rather than netted into churn. Netting them makes a product quality issue look like a retention issue.

The traps worth writing into your definition doc: currency conversion happens at a fixed date rule and not at whatever today's rate is, proration events on mid-cycle upgrades belong in expansion rather than in new, and test-mode data never enters production reporting. If you sell through more than one biller, for example Stripe for self-serve and invoices for enterprise, one of them still has to be the arithmetic authority and the other has to be reconciled into it on a schedule with a named owner.

Retention KPIs: cohorts or nothing

Retention metrics computed on the whole customer base at once are close to meaningless because new customers dilute the denominator every month. The cohort is the unit.

Gross revenue churn answers "how much of what we already had did we lose," with the denominator locked at the start of the period. Net revenue retention adds expansion from that same cohort, and it is the number that tells you whether the existing base is a growing asset or a leaking one. The single most common error is letting new customers into the NRR numerator, which turns a retention metric into a growth metric that will flatter you right up until growth slows.

Logo churn is a different question from revenue churn and often points the opposite way. Losing many small accounts while expanding a few large ones produces excellent NRR and a quietly dangerous concentration profile. Track both and look at them side by side.

One state-machine detail decides whether logo churn is trustworthy: paused, delinquent, cancelled, and expired are four distinct statuses in most billing systems, and teams routinely count only the last one. A customer who has failed payment for three cycles is churned in every sense that matters to a forecast. Write the rule down, including the day count at which delinquent becomes churned, and apply it identically every month.

Cash KPIs: the ledger, not the forecast deck

Cash metrics are the ones where the wrong source is most tempting, because the finance model is right there and it already has runway in it. The model is a projection. The KPI is an actual.

• Cash balance: from the bank feed, all accounts, one date convention.
• Net burn: cash out minus cash in over the period, trailing three months averaged, because single months are noisy for reasons that have nothing to do with the business (annual insurance, a big invoice landing on the 30th instead of the 1st).
• Runway: cash balance divided by trailing net burn. Not divided by planned burn. If the plan is the divisor, runway becomes a statement of intent.
• Gross margin: decide once whether hosting, payment processing fees, and customer support salaries sit above or below the line, document it, and never quietly change it mid-year.
• DSO and AR aging: from the ledger, and worth watching weekly rather than monthly if you invoice enterprise customers, because a slipping collections cycle shows up in cash long before it shows up in revenue.

Deferred revenue deserves its own line. Cash collected for an annual contract is not revenue earned, and treating it as such is the fastest way to build a runway number that is confidently wrong.

Pipeline KPIs: the CRM owns them, if hygiene exists

Pipeline is the function where source-of-truth discipline is necessary but not sufficient. HubSpot, Salesforce, Pipedrive, or whatever you run genuinely is the system of record for deals. The problem is that CRM records are mutable and humans mutate them.

The metrics that matter:

Two hygiene notes carry most of the weight. First, current stage is not stage history. If your CRM only exposes where a deal is now, you cannot compute conversion between stages honestly, and you will need the stage-change history object or a nightly snapshot. Second, close dates get pushed. A forecast that reads current close dates without tracking how many times each has moved is measuring optimism. Counting close-date changes per deal is one of the highest-value derived metrics most sales teams never build.

Win rate deserves a specific warning because it is the most abused business intelligence KPI example in sales reporting. Wins divided by all deals, including open ones, produces a number that improves every time a rep creates a new opportunity. The denominator is closed deals, won plus lost, in a stated period.

Marketing KPIs: analytics for behavior, ad platforms for spend

The rule here is a split: your analytics tool owns what happened on your property, the ad platforms own what you spent, and your billing system owns what came back. Never let one of those three answer for another.

Google Analytics is the source for sessions, users, engaged sessions, on-site conversion events, and channel grouping. GA4's event model means "sessions" and "pageviews" are not what they were in the previous generation of the product, and consent settings, ad blockers, and cross-device behavior all shift totals in ways that are real and unfixable. That is fine as long as you compare GA to GA over time rather than to a platform's own numbers.

The ad platforms own spend, impressions, clicks, and their own reported conversions. Their conversion counts will exceed yours, because their attribution windows and view-through rules are theirs. Use platform conversions to optimize campaigns inside the platform, and use your own analytics plus billing data to decide how much to spend overall.

CAC is the metric most often published without a definition. Blended CAC is total sales and marketing spend divided by all new customers. Paid CAC is paid spend divided by paid-attributed new customers. They differ by a lot, both are legitimate, and using them interchangeably across a deck is how boards get misled by teams that were not trying to mislead anyone. Payback period is CAC divided by gross-margin-adjusted monthly revenue per customer, and it is a more honest health signal than any lifetime value ratio, because LTV requires a churn assumption that is often a guess dressed as a number.

Industry context changes which acquisition metrics matter. A hotel group cares about channel mix and pace far more than about session counts, which is why the metric set in Hospitality Business Intelligence: 2026 Software Guide looks so different from a SaaS set. Retailers face the same divergence around inventory-linked demand metrics, covered in Retail Analytics Solutions Compared: 2026 Buyer's Guide.

Support and product KPIs

Support metrics are simple to define and easy to game, so the definition should include the calendar. First response time measured on a 24/7 clock and the same metric measured on a business-hours calendar produce very different results, and only one of them reflects the customer's actual experience of your stated support hours. Pick one, state it, and keep it.

Track first response time, full resolution time, backlog by age, reopen rate, and ticket volume normalized per hundred customers. That last normalization is what turns support volume from a growth artifact into a product quality signal: raw ticket counts rise with the customer base and tell you nothing.

On the product side, activation rate is the highest-leverage number most teams under-instrument. Define activation as a specific event or set of events that correlates with retention, not as "completed onboarding," and measure it at the account level if you sell to teams and at the user level if you sell to individuals. Both are valid. Mixing them within one chart is not.

Choosing a delivery method for business intelligence KPIs

Once the definitions are settled, there is a separate decision that most KPI guides skip entirely: how the number reaches a human. There are three real options, and most companies need more than one.

Dashboards are the default and they remain correct for a stable, owned metric set. If you are choosing a platform for that job, the head-to-head in Domo vs Power BI in 2026: Pricing, Features, Verdict covers the two most common shortlist entries, and Microsoft BI Solutions in 2026: The Full Stack Explained explains how the Power BI licensing and Fabric pieces fit together if you are already a Microsoft shop.

The honest limitation of dashboards for KPI work is that they answer only the questions you anticipated. A dashboard showing churn at a certain level cannot tell you which accounts churned, what their support history looked like, or whether their usage had dropped first. Those follow-up questions are where the actual decision lives, and answering them means opening three other tools.

Where Skopx fits in a business intelligence KPI stack

Skopx is not a dashboard builder and does not try to be. There is no chart designer, no semantic modeling layer, no report scheduler that renders PDFs. If your requirement is a governed set of visual dashboards for a hundred people, buy a BI platform and treat the rest of this section as complementary.

What Skopx does is connect to nearly 1,000 tools a company already uses, Stripe, HubSpot, Google Analytics, QuickBooks, Gmail, Slack and the rest, and let you ask for a number instead of building a view of it. "What was net revenue retention for the accounts that signed in Q1" gets answered from the billing data with the records it used cited underneath, so the definition is inspectable rather than buried in a measure somebody wrote two years ago. That property matters more for KPIs than it does for most analytics work, because a KPI you cannot audit is a KPI you will eventually stop believing.

Three other pieces are relevant to KPI programs specifically. The morning brief pushes the numbers you care about into your inbox before you go looking, which is the delivery mode that beats dashboards for anything a leader should see daily. The insights engine watches connected data for anomalies and surfaces the ones worth attention, for example a failed payment rate stepping up or pipeline creation dropping against its own trend. And workflows, built by describing them in chat rather than by wiring nodes, can assemble a recurring KPI package on a schedule.

Monday KPI brief

Monday 07:00

Runs before the weekly leadership standup

Stripe billing

MRR plus the four movements and failed payments

HubSpot pipeline

Pipeline created, coverage, aging deals

Google Analytics

Sessions and on-site conversion by channel

Accounting ledger

Cash balance and trailing net burn

Compose brief

Week over week deltas with sources cited

Flag outliers

Only metrics outside their normal range

Post to Slack and email

One message, no dashboard to open

Skopx runs on your own AI key with zero markup on model usage, and seats are $5 per month for Solo and $16 per seat per month for Team, listed on pricing. More detail on building this kind of recurring package lives on the workflows page.

A KPI program you can stand up in four weeks

Week one: pick twelve. Two or three per function, no more. Every additional KPI costs review attention, and attention is the scarce resource. If you cannot name the decision a metric would change, cut it.

Week two: write the definitions. One page per metric: the formula, the source system, the field names, the edge cases, the owner, the review cadence. This is the deliverable that survives staff turnover.

Week three: connect the sources and reconcile. Compute each metric from its named source and compare it against whatever number the business has been using. Where they differ, find out why before changing anything. The reconciliation memo is often more valuable than the dashboard.

Week four: choose delivery per metric. Daily operational numbers go into a brief or an alert. Weekly management numbers go into the standup packet. Quarterly strategic numbers get built as a proper report. Ad hoc questions stay ad hoc, and should be answerable in chat rather than by filing a request with an analyst.

Smaller companies can compress all of this and should skip the warehouse entirely until a real modeling need appears, a sequence covered in Business Intelligence for Small Business: 2026 Guide.

Frequently asked questions

How many business intelligence KPIs should a company track?

Roughly ten to fifteen at the leadership level, with deeper metric sets owned inside each function. The constraint is not storage or tooling, it is review attention: a list of forty numbers gets skimmed, and skimmed numbers do not change decisions. Every KPI on the leadership list should have a named owner and a clear answer to "what would we do differently if this moved."

What is the difference between a KPI and a metric?

A metric is any measurement. A KPI is a metric that a specific person is accountable for and that a specific decision depends on. Sessions by browser is a metric. Activation rate with the head of product accountable for it and a quarterly target attached is a KPI. Promoting every metric to KPI status is how dashboards become wallpaper.

Should MRR come from Stripe or the CRM?

From the biller, always. The CRM records what sales expected to happen; the billing system records what customers were actually charged. Keeping CRM contract values alongside billing MRR is useful for finding provisioning gaps and billing errors, but only one of the two can be the reported number, and it should be the one where money changed hands.

Do I need a data warehouse to track business intelligence KPIs?

Not at first. If your KPIs come from five or six SaaS tools and nobody needs multi-year cohort analysis across joined datasets, connecting to the sources directly is faster and cheaper. A warehouse becomes worth it when you need joins the source APIs cannot do, history the sources do not retain, or one governed definition shared across several downstream tools.

How often should each KPI be reviewed?

Match cadence to how fast the metric can actually move. Cash position and failed payments are daily. Pipeline created, support backlog, and traffic are weekly. MRR movements, retention, and CAC are monthly, because shorter windows are mostly noise. Reviewing a monthly metric daily produces reaction to randomness, which is worse than not looking.

What is the best BI tool for KPI tracking?

There is no single answer, because the delivery mode decides the tool. For governed recurring dashboards, the mainstream platforms are all credible and the choice usually comes down to your existing cloud and data stack, with the options laid out in Power BI Alternatives: 10 Tools to Consider in 2026. For ad hoc questions that cross tool boundaries, and for pushed briefings rather than pulled dashboards, a chat-based answer layer over your connected tools does the job with far less maintenance. Most companies past a certain size end up running both, and the source-of-truth definitions are what keep the two agreeing.