The IMPACT engine · AI agents + human agents
AI agents score the signal. Human agents call the move.
The proprietary platform that runs underneath every Diray plan. Machine learning models score millions of TV signals in real time. Four decades of operator judgment translate the score into the actual buy. Monday produces a scored read. Friday produces the move. No black-box reporting.
Built on four decades of media operating discipline
39yrs
Buying media
$2B+
Spend stewarded
+18%
Velocity lift
+11pts
Intent lift
The loop · the IMPACT engine in motion
Five moves. Every week. Every plan.
The IMPACT engine does not sit in a slide. It runs every week. No campaign launches into a black box. No campaign exits without a written read. The same loop runs on a $50k pilot or a $50M plan. Discipline is the deliverable.
Mon · Sense
Pull the weekly signal. Sales velocity, intent panel, exposure log. The market moved since Friday. The plan responds to what changed, not what was true last quarter.
Tue · Score
The IMPACT engine grades every line item against your business outcome: Retail Velocity, Purchase Intent, Category Share. One IMPACT Score per placement. Statistical significance flagged. The score is honest about what we know before the decision gets made.
Wed · Decide
Invest more. Hold. Reallocate. One of three. The decision goes in writing before the budget moves. The operator owns the call. Data informed it. Nobody hides.
Thu · Prove
Held-out control vs. exposed audience. The lift is real, or it isn’t. No “promising signal” language. The week’s claim is defensible on the record.
Fri · Improve
What goes into next week. What gets retired. What the team learned that the plan didn’t know on Monday. The loop is how the plan gets sharper, not just bigger.
Want to see the loop run on your category? The pilot is the cheapest way to find out if the discipline holds.
Pilot the Impact Engine →The honesty check, built in
For every product, the engine also runs a baseline that predicts outcomes from recent history alone, with zero media features. If the media model does not beat simple momentum, the weekly read says so. You will never get a report that assumes media worked.
Signal · three pipes in
What goes into the engine.
Diray doesn’t score against impressions. Three outcome signals feed the engine. Sales data is the ground truth. The other two triangulate.
Sales velocity
Units off shelf per store per week. The metric the buyer cares about. Pulled weekly from Walmart Connect, Kroger Precision, NCS, Circana.
Intent panel
Brand-lift and purchase-intent panel. Lifted consideration measured pre and post exposure against a held-out control. The leading indicator before the cart moves.
Exposure log
Server-side log of who saw what, when, on which screen. Reconciled against the audience panel. The exposure side of every lift claim is auditable, not modeled.
Signal to shelf · the loop, off the whiteboard
Software scores. Operators decide.
The Impact Engine reads millions of signals, but a person with four decades of pattern recognition signs every move. Follow one decision from the desk to the shelf.
Proof · what the method has done, with the range
Median outcome. Full range. Source on every line.
Three numbers, taken as the median across the last 12 pilots running the full loop. The range is shown so you see the floor as well as the ceiling. No cherry-pick.
N = 12 pilots, Q1 2024 – Q4 2025. Two pilots excluded for incomplete control panels. Full methodology available in the walkthrough.
Voice · operator-to-operator
Operators on what changed.
Three quote slots, every one carrying a named human, a named title, a named brand, and the pilot cycle the quote came from. If a brand cannot go on record, the slot stays empty until they can. We do not run anonymous testimonials.
“Diray is the only partner that hands us a written read every Friday. The discipline is the product.
[Operator name — pending release] VP, Growth · [DTC apparel] · Pilot cycle 04
“Our Tuesday read used to be a meeting. Now it’s a decision before 10am.
[Operator name — pending release] Head of Performance · [Specialty retail] · Q3 2025
“Five years in and Diray still ships a written log every Friday. That’s the moat.
[Client name — pending release] VP, Marketing · [CPG] · 5-year client
Offer · 30-day method pilot
One channel. One outcome. The full loop, once.
The fastest way to see whether the discipline holds in your category. Diray runs the full five-move loop on one channel for 30 days. Day 30 ends with a written readout and one decision: scale, hold, or reallocate.
Week 1 · Wire
Signal pipes connected. Held-out control defined. The one outcome and the threshold that defines “moved” agreed on paper.
Weeks 2–4 · Loop
The five moves run. Weekly scored read. Weekly decision. Three full cycles of the operating loop on live spend, on live surfaces.
Day 30 · Read
Written record with the three numbers and one decision. The pilot ends with a yes, a no, or a sharper scope. Not a deck.
Honest range · what the loop will not promise
Three things the loop does not do.
A method that promises everything earns no operators. These are the three places the loop will not commit, and the reason each one is off the page. If a vendor promised you the opposite, ask them how they measured it.
Run without a control.
No held-out audience, no honest lift number. If your category will not support a control cell, the loop stops at directional reads. We will tell you that on day one, in writing.
Pull a stat-sig read in week one.
Most pilots hit directional reads in week three and stat-sig reads in week six. Anyone promising significance on day seven is either spending wrong or measuring wrong.
Replace the platforms you already run.
The loop runs on top of MNTN, Tatari, Vibe, your DSP, your MMM. We do not displace them. If your existing stack is working, the loop sharpens it. If it is not, the loop tells you so honestly.
FAQ · what operators ask first
The questions before the pilot.
How is this different from MNTN, Tatari, or Vibe?
Those are buying platforms. Diray is the operating loop that runs on top of whatever platform you use. We don’t sell inventory. We sell the written weekly read, the decision in writing, and the held-out lift test. Platforms optimize. The loop decides.
Do you replace our current measurement vendor or work with it?
Work with it. Your MMM, your incrementality vendor, your panel partner. They feed the scoring layer. The loop’s job is to turn their outputs into a weekly decision instead of a quarterly slide. Most engagements keep the existing stack and add discipline on top.
What if our category doesn’t have retail sales data?
Sales velocity is the preferred ground truth. When it isn’t available, the loop runs on the closest defensible outcome you do have. Pipeline, qualified leads, app installs, paid trials. The framework is the same. The signal swaps in.
How long until the loop produces a stat-sig read?
Depends on spend, audience size, and lift magnitude. Most pilots produce a directional read by week three and a stat-sig read by week six. The 30-day pilot is scoped to give an honest answer about whether the signal is strong enough to commit.
How the engine comes to life · six IMPACT products
The engine ships as six products.
The loop runs underneath all six. Each product is the IMPACT engine pointed at a specific outcome. Pick by the move the brand needs to make next.
IMPACT shelf
Drive retail sales through the full funnel. TV to retail media networks to purchase. Measured by closed-loop sales lift.
IMPACT demand
Capture rising category demand. Share of active demand reads intent in real time and predicts where share moves next.
IMPACT market
Build brand awareness at scale. Mass reach through linear and streaming. Measured by brand lift and category share.
IMPACT launch
Take a new product from zero to market. Build the strategy from scratch, prove viability, scale through retail.
IMPACT efficiency
Make your existing budget perform harder. Optimize across networks, dayparts, and platforms to stretch every dollar.
IMPACT surge
Activate on weather, sports, and cultural moments. Media that triggers the instant conditions align.
Next step
Scope the pilot.
45 minutes with Diray operators. You bring the channel and the outcome. We bring the loop. We leave with a 30-day pilot scoped on paper.