Data

Business Intelligence Services

You already have dashboards. The question is what happened the last time one of them turned red. In most companies the honest answer is that someone noticed a week later, asked three people what the number meant, got two different definitions, and by then it had moved back on its own. The reporting worked exactly as designed. Nothing happened anyway. Facet's Data division exists because the gap between seeing a number and acting on it is where almost all of the value in business intelligence services is currently being lost. We built the division on our own problem first: Facet runs its marketing and delivery on Agent Loops, which means our reporting is read and acted on by something other than a person before a person ever opens it — and that forced us to define our own numbers to a standard no human dashboard had ever demanded of them.

Why hire a business intelligence services partner?

Because visibility stopped being your constraint some time ago, and you are probably still buying for it.

A dashboard is a place where a decision waits for a human

Business intelligence was designed for a world where the scarce resource was information. Getting a number in front of a decision-maker was genuinely hard, genuinely valuable, and worth building a whole industry around. That world is gone. You can now see almost anything about your business within a day of wanting to. And yet the decisions do not move much faster, because the bottleneck quietly relocated.

• The number is available; the attention is not: A dashboard assumes someone looks at it, understands the movement, knows what to do, and has the authority to do it — four conditions that have to line up on the same afternoon. Most weeks they do not.
• Nobody agrees what the number means: Two teams report revenue differently, both are defensible, and the resulting meeting is about the definition rather than the decision. This is the single most common thing we find, and it has nothing to do with tooling.
• The alert that fires every day is not an alert: Thresholds set once and never tuned become noise within a quarter, and noise trains people to ignore the channel that matters.
• Reporting rewards the retrospective: A monthly dashboard is a very good instrument for explaining what happened and a very poor one for changing it. Companies buy the first and expect the second.

Action item: Take the last five decisions your leadership team made. For each, write down how many days passed between the underlying number moving and the decision being made. That interval is what you are actually buying when you buy business intelligence services.

The reframe: your BI layer's most valuable consumer is no longer a person

Here is the argument this division is built on, and it is the one thing on this page a competitor would not write.

When a human is the only consumer of a metric, "good enough" is generous. Fuzzy definitions survive because a person compensates by knowing what was meant. Stale data survives because a person knows Tuesday's number is Monday's. Inconsistent naming survives because a person recognizes both names.

The moment an agent consumes that metric — to trigger work, route an exception, or decide whether a threshold has been breached — every one of those tolerances becomes a defect. Which is the opportunity, not the problem. Building a data layer an agent can act on forces the rigor your human reporting always deserved and never got, and it converts your dashboards from a viewing surface into a control surface. The reason to fix your definitions is no longer hygiene. It is that a defined metric can have a loop behind it and an undefined one cannot.

That is why discovery, information architecture, and requirements work live in this division rather than in a project function. You cannot automate a process you cannot define, and you cannot define a process whose data does not mean anything specific. Making the business legible is the prerequisite for making it agentic.

We had to make our own numbers agent-legible first

The argument above is not a prediction. It is a description of what happened to us.

Facet's own delivery — engineering, content, research, quality review and reporting — runs through Agent Loops under human command. Our content operation is a working example of it: agents pull the performance data for a published piece, score it against the rubric the piece was briefed on, and feed what they find into the next brief. Nobody reads a dashboard in that sequence. Something acts on the number directly.

The first thing that broke was not the pipeline. It was the definitions. A metric that two people had been quietly interpreting differently for a year is harmless right up until something with no context reads it and acts. We spent the opening stretch of that work not building anything, but settling what our own terms meant — and that experience is the reason this page opens with definitions rather than with tooling.

What it buys you is the expensive half of a BI engagement. The slow, hourly part of this work has always been discovery: senior consultants reading your systems by hand to find out where a number comes from and what it excludes. We put agents on that pass — inventorying sources, profiling fields, drafting a first cut of every metric contract for a person to settle — so the engagement is time-boxed and fixed-fee rather than an open-ended reading exercise you fund by the hour.

• Humans settle the definitions. An agent can show you that two systems disagree about what a customer is. It cannot tell you which one your business meant, and it should not try.
• The failure mode is a confident wrong number. A model asked to interpret an ambiguous field will produce something plausible rather than stop. That is precisely why quality gates run before publication and why a metric without a written owner does not go on a dashboard.
• Agent-produced work clears the same gates as human work. A definition an agent drafted still gets argued about in a room before anyone builds on it.

What the Data division does

Two jobs, in order. Make the business legible, then make the numbers trustworthy enough to act on automatically.

Make it legible

Discovery at Facet is not a warm-up phase before the real work. It is the work that determines whether anything downstream can be automated at all.

• We document how the work really happens: Including the exceptions that live in one person's head, because those are where every automation attempt goes to die.
• We name things once: An information architecture pass settles what an account is, what a customer is, and whether those are the same thing — questions that sound pedantic until two systems disagree in front of a client.
• We write requirements that survive handoff: Specific enough that a builder and a reviewer reach the same conclusion without a meeting. That is a higher bar than most requirements documents clear.
• We surface the decisions nobody has made: Most stalled data projects are not blocked on engineering. They are blocked on an unmade business decision that everyone has routed around for two years.

Action item: Ask three people in different teams to define your primary revenue metric in one sentence, separately. If you get three sentences, that is your first project, and it is not a tooling project.

Make it trustworthy

Reporting people trust is a product of a small number of unglamorous commitments, held consistently.

• Every metric has a written contract: What it counts, what it excludes, where it comes from, how fresh it is, and who owns it. If a number cannot be described that way, it does not go on a dashboard.
• Freshness is a stated commitment, not a hope: Each metric carries an expectation for how current it is, and a visible signal when it is not met. Silently stale data is worse than a missing chart.
• Quality checks run as gates: Row counts, nulls, ranges, and referential checks run on every load, and a failure stops the number from publishing rather than publishing it wrong.
• We build on the modern stack you can staff: Cloud warehousing with dbt for transformation, because the transformation logic should be version-controlled, tested, and readable by someone you hire next year.

Marketing is where this gets tested hardest

Marketing is usually where a company first discovers its numbers cannot be defended, because marketing is the function most often asked to justify its budget with a figure somebody else is entitled to challenge.

The test is never the first question. It is the second one. Pipeline was up last quarter survives a leadership meeting comfortably. Which channel, over what window, and what happens to that figure if we credit first touch instead of last usually does not — not because the answer is wrong, but because there is no written definition to fall back on and three systems each hold their own version of it.

• Attribution is a definition problem before it is a modeling problem: What counts as a lead, when an opportunity is sourced rather than influenced, and how long the window stays open are business decisions. Settle them deliberately, or whichever tool you bought settles them for you.
• One revenue definition, shared with finance: The fastest way to lose a marketing argument is to arrive with a number finance does not recognize. A metric contract settles that in advance rather than in the meeting, which is usually the difference between defending the budget and defending the spreadsheet.
• Content and campaign performance on the same spine: Which pieces and which programs actually preceded revenue, measured against the definitions the rest of the business already uses — so content strategy gets argued from evidence instead of from a shared sense of what feels like it is working. That is what Analytics & BI builds on top of these contracts.
• Then the number can act: A defined acquisition-cost or pipeline metric can have an Agent Loop behind it. Cost per acquisition crossing its threshold opens the work item with the campaign context attached, on the day it happens, rather than being spotted at the monthly review three weeks later.

This is also the most common place a first metric contract gets written, because marketing feels the cost of an undefined number sooner, and far more publicly, than any other function does.

What the loop actually does

Once a metric is defined precisely enough to be acted on, you can put an Agent Loop behind it. Three passes sit inside that loop, and they are agent work for an unglamorous reason — each is continuous, mechanically checkable, and the kind of thing a person does well for a month and then stops doing.

• It watches, so nobody has to: Continuous evaluation against the threshold, rather than a person remembering to look on Thursday.
• It routes, rather than notifies: A breach opens the specific work item with the relevant context attached, assigned to whoever owns it — instead of a message in a channel that everybody assumes somebody else is reading.
• It records what happened next: The action taken gets logged against the trigger, so you accumulate a history of which thresholds actually predicted something worth doing and which were noise. That history is what lets you tune them.

What a person still decides

The loop escalates. It does not set policy, and a data partner who blurs that line is selling you something you should not buy.

• What the metric means. An agent can show you that two systems disagree about what a customer is. It cannot tell you which one your business meant, and it should not try.
• Which thresholds deserve a trigger at all. Set them tight and the loop becomes the daily alert nobody reads. That number is a judgment about how much interruption your business is willing to absorb, revisited as that history accumulates.
• What the loop may act on alone. What runs unattended, what needs a signature, and who owns an exception is agreed and written down before anything runs. That is a risk decision about your business, not a configuration setting.
• What to do about the number. Reporting narrows the options; it does not choose between them. An engagement that is quietly hoping the data will make an unmade decision ends unhappily.

Putting that loop in place is also where this division hands off to the rest of Facet, and it is a ladder rather than a bundle. Building the loop behind a defined metric is Agentic Engineering. Keeping it accurate as your business, your data and your thresholds move is AI Managed Services. Running that pattern across every operating area at once — not just the numbers, but the work they trigger — is the Agentic Operating System. You can stop at any rung, and plenty of clients stop at the first.

How it works

The cost nobody puts on an invoice

Numbers below are illustrative. Run your own and the conclusion usually survives.

Take a team making twelve operationally significant decisions a month off reported numbers — pricing, staffing, spend reallocation, inventory, escalation.

Ninety-six decision-days a month, recovered (illustrative). The reason this never gets funded is that decision latency has no invoice. Nobody submits an expense report for the eight days a price change sat unmade. It shows up instead as a margin number that was slightly worse than it needed to be, in a quarter where nothing obviously went wrong — which is exactly the kind of cost an organization can carry indefinitely without ever deciding to.

Action item: Pick your single most consequential recurring decision and measure only its latency for one month. One number, honestly measured, makes this case better than any dashboard.

Which of your numbers could have a loop behind it instead of a meeting? Book the Metric Contract Workshop — one working session plus one week, fixed fee. You leave with your top operating decisions written down, the metrics behind them defined in one sentence each, and an honest map of which ones your current data can support and which ones it cannot.

Who should not hire this division

• You want a dashboard by the end of the month: If the goal is a chart for a board meeting, a business analyst and your existing tool will get you there faster. We will end up asking questions about definitions that you do not have time to answer.
• Nobody is empowered to settle a definition dispute: If two departments have disagreed about a metric for two years, that is a governance problem. We can facilitate the decision; we cannot make it, and an engagement without an internal decider stalls in week three.
• You are hoping data will make an unmade decision for you: Numbers narrow options. They do not choose. Engagements that begin as analysis and are secretly hoping for permission tend to end unhappily.
• The data genuinely does not exist yet: Sometimes the honest finding is that you need instrumentation before you need analysis. We will say so, and that conversation is cheaper before the engagement than during it.

Business intelligence services FAQ

What are business intelligence services? They are the design, build, and governance of how a company turns its raw operational data into numbers it can act on — covering the definitions, the pipelines, the reporting, and increasingly the automation that sits behind a threshold.

• Definition comes before tooling: Agreeing what each metric counts, excludes, and is sourced from is the step that determines whether anything downstream is trusted.
• Pipelines make it repeatable: Moving and transforming data on a schedule, with tests, so the number is produced the same way every time rather than rebuilt by hand each month.
• Reporting makes it visible: Dashboards and scheduled views aimed at the specific decisions people actually make, not at everything that could be counted.
• Governance keeps it true: Ownership, freshness expectations, and quality gates, so the reporting stays reliable after the people who built it move on.
• Automation makes it act: Loops that watch defined metrics and route work when a threshold moves, which is where the measurable return now concentrates.

How is this different from just buying a BI tool? A tool renders whatever you give it, faithfully and fast, including a wrong number. The work that makes the output trustworthy happens before the tool.

• Tools do not settle definitions: Two teams can build contradictory dashboards in the same tool in the same week, and both will look authoritative.
• Tools do not create ownership: Somebody has to be accountable for a metric being right. Software does not volunteer.
• Tools do not fix the source: Reporting reflects the state of your operational systems. If the data is entered inconsistently, better visualization only distributes the inconsistency more widely.
• Tools do not act: Rendering a chart and doing something about it are different problems, and the second one is where the time is actually being lost.

How long before we get something usable? The first working reporting lands in weeks, because we scope to a small set of decisions rather than attempting to model the whole business at once.

• We start from decisions, not tables: The first question is which recurring decisions matter most, and we work backward to the minimum data required for those.
• Definitions are settled early: Metric contracts for the first set get agreed in the opening week, because everything after that depends on them.
• First reporting comes before full coverage: A small number of trustworthy metrics beats broad coverage nobody believes.
• Coverage expands as it earns it: We widen once the first set is used in real decisions, rather than on a predetermined build plan.

Do we need a data warehouse to start? Not necessarily, and buying one first is a common and expensive way to defer the real problem.

• Start where the decisions are: If the numbers that matter live in three systems, we work with those three. A warehouse becomes worth it when joining sources is the actual constraint.
• The definitions transfer: Metric contracts agreed now remain valid when the architecture changes later. That work is never wasted.
• Warehouses do not clean anything: Centralizing inconsistent data produces centralized inconsistent data, on a larger bill.
• We will tell you when it is time: Usually when the manual joining becomes the bottleneck, or when the volume makes querying the sources directly impractical.

How does this connect to your agentic work? Directly — a defined, governed metric is the precondition for putting a loop behind it, and this division produces the definitions.

• Agents need unambiguous inputs: A metric a person can interpret by context is a metric an agent will misread. Precision stops being pedantry.
• Discovery output is agent-ready by design: The process documentation and requirements we produce are exactly what a decomposition needs, so the work carries forward rather than being redone.
• You can stop at legible: Plenty of clients take the definitions, the reporting, and the governance and never automate a thing. That is a complete engagement, not a failed one.

Make the Business Legible

The unglamorous truth of this work is that most reporting problems are definition problems, and most definition problems are decisions nobody has made. Fixing that is not a technology exercise, which is why we put discovery in the same division as analytics rather than treating it as a preamble.

Facet has been systematizing how businesses run since 2014, was named a CIOReview Top 10 Digital Transformation Services Company in 2021, and builds on the modern open-source data stack — cloud warehousing and dbt — so the logic behind your numbers is version-controlled and readable by whoever you hire next. Pricing is a fixed fee, quoted before work starts, with no hourly billing.

Which of your numbers could have a loop behind it? Start with the Metric Contract Workshop: one working session plus one week, fixed fee. You get your top ten operating decisions written down, the metrics behind them defined in one sentence each with source and freshness stated, a named owner per metric, and a map of which ones your current data can support and which ones it cannot. The definitions are yours whether or not we build anything.

Some of those workshops end with the opposite of a proposal. If the map shows your data cannot yet answer the decisions you are trying to make — that what you need is instrumentation, not analysis — that is the finding, and it is the one worth the most to you. It also means we are pointing you at a smaller piece of work than the one you arrived asking for, which is a trade we will make every time rather than build reporting on numbers that are not there.