What Series A AI founders should bring to a Q3 raise this quarter

The problem this solves

Two things happened in two days that hardened the Series A AI proof bar, and most founders haven’t noticed yet.

On May 5, CopilotKit closed a $27M Series A from Glilot Capital, NFX, and SignalFire on the back of one specific story. TechCrunch’s announcement that day named Deutsche Telekom, Docusign, Cisco, and S&P Global running their protocol in production, with millions of installs per week and a large portion of the Fortune 500 already on it. The Series A check followed because the production proof was already in the protocol’s adoption curve, not stitched together from pilot-deck slides.

On the same day, Sierra raised $950M at a $15B valuation and walked the world through what that proof looks like at full enterprise scale.

Twenty-four hours apart, two announcements told every Series A AI founder going into a Q3 raise the same thing. Pilot logos and total ARR are not going to clear the bar this quarter. One named institutional customer in real production, with a measurable reliability or deployment-speed metric attached, will.

The approach

A Q3 deck that reads “27 logos in trial, $4.8M ARR, 6-month sales cycle” is a 2024 deck heading into a 2026 fundraise. Here is what to bring instead.

Five moves, in order. None require Sierra’s scale. They require picking one customer and one number and doing the work.

1. Name one customer in production, not in trial.

   Series A investors are now scoring "in production" as a separate category from "in trial." That means one customer running the product on a real workflow, with their own people, their own data, their own budget approving the spend. If that customer exists, name them. If not, your job between now and Q3 is to convert one.

2. Attach one measurable outcome to that customer.

   Sierra's deck quotes 8 weeks for Cigna with an 80% drop in patient authentication time. A Series A version does not need that scale. It needs to be specific. Time saved per workflow, error rate reduction, deployment speed, resolution rate. One number, one customer, both verifiable on a reference call.

3. Write a one-pager on the agent-incident question.

   A LiveScience report from April 29 described an AI coding agent that deleted PocketOS's production database and three months of backups in nine seconds. Investors reading that headline are now asking Series A founders at due diligence what happens in the first nine seconds if an agent goes wrong. The answer needs to fit on a single page. Agent inventory, scoped tokens at every connection, kill switch, backup recovery time.

4. Define deployment speed as a number, not a marketing claim.

   Sierra is shipping Nordstrom in five weeks and Cigna in eight. A Series A startup with a five-month onboarding cycle is competing on a different procurement clock than the company setting the market benchmark this quarter. Pick a number, commit to it, track it across customers, bring the data to the next pitch.

5. Pick one of three valid Series A archetypes.

   This week showed three valid shapes. Named-enterprise production proof (CopilotKit's $27M behind Deutsche Telekom, Docusign, Cisco, S&P Global). Elite technical credibility with a future-capability bet (Standard Intelligence's $75M Series A from Sequoia and Spark Capital, with Andrej Karpathy as angel). Vertical workflow tied to a real customer (Nova Intelligence's $31.5M for SAP S/4HANA migration). Pick one shape and tell that story all the way through. Investors will pass on a deck that tries to be all three at once.

Why most teams get this wrong

The single mistake I keep seeing at Series A is founders treating their pilot count as their proof.

It feels logical. More logos look like more momentum. But a pilot count does not survive an investor reference call. The first question on that call is “how is this used day to day,” and the second is “what would happen if it got turned off tomorrow.” If the answer to either is fuzzy, the round prices very differently.

There is a piece by Mean CEO from May 4 that named this in plain language. Investors want evidence that the product works inside a repeated workflow with usage frequency, retention, customer urgency, and trust features. Demos do not carry that weight anymore. Workflow does.

The other mistake is bigger. It is bringing a deck full of metrics that do not belong to the company. Total addressable market. Industry growth rate. Foundation model release timeline. None of that proves anyone can run in production. Investors do not need a TAM lecture this quarter. They already know the AI market is big. They need to see one real customer surviving contact with a real product.

The numbers

A few specific reference points worth bringing to prep. None of them require matching Sierra’s scale. They give the shape of what investors are now anchoring on.

Sierra reached $150M ARR in eight quarters since launch and serves 40% of the Fortune 50. Nordstrom’s voice agent went live in five weeks. Cigna deployed in eight weeks and cut patient authentication time by 80%. Singtel went live in ten weeks with resolution rates above 70%. CMSWire’s coverage on May 6 noted that organizations using autonomous AI systems reported a 28% improvement in issue resolution time and a 19% lift in first-contact resolution rates.

Those are the operational numbers. Read them once, then think about what one row of the same table would look like for one of your customers. Not TAM. Not team headcount. One row: customer name, deployment time, measurable outcome, time-to-value. A founder who can fill in that row honestly with a real customer walks into Q3 with the strongest version of the Series A AI deck.

CopilotKit’s row reads as four named Fortune 500s plus AG-UI protocol adoption by Google, Microsoft, Amazon, and Oracle. Millions of installs per week. A large portion of the Fortune 500 in production. That is a row.

Ship it

If you’re raising in Q3, here is what I would do this week.

Pick one customer. The one closest to “in production” with a real budget and a real workflow. Get on a call with them by Friday. Find one number both sides can look at honestly. That number is what the Series A deck gets built around.

Then write the agent-incident page. One page. What happens in the first nine seconds, the first hour, the first day. What gets revoked. What gets restored. Who gets the call. Bret Taylor told CMSWire this week that Sierra had digitised the last remaining analogue channel, the telephone line. He did not say it casually. He said it because shipping into a customer’s primary revenue channel is the bar Series E money is now priced on.

A Series A founder does not need to be Bret Taylor. The need is for one customer, one number, one incident-response page, and a clear archetype. Bring those four things, and a Q3 raise becomes a meeting about specific evidence. Bring a logo wall instead, and the same meeting becomes a meeting about why the round is hard.

That is the change. The good news is the work is doable in five weeks. So is shipping at Cigna.