The Mars Climate Orbiter Lesson Every GTM Team Still Hasn't Learned

In 1999, NASA lost the Mars Climate Orbiter because one team worked in metric units and another in imperial, and no layer in between caught it. Both teams were right. Both did their jobs. The spacecraft still burned up, because "right" in two incompatible systems does not add up to a working whole. Your GTM stack runs on the exact same flaw. Marketing's MQL is not sales' MQL. RevSure's cohorted conversion is not the CRM's snapshot. Every quarter, your numbers fail to reconcile, and nobody catches it until it is too late to act.

The consensus reading of that NASA failure is that someone made a mistake. Tighten the process, add a checklist, and it will not happen again. That reading is comforting and wrong. Nobody fumbled a calculation. Two teams each computed correctly inside their own definition, and the system had no shared layer where those definitions had to agree. The failure was structural. The same structure is sitting in your revenue stack right now.

Both teams are right. That is the problem.

Here is what makes this failure mode so hard to see: there is no villain. In the Mars case, the navigation team and the engineering team were each internally consistent. The error lived in the seam between them.

GTM runs on seams like that. Consider the definition of a single metric. In the case files, one team's internal MEL-to-MQL conversion runs 75 to 80 percent. RevSure's cohorted view of the same funnel reads 27 to 30 percent. Neither is wrong. A cohort that follows the same leads over time is measuring something genuinely different from a snapshot that overwrites itself. But put the two numbers in a board deck side by side and someone has to explain a 50-point gap, and the honest explanation, "they are different units," sounds like an excuse.

It gets more concrete. One team defines an MQL by campaign-member first-responded date. Another routes essentially everyone through as an MQL unless they are flat-out junk or in an unsupported country. A third abandoned scoring entirely for a hot-warm-lava model because nobody had confidence in how the scores added up. Three teams, three definitions of the same word, all feeding the same pipeline. This is metric, not arithmetic, but the failure is identical to the orbiter: locally correct, globally incoherent.

The 24-hour version

The orbiter failed once, catastrophically. GTM fails this way continuously, in smaller increments, all the time.

Last-touch attribution loses ROI visibility every time a cookie resets on a 24-hour cycle. Salesforce, as one operator described it, is just a snapshot that overwrites its own history, so the record of what an account looked like last month may simply be gone. A spend number lives in one team's spreadsheet using a naming convention that does not tie back to the CRM. Each of these is a local unit mismatch. Each one is small. Together they mean that the answer to "what is performing this quarter" is, as one CMO put it, always a post-mortem, known only after it is too late to change the outcome.

That is the Mars lesson at quarterly cadence. Not one dramatic loss, but a steady leak of coherence, paid for in decisions made on numbers that never reconciled.

Now hand the controls to AI agents

Here is why this stops being a tolerable annoyance and becomes an existential one.

For a hundred years, humans have absorbed these mismatches manually. Someone notices the board number, looks off, walks down the hall, and reconciles two definitions by hand before the meeting. Slow, expensive, but it works as a safety layer. The orbiter had no such layer, which is why it was lost.

Your agentic stack will not have one either, unless you build it. An SDR agent reading one definition of a qualified lead, a forecasting agent reading another, a campaign agent optimizing to a third: each correct in its own units, none reconciled. The AI agents will not pause at the seam the way a human does. They will act, at machine speed, and write their incompatible answers back for each other to consume. By 2030 the vision memo projects over 100 agents across GTM. A hundred Mars orbiters, each navigating in its own units, all flying the same mission.

AI is only as good as its context. Garbage in, garbage out, and nowhere is it felt more than in GTM. The orbiter is what garbage-in looks like when the actor is fast, confident, and given no shared frame of reference.

The lesson is the layer

The fix for the Mars Climate Orbiter was never a better navigation team. Both teams were already excellent. The fix was a layer that forced the units to agree before anything acted on them.

That is the lesson GTM still has not learned. The answer is not a better attribution tool or a smarter forecasting model competing to produce its own number. It is a shared context layer underneath all of them, where an MQL is resolved to one definition, an account to one entity, a quarter to one consistent frame, before any human or agent reads it. The case files show what that looks like when it exists: counts that reconcile to within 0.5 percent of the system of record, and disagreements that become findings instead of fights.

Two right answers in two unit systems crash the spacecraft. One shared frame of reference flies the mission. GTM has spent decades treating its reconciliation problem as a series of individual mistakes to be cleaned up after the fact. It was always the missing layer. Build the layer before you put a hundred agents at the controls.

Frequently asked questions

What is the Mars Climate Orbiter lesson for business systems?

The lesson is that two internally correct systems with incompatible definitions produce a broken whole, with no single person at fault. NASA lost the orbiter because one team used metric units and another imperial, with no layer reconciling them. Any business system fails the same way when separate teams define the same metric differently and nothing forces those definitions to agree before action is taken.

Why do two GTM teams get different numbers for the same metric?

Because they are measuring in different units without realizing it. A cohorted conversion rate that follows the same leads over time is genuinely different from a snapshot that overwrites itself, which is how one funnel can read 27 to 30 percent in one view and 75 to 80 percent in another. Both can be correct inside their own definition.

How do AI agents make data mismatch worse?

Humans absorb unit mismatches by reconciling them manually before acting. AI agents do not pause at the seam. They act at machine speed on whatever definition they read and write the result back for other agents to consume. With over 100 agents projected across GTM by 2030, incompatible definitions propagate faster than any human can catch.

Is GTM data mismatch a data quality issue or a definitions problem?

It is a definition problem that looks like a data quality problem. The underlying records may each be accurate. The failure lives in the seam between systems that define a stage, an account, or a quarter differently. Cleaning individual records does not fix it; reconciling the definitions in a shared layer does.

What fixes inconsistent definitions across GTM systems?

A shared context layer that resolves definitions before anything acts on them, so an MQL means one thing, an account resolves to one entity, and a quarter is one consistent frame. In RevSure deployments, this is what lets counts reconcile to within 0.5 percent of the system of record, turning disagreements into findings rather than fights.

Can a better attribution or forecasting tool solve this?

No, because another tool just adds one more number competing with the others. The Mars fix was not a better navigation team; both were already excellent. It was a layer forcing the units to agree first. GTM needs the same: one context layer underneath the tools, not another tool alongside them.

‍