Your Enterprise Customers Are Re-Shopping You Every Six Months

Venture firm Madrona surveyed 150 enterprise IT professionals this summer. Two findings got the headlines: 74% plan to expand their AI budgets over the next twelve months, and the rest plan to hold steady. IDC expects about $4.25 trillion in technology spending in 2026, almost all growth driven by AI. That is the good news, and founders have been repeating it for weeks.

The finding further down is the one that should change how you plan. Some 77% of the enterprises surveyed reevaluate their AI vendors every six months or on a rolling basis. Madrona calls it a "fast in, fast out" dynamic, fundamentally different from traditional enterprise SaaS, where multi-year contracts provided a moat of inertia.

Read that again with your pipeline in mind. The logo you signed in March is being re-shopped in September, not because you did anything wrong, but because that is now the buying process.

The old enterprise motion had a quiet gift built into it. A buyer took nine months to decide, signed a three-year agreement, wired the software into six other systems, trained 400 people on it, and then stopped thinking about it. Your product did not have to keep winning. Inertia won for you.

That inertia was never a line item, but it was the most valuable asset most SaaS companies owned. It is what let founders forecast with confidence.

The AI buying cycle removed it from both ends. Procurement got faster, which is why so many startups have posted revenue curves that would have been impossible in 2019. TechCrunch reported earlier this year that more companies are hitting $10 million in ARR within three months of launch than at any previous point. The same speed that let you in lets the next vendor in. Switching costs are lower, the models underneath are increasingly interchangeable, and the re-evaluation cadence, in Madrona's words, is relentless.

You did not get a faster sales cycle. You traded a slow one with a moat for a fast one without.

The same Madrona research found that fewer than half of enterprise AI pilots ever make it into full production. Depending on your mood, that is either grim or an improvement. MIT's widely circulated 2025 study reported that 95% of enterprise AI projects had failed to deliver measurable ROI, so moving from a 5% success rate to something closer to 45% is real progress for the category.

It is not progress for your forecast, though. The pilot-to-production conversion used to be the finish line. You survived the bake-off, you got budget, you got a signature, and the revenue was yours to lose slowly.

Now production is a checkpoint. For the first time, enterprise revenue stays insecure after the product graduates from pilot and gets adopted company-wide. A customer can be live, happy, and quietly running a comparison against two competitors in the same quarter. If your board deck treats "moved to production" as the point where an account becomes durable, it is describing a market that stopped existing.

Part of the churn risk is self-inflicted, and it lives in your pricing page.

New research from Andreessen Horowitz surveyed 50 technical AI buyers and found that more than half want fees tied to the work produced or another outcome, rather than to usage like tokens consumed. Partners Tugce Erten and Sarah Wang argue that pricing "around the recognizable work" is what proves your worth to the customer.

Per-seat and per-token pricing are both inherited from the SaaS era, and they share a flaw here. Once a company knows it needs email or HR software, the only question is how many employees. The value is assumed. With AI tooling, the value is exactly what is under review every six months, and a token meter tells the buyer nothing about whether you delivered any.

Price around tickets closed, reports processed, leads qualified, or documents reviewed, and the invoice becomes the ROI report. When the six-month review comes around, the person defending your line item holds a number that maps to their own goals rather than a consumption chart they have to interpret.

That is not a growth hack. It changes who has to do the work of justifying you.

Revenue that renews by default and revenue that gets re-won every two quarters should not carry the same name on your dashboard. Track these four separately.

Gross revenue retention, not net. Net retention lets expansion from your three happiest accounts hide churn everywhere else. Gross retention is the floor.

Contracted months remaining, weighted by ARR. If the weighted average is under nine months, you do not have annual recurring revenue. You have a subscription business with a quarterly renewal cycle, and you should plan cash accordingly.

Depth of use per account. Count the distinct teams, workflows, and systems your product touches inside each customer. One workflow in one team is a pilot with a longer invoice. Four workflows across three teams is something a competitor has to dislodge rather than beat on a demo.

Time to first defensible outcome. Days from signature until the customer can point at a result they would lose by switching. Anything over 60 days in a market with a six-month review cadence means you run out of clock before you have proven anything.

Founders doing this the first time often find it useful to build the model out loud with a tool like Foundra rather than reverse-engineering it from a board deck template, because these four numbers interact and the interaction is where the surprises are.

Inertia is gone as a free gift. It is still available as work. Four things create it.

Own the data record, not just the inference. If the customer's history, corrections, exceptions, and labels accumulate inside your product, leaving means starting over. If you are a thin layer over a model the buyer can also call directly, leaving means changing an API key.

Get written into the workflow, not the tool list. A product a team opens is replaceable. A product that receives a webhook from the ticketing system, writes back to the CRM, and posts into the channel where decisions happen is a migration project.

Build the review into your own calendar. If the enterprise reevaluates you every six months, run that review yourself at month four with the numbers already assembled. The vendor who arrives with an ROI summary before procurement asks for one is rarely the vendor who gets replaced.

Multi-thread past your champion. Champions move roles constantly. Every account needs three people who can describe, unprompted, what breaks if you disappear.

Expect the diligence question to change this cycle. It is moving from "how fast did you grow" to "how much of that growth survives a vendor review."

Answer it before it is asked. Show gross retention alongside net. Show weighted contract length. Show what share of ARR sits in accounts with three or more active workflows. Name the accounts that came up for review in the last two quarters and say which you kept, which you lost, and what the losses had in common.

That last part feels risky and it is the part that lands. Every investor looking at AI-native companies is trying to work out which revenue is durable and which is a long trial. A founder who has already done that separation understands their own business.

You can do this without new tooling.

Week one: pull every account and record contract end date, months remaining, gross retention over four quarters, and the number of distinct workflows in use. Sort by ARR at risk in the next two quarters.

Week two: call the ten largest. Do not pitch. Ask three questions. How would you describe what we do to your CFO? When does your team next review tools in this category? What would have to be true for us to be an obvious keep?

Week three: rewrite pricing for your next ten deals around a unit of work the buyer already reports on. You do not need to reprice the existing base to learn whether the new frame closes faster.

Week four: ship the highest-value integration you have been postponing into your top five accounts. Depth of use is the only defense that compounds while you sleep.

None of this is exciting. It decides whether the ARR number in your deck next year is a floor or a high-water mark.

Does this apply if we sell to small businesses rather than enterprises? The Madrona research covered enterprise IT buyers, so the six-month cadence is an enterprise finding. The transferable part is the principle: assume the purchase decision gets remade, and build depth of use accordingly.

Should we move to outcome-based pricing right now? Not across your whole book at once. Outcome pricing requires that you and the customer agree on how the outcome is measured, and getting that wrong creates disputes worse than churn. Test it on new deals, in one segment.

Is a six-month re-evaluation the same as churn risk? No, and treating them as the same causes bad decisions. A review is a chance to expand as well as to lose. The risk is not being reviewed, it is being reviewed by people who cannot articulate what you delivered.

How do we forecast when contracts are this short? Forecast in two books. Book one is revenue under contract for more than nine months with multiple active workflows. Book two is everything else, discounted by your observed retention on similar accounts. Run cash planning off book one only.

What if enterprises go back to multi-year commitments? That would change the urgency, not the work. Whether buying habits revert is an open question, and depth of use plus outcome-linked pricing wins in either version of the market.

1. Startup ARR is less secure than ever, new research shows (TechCrunch, September 3, 2026)
2. Madrona: Harnessing Enterprise Value, the ROI of AI (research report, August 2026)
3. a16z: You Are Not a Model, Don’t Price Per Token
4. MIT NANDA: The State of AI in Business 2025 report
5. IDC: Worldwide technology spending forecast
6. More startups are hitting $10M ARR in 3 months than ever before (TechCrunch)