PerspectiveEcosystem Mechanics

Microsoft Now Grades What You Share. The Grade Is Set Upstream.

QRP is retired, automated quality checks run at submission, and every referral now carries an ML-computed confidence score that Microsoft sellers triage by. Every input the model grades — customer need, deal value, close date — is determined before the referral is shared. Referral tooling moves the artifact; it cannot improve what the artifact says.

Ravi Paladugu10 min readSeptember 3, 2026Alliances Directors, Azure Partners, ISV Leaders

TL;DR

Microsoft retired QRP in March 2026, added automated quality checks at referral submission in May, and now assigns every outbound referral an ML-computed Referral Confidence Score (High/Medium/Low) that Microsoft sellers use to prioritize. The score's inputs — customer need, deal value, close date, solution play — are set by upstream offering and research work, before the referral is shared.

Definition — Referral Confidence Score

A per-referral quality grade in Microsoft Partner Center. Per Microsoft's documentation (July 2026): “All outbound opportunities will have confidence score as High, Medium and Low. This feature is to enhance the quality of opportunity creation with guidelines from ML model.” Sellers use it to prioritize opportunities; it is not exposed back through the Referrals API.

There is a sentence in Microsoft’s Partner Center documentation that most alliance teams haven’t read, and it changes what a referral is. In the co-sell opportunities documentation, updated July 2026, Microsoft states: “All outbound opportunities will have confidence score as High, Medium and Low. This feature is to enhance the quality of opportunity creation with guidelines from ML model.”

Read it twice. Every referral you share with Microsoft now carries a grade, the grade comes from a machine-learning model, and Microsoft sellers see it. The referral stopped being an administrative artifact the moment that shipped. It became a scored submission — and the score is computed from work that happened, or didn’t happen, long before anyone opened Partner Center.

That is the argument of this piece: Microsoft has moved from collecting partner referrals to grading them, and every input the model grades is determined upstream of the share. Referral tooling can move the artifact between systems. It cannot improve what the artifact says.

What happened to QRP?

The Qualified Referral Program is gone. Microsoft’s March 2026 Partner Center announcement put it plainly: “QRP is retired at the end of March 2026 and replaced by a unified co-sell experience centered on Partner Center referrals. This change reduces duplicate workflows, improves pipeline transparency for both partners and Microsoft sellers, and accelerates response times.” Shared referrals now arrive in the Referrals workspace’s Leads tab as SMB opportunities, QRP’s incentives folded into the standard frameworks, and historical QRP data entered a read-only state for 90 days to allow exports.

Retirements like this are usually read as plumbing changes — a program consolidated, a workflow renamed. This one is worth reading differently, because of what Microsoft shipped in the two months that followed. QRP’s retirement removed the last separate lane for partner referrals. Everything now flows through one pipe — and then Microsoft put a quality gate on the pipe.

Why do co-sell referrals get held or declined?

In May 2026, Microsoft turned on automated quality checks at submission: “Complete, high-quality referrals are automatically accepted and routed to the appropriate seller.” And the other side of that sentence: “If any required information is missing, the referral might be held for follow-up, or declined with the option to resubmit once updated.” The required information is specific: solution area and play, estimated deal value, estimated close date, customer need, customer contact.

Look at that field list as a GTM person rather than an administrator. A credible estimated close date exists because someone had a real conversation with the buyer about their timeline. A customer need that reads as genuine exists because someone researched the account and can articulate why this company, now. A defensible deal value exists because the offering is scoped tightly enough to price. None of these can be produced at the form. They are records of work already done.

The referral form doesn’t ask for information. It asks for evidence — and evidence can’t be back-filled at the moment of submission.

What is the Referral Confidence Score?

The score is the grading layer on top of those checks. Microsoft’s documentation describes it from both sides of the table. For the partner submitting: partners “receive instant recommendations to strengthen their submissions… our system highlights what’s missing and suggests improvements on the spot.” For the Microsoft seller receiving: “Every referral comes with a score—think of it as a quality snapshot… helps sellers prioritize the best opportunities and boosts overall acceptance rates.”

That second quote is the one that should change how your team spends its week. Microsoft sellers triage by the score. A High-confidence referral and a Low-confidence referral are not two entries in the same queue — they are, functionally, two different products arriving on a seller’s desk, and the seller was just told which one to pick up first.

We’ve published the verified mechanics of the score — the High/Medium/Low bands, the required fields, what the API does and doesn’t expose — at the Referral Confidence Score explainer. The short version that matters here: the score is not readable back through the Referrals API. You cannot instrument it, dashboard it, or optimize it after the fact. You can only influence what it is computed from.

What do Microsoft sellers see when you share a referral?

Picture the referral from the seller’s side, because that is where its fate is decided. Say your company sells a compliance automation product, and the referral describes a 150-person healthcare provider — a mid-market buyer with an audit deadline two quarters out, named in the customer-need field, with a close date that matches the audit calendar and a deal value that matches the scope you actually proposed. The seller sees a High-confidence referral whose story hangs together, auto-accepted and routed — and a reason to engage this week.

Now the same referral built the other way: an ICP-filtered company name, a customer need pasted from your solution description, a close date that is a guess wearing a date format. The checks may hold it for follow-up. If it passes, it arrives graded — and the grade tells the seller what the submission couldn’t hide: nobody has done the work yet. The referral isn’t evidence of a deal. It is a request that Microsoft go find one.

A complete referral gets routed. A strong referral gets worked. The difference is everything the form can’t see — and everything the model now scores.

“Isn’t this just a forms problem?”

The tempting read of all this: fill in every required field and the referral auto-accepts, so buy better referral plumbing and move on. It’s tempting because half of it is true. Complete referrals do auto-accept — Microsoft says so. But acceptance was never the scarce resource. Seller attention is. And Microsoft’s own language separates the two: completeness gets you “automatically accepted and routed,” while the score exists so sellers can “prioritize the best opportunities.” Routing is not engagement.

The score’s inputs are exactly the ones a form cannot manufacture. Whether the customer need is specific to this buyer. Whether the deal value and close date are credible together. Whether the solution play matches what the account is actually doing. Those are functions of the offering you built, the research you ran, and the conversation you had — upstream, before the referral existed. Tooling that moves referrals between your CRM and Partner Center faster is moving the same evidence faster. If the evidence is thin, you have automated the production of Low-confidence referrals.

FY27 raises the stakes on the same bet

Everything above would matter on its own. Microsoft’s FY27 direction makes it compound. At the FY27 MCAPS Start for Partners, the priorities were AI skilling for what Microsoft calls the agentic AI era, go-to-market acceleration, broadened co-sell and incentives, and enhanced investment in Microsoft Marketplace. Mira Ayad, General Manager of Global Marketplace, put Microsoft’s own number on the momentum in a July 2026 partner blog: Marketplace “sales doubling year-over-year, for the third consecutive year” — Microsoft’s claim, worth reading in its original context. Her framing: “In an AI-driven economy, Marketplace is becoming the commercial engine that connects innovation to customer value.”

And in September 2026, Microsoft is launching Frontier Accelerate for Marketplace — “a new unified offering… that brings together go-to-market benefits from ISV Success, Marketplace Rewards, Azure IP co-sell, and Solutions Partner with certified software designations into a single path for Marketplace growth.” The pattern across all of it is one pattern: fewer, more unified paths to Microsoft’s field and its customers — each with a quality mechanism at the gate. The partners who benefit from consolidation are the ones whose submissions grade well once everything flows through the graded pipe.

The grade is set upstream

Here is the structural point, stated without the vendor fog. The confidence score makes upstream GTM work measurable for the first time — by Microsoft, on every referral, in a number a seller acts on. The offering that is scoped tightly enough to price. The research that finds the mid-market logistics company actually consolidating onto Teams — not every company that fits a filter. The written record of why this buyer, now, that becomes a customer-need field a seller believes. That work used to be invisible until win rates revealed it quarters later. Now it is graded at the moment of submission.

Microsoft built the grading machine. The grade it assigns was always being earned — or forfeited — before the referral existed.

This is the layer Wyra operates. Lex, Wyra’s agent for the Microsoft ecosystem, builds the upstream record the score is computed from: offerings grounded in your Microsoft practice, prospects qualified in writing — or disqualified with the reason written out — and the why-now on every account captured before outreach begins. The deals your team chooses to share sync two-way with the Referrals workspace from your CRM, so what Microsoft sellers see is the evidence, current, rather than a re-keyed copy. And we run this motion on ourselves: Wyra is listed on Microsoft Marketplace and enrolled in ISV Success — the same graded pipe our customers share into.

The referral was never the work. It was always the receipt. Microsoft just started reading the receipts.

Frequently asked questions

What happened to Microsoft's Qualified Referral Program (QRP)?

Microsoft retired QRP at the end of March 2026 and replaced it with a unified co-sell experience centered on Partner Center referrals. Shared referrals now arrive in the Referrals workspace's Leads tab as SMB opportunities, QRP incentives were folded into standard frameworks, and historical QRP data entered a read-only state for 90 days to allow exports. Every partner referral now flows through one pipe — the one Microsoft added quality checks and a confidence score to in the months that followed.

What is the Referral Confidence Score in Partner Center?

An ML-computed grade — High, Medium, or Low — attached to every outbound co-sell referral. Partners get instant recommendations to strengthen submissions before sharing; Microsoft sellers see the score and, in Microsoft's own words, use it to prioritize the best opportunities. The score is not readable back through the Referrals API, so it cannot be dashboarded after the fact — only influenced upstream, by the quality of the customer need, deal value, close date, and solution play the referral carries.

Why do Microsoft co-sell referrals get held or declined?

Since May 2026, automated checks run at submission. Complete, high-quality referrals are automatically accepted and routed to the appropriate seller; if required information is missing — solution area and play, estimated deal value, estimated close date, customer need, customer contact — the referral can be held for follow-up or declined with the option to resubmit. The fields that trip submissions are records of work: a credible close date comes from a real buyer conversation, a genuine customer need from actual account research.

What do Microsoft sellers see when a partner shares a referral?

The referral's content plus its confidence score — and they triage by the score. A High-confidence referral whose customer need, deal value, and close date tell one coherent story about a specific buyer gets auto-accepted, routed, and worked. A complete-but-thin referral also routes, but arrives graded, and the grade tells the seller the upstream work hasn't been done. Completeness gets a referral routed; strength gets it worked.

Key Takeaways

QRP retired at the end of March 2026; all partner referrals now flow through one Partner Center pipe — and in May 2026 Microsoft put automated quality checks on that pipe, auto-accepting complete referrals and holding or declining incomplete ones.
Every outbound referral carries a Referral Confidence Score (High/Medium/Low) from an ML model. Microsoft sellers use it to prioritize — routing is automatic, engagement is graded.
The score’s inputs — credible customer need, deal value, close date, solution play — are records of upstream work: offering clarity, prospect research, real buyer conversations. They cannot be produced at the form.
FY27 compounds the shift: consolidated paths to Microsoft’s field (Frontier Accelerate for Marketplace, broadened co-sell) each carry a quality mechanism at the gate. The partners who benefit are the ones whose submissions grade well.

Book a Demo

Related Resources

View all

PerspectiveEcosystem Mechanics

Azure Partner Center Is Not Partner Enablement. It’s Infrastructure.

Every Director of Alliances at an Azure partner knows Partner Center well. Most also know it isn’t where their GTM happens. This piece names that gap: Partner Center is operational infrastructure — and the intelligence layer partner-led growth requires is built from the partner’s vantage point, not the portal’s.

10 min readApril 9, 2026

PerspectiveCategory Definition

The Co-Sell Stack, Explained

Every product in the co-sell market operates at one of four layers, and each is defined by one question: what does this software assume already exists when you start using it? Three layers are well served. The fourth — where opportunities come from — is almost entirely unoccupied.

10 min readAugust 7, 2026