Illustrative scenario. The numbers and charts on this page are a designed example built from hand-set parameters chosen to demonstrate the interface. No Meridian model has been fit to customer data here, and no sampler has run. Nothing on this page is a measured fit-quality claim.
Marketing Measurement · Robyn & Meridian methods

The marketing operating system that measures, executes, and learns.

Compass measures what your marketing actually caused, then tells you which of those numbers you can trust. Fitted response curves per channel, a budget optimiser that respects your constraints, and a stability check that shows which channels the data identifies and which the modelling choices decide.

Five design partner slots open for Q3 2026
Method
Robyn-style
ridge + RSSD, Meridian ROI priors
Tuned for
B2B sales cycles
26-week adstock memory
KPI
Pipeline created
Not last-touch revenue
Channels
Six B2B channels
Including field events & SDR
The problem

Last-touch broke. MMMs built for DTC don't fit B2B.

B2B SaaS marketing leaders are stuck between attribution dashboards that can't see anything but the last click, and enterprise MMM engagements priced for Fortune 500 brands. Neither fits a company with a six-month sales cycle and $2–10M in annual paid spend.

01

Your six-month sales cycle laughs at last-touch

By the time pipeline turns to revenue, the original marketing touch is far outside the attribution window. Last-touch and multi-touch dashboards systematically under-credit the channels, events, podcasts, brand, that warm the deal in the first place.

02

DTC-shaped MMMs miss your biggest channels

Off-the-shelf marketing mix models assume an 8-week attention span and a DTC channel mix. They literally have no way to model field events, SDR outbound, or the way a LinkedIn ad in March still influences a deal that closes in September.

03

Spreadsheets can't carry uncertainty

CFOs don't want a number. They want a number with confidence around it. The Excel models marketing teams use to defend budget can't express "ROI between 2.4× and 3.8× with 90% probability", so the conversation defaults to "trust me."

Measure

See which channels actually create pipeline

Compass fits a Bayesian hierarchical model with Hill saturation and 26-week geometric adstock tuned for B2B sales cycles. Every channel gets a fitted response curve with an uncertainty band, not a point estimate, so you can see exactly where each channel is under-invested, healthy, or past the knee.

  • Fitted response curves for every channel, with uncertainty bands
  • Adstock decay tuned for B2B sales-cycle memory, not DTC click windows
  • Contribution decomposition that separates paid lift from baseline pipeline
Response curvesdesigned median + assumed range
Pipeline created ($) Quarterly spend ($) Field Events · current LinkedIn · current Content Syn · saturated $0 $300K $600K $900K $1.2M
Allocate

Get a defensible recommendation, not a guess

Compass runs a Budget Optimizer over the fitted response curves and returns the budget reallocation that maximizes incremental pipeline subject to per-channel floors and ceilings. You don't get a recommendation pulled from a spreadsheet, you get one walked along the fitted response curves, with the uncertainty in each channel's ROI carried through to the answer.

  • Constrained reallocation with per-channel multiplicative bounds
  • Expected lift reported with a bootstrap interval, not a single point
  • What-if sliders to explore alternatives interactively
Budget reallocationCurrent → Optimal · Q3 2026
Field Events $300K $750K +150% SDR Outbound $220K $258K +17% Podcast Sponsorships $180K $205K +14% LinkedIn Ads $550K · $563K +2% Paid Search $850K $600K −29% Content Syndication $400K $125K −69% Projected pipeline lift +$1.57M Assumed range: $1.10M to $2.04M
Defend

Walk into the CFO meeting with the uncertainty attached

Every per-channel ROI lands with an interval, and every interval is labelled with what it is conditional on. Compass also refits the same panel across a range of defensible specifications and shows you how far each channel's ROI moves. When finance asks "how sure are you?", the honest answer is different for each channel, and this is the number that tells you which.

  • Per-channel ROI intervals from a block bootstrap, rendered as distributions
  • Holdout error and a multi-specification stability sweep on every fit, separating channels the data identifies from channels the specification decides
  • One-page analyst brief you can hand to finance unedited
Per-channel ROI rangesdesigned median + assumed range
CHANNEL ROI (× pipeline / $ spent) Paid Search 5.99× LinkedIn Ads 5.75× SDR Outbound 3.95× Podcast Sponsorships 3.40× Content Syndication 2.53× Field Events 2.39× break-even (1×) ROI > 1.0× = channel pays for itself in pipeline created. Width of bar = assumed range for this illustrative scenario.
Roadmap · Next build

One agent, once the measurement layer earns it.

Most MMM tools fail before the model runs, because the panel that went in could never have supported the answer that came out. So the first thing we are automating is the gate, not the commentary.

🔍
In development
Data Quality Auditor
Runs fifteen deterministic checks over a weekly spend panel before anything is fitted: missing weeks, unit mismatches, spend that never varies enough to be identifiable, single-week outliers, and mid-series regime changes. No language model, no judgement calls, no per-run cost — just thresholds that either pass or fail.
OUTPUT: A verdict — ready to fit, needs fixes, or cannot be fitted — plus every failing check, the value that failed it, and the threshold it missed.

Earlier drafts of Compass described five agents. Four are shelved until the measurement layer is proven on real client data.

Command Center Build · Execution Side

The execution side of the operating system.

Measurement tells you where to spend. The Command Center is where the work actually gets made. AI agents draft your comparison pages, paid creative, lifecycle email, and content. A shared Playbook grades every draft against your brand and channel rules before a human sees it. You review, approve, or reject. Every rejection gets logged. Overnight, a Marketing Systems Agent turns recurring rejection reasons into proposed Playbook updates that a Director approves. The system only ever gets more specific, never looser.

  • Five specialized content agents, one per asset type
  • Shared Playbook enforcement gate before human review
  • Tiered approval queue: analyst, growth lead, director, CMO
  • Nightly self-improvement loop with human approval on every Playbook change
  • Measurement Lab tab wires the MMM directly into the same governance model
View live Command Center demo → Talk to us about a custom build

Delivered as a 30 to 45 day engagement. Your version of the seven-tab Command Center, tuned to your funnel, channels, review criteria, and approval tiers. Available as a design-partner engagement or a standard commercial build. Post-delivery retainer optional.

Command Center · Live prototypeseven tabs · self-improving
Daily Pulse
Signups · funnel health · channel mix
Acquisition Funnel
Free → activated → paid
Content Systems
5 agents · shared Playbook
Quality Gate
Human review · logged reasons
Learning Loop
Nightly Playbook proposals
Measurement Lab
Holdouts · geo lift · MMM
Action Center · tiered approval queue with pre-committed dose branches
How it works

From CSV to readout in days, not months.

No SaaS sign-up, no integrations to wire. For our first five design partners, getting started is two days of CSV prep on your end and a full model fit on ours.

Step 01

Send a CSV

Weekly paid spend per channel, weekly pipeline created, and a few simple controls. Six quarters of history minimum, most CRMs export this in under an hour.

~2 hours of work on your end
Step 02

We fit the model on your data

Compass fits a regularized response model with B2B-tuned adstock and Hill saturation, carrying explicit priors on each channel's ROI so experiment results can be fed back in as evidence. Every fit is then refit across a range of defensible specifications, because on real panels the choice of baseline can move a channel's ROI further than the data does. Model fit alone does not tell you whether attribution is right, so we report both. A Meridian-based reference fit for the finance-facing artifact is on the roadmap, not shipping today.

~3 business days turnaround
Step 03

We walk you through the results

A 30-minute screen-share readout of the channel-by-channel response curves, the optimizer's recommendation, and the one-page brief you can take straight to your CFO.

Plus a follow-up readout each quarter
Method borrowed from
Meta Robyn
&
Google Meridian
Their methods, reimplemented — not the libraries themselves
Hill saturation 26-wk adstock RSSD spend-effect penalty Bootstrap intervals
Methodology

Two open-source methods, one honest implementation.

Compass does not run Meridian or Robyn. It implements their published methods directly, and the honest lineage is mostly Robyn: a regularized regression over Hill-saturated, adstocked media, with hyperparameters chosen by derivative-free search against a composite loss — holdout error plus Robyn's RSSD term, which penalises solutions where a channel's share of effect drifts far from its share of spend. Uncertainty comes from a moving-block residual bootstrap, not a sampler. From Meridian we take two ideas rather than the machinery: ROI priors, so external experiment results can anchor a channel instead of the fit inventing its effect, and a knotted trend baseline that absorbs slow-moving demand without eating media credit.

We are specific about this because the distinction is the first thing a competent analytics lead will test. Robyn is frequentist and Compass is too; Meridian is Bayesian and Compass is not, so nothing here is a posterior and nothing here is a credible interval. What you gain is that every number is reproducible and auditable in seconds rather than a sampler run you have to take on faith. A blessed Meridian reference fit, for the cases that genuinely need a Bayesian treatment, is on the roadmap and is not shipping today.

The model is tuned for the workloads that legacy MMM tools miss. B2B SaaS channels including paid search, LinkedIn, field events, podcast, content syndication, and SDR outbound. A 26-week adstock memory calibrated to real sales-cycle lag. Pipeline-created as the primary KPI instead of closed-won revenue, so you can act on the model before the deal closes six months later. Same underlying rigor when we run the model on D2C accounts, with the channel registry and adstock retuned for shorter conversion cycles.

How to work with us

Two ways to engage.

Compass is a founder-led platform in its early customer phase. Two engagement types are open right now. Both include a fully custom Compass instance tuned to your funnel, channels, and review criteria. The right fit depends on how much you want to shape the product versus how much you want a turnkey delivery.

Design Partner Program (limited slots)

Reduced-price engagement for teams who want to help shape Compass. Best fit for marketing leaders who have opinions on how MMM and AI agents should work in their industry.

  • Fully custom Compass instance built in 30 to 45 days
  • Founder-led delivery with monthly working sessions for the first 90 days post-launch
  • Direct input into the product roadmap
  • Reduced pricing in exchange for agreed case study rights and referenceability upon successful outcome

Command Center Build (open enrollment)

Standard commercial engagement. Best fit for teams who want a scoped, priced, dated deliverable without a longer product-partnership commitment.

  • Fully custom Compass instance built in 30 to 45 days
  • Fixed-price scope agreed upfront, no billable-hour surprises
  • 30 days of post-delivery support included
  • Optional maintenance retainer available thereafter for ongoing model refits and agent iterations
FAQ

What B2B marketing leaders ask first.

Do we have enough data for an MMM to work?

For Compass we recommend at least 18 months, ideally 24, of weekly data with reasonably consistent channel mix. That's roughly what a Bayesian MMM needs to identify Hill curves and adstock decays separately for each channel. If your channel mix changed dramatically inside that window (a brand new motion, a paused channel), we'll surface that as identifiability uncertainty in the response curves rather than pretend the model is more confident than it is.

How is Compass different from just running Google Meridian ourselves?

You can. Meridian is open source and well documented, and if you have a data scientist who wants to own the pipeline, that's a valid path. Compass exists because most growth teams don't have that person, and the ones who do would rather have them working on modeling questions than on data prep, refit orchestration, response-curve rendering, and finance-facing storytelling. Compass handles the parts around the model: data validation, refit orchestration, response-curve rendering, the budget optimiser, and the finance-facing artifact. It also does one thing a single Meridian run will not do for you — refit the same data under a dozen defensible specifications and show you which channel ROIs hold still and which ones move by an order of magnitude. That stability sweep is usually the most uncomfortable and most useful output. If you have the appetite to build all of that in-house, Meridian gets you a model. Compass gets you the workflow around it, and a straight answer about which numbers you can take to a CFO.

How is this different from Recast?

Recast is an excellent product, they pioneered modern MMM for digital-native brands. Their core focus is DTC and consumer brands, and their pricing reflects the enterprise contracts they sign. Compass is positioned for B2B SaaS specifically: pipeline-created as the KPI instead of revenue, 26-week adstock for long sales cycles, and field events / SDR outbound as first-class channels. If you're an omnichannel DTC brand, Recast is probably the better fit. If you're a Series B-D B2B SaaS company, we're built for you.

What about Prescient AI?

Same answer with a slightly different shape: Prescient's core focus is DTC and omnichannel consumer brands, and they're excellent at it. Their case studies, integrations, and pricing all reflect that ICP. Compass is wedged into B2B SaaS, a vertical no modern MMM platform is seriously chasing yet.

How do you handle our six-month sales cycle?

Two ways. First, the dependent variable is weekly pipeline value created, opportunities generated, not deals closed. Pipeline sits 1–4 weeks after a marketing touch instead of 3–9 months, which gives the model enough signal to fit. Second, adstock max_lag is set to 26 weeks (vs. 8 in a default DTC model), so a LinkedIn ad seen in March can still be credited for a deal that closes in September.

What does the input CSV need to look like?

One row per week, with columns for: week-starting-Monday, weekly pipeline value created (the KPI), weekly spend per paid channel, and a few simple controls (price index, holiday flag, competitor index). Most B2B SaaS companies can export this from Salesforce + their ad platforms in under two hours. The full schema is available on request, email contact@compass-mmm.com and we'll send it.

What's the pricing after the design partner program?

We're not publishing public pricing yet, partly because we want design partners to influence it, partly because the right price for a $50M ARR Series C is different from a $200M ARR Series D. For comparison, Recast and Prescient are typically in the high-five-figure to low-six-figure ARR range for B2B SaaS scale. We expect to come in materially below that for design partners and modestly below it for general availability, but the exact number depends on what we build in the next six months.

Stop guessing. Start defending.

Five design partner slots open for Q3 2026. Apply now to get a full model fit on your data and the brief you'll wish you'd had at the last board meeting.

Work with us