Executive Case Study

The first 90 days of a revenue operations transformation

A fictional but entirely recognizable situation: $75M ARR, a CRM nobody trusts, 25% forecast variance, comp disputes every quarter, renewals in spreadsheets, 14 GTM tools, a CRO who wants a forecasting platform and a CEO who wants AI. Thirty-seven things are broken. This is how I choose the five that matter — and what I deliberately refuse to do first.

The situation

ARR

$75M

Growing 28%, decelerating

Forecast variance

25%

Quarter to quarter, both directions

GTM tools

Four overlap materially

Sources of ARR truth

CRM, billing, and a finance workbook

What people tell me in week one

CRO: "The forecast is wrong every quarter. We need a forecasting tool."
CEO: "Every company our size is using AI. What's our plan?"
CFO: "I can't tie your ARR number to my ARR number, so I use mine."
VP Marketing: "We hit MQL target. Sales says the leads are garbage."
VP Sales: "My reps spend Friday afternoons fixing Salesforce instead of selling."
Top AE: "My commission has been wrong three quarters running."

What's actually true underneath

No shared definition of a qualified opportunity, so the funnel can't be measured end to end.
Stage criteria are subjective, so stage-based probability is fiction — which is why the forecast misses.
Three ARR definitions exist because nobody ever adjudicated one.
Renewals in spreadsheets means churn is discovered, not forecast.
Comp disputes are a data-integrity symptom, not a plan-design problem.
Tool sprawl is the residue of six years of point solutions bought to route around bad data.

Almost none of this is a tooling problem. It is presenting as one because tooling is the only thing anyone has tried.

The diagnosis

Forecast variance of 25% at $75M ARR is not a modeling failure. A model can only be as good as the stage definitions, the close-date discipline and the ARR definition feeding it. Buying a forecasting platform in month one takes a company that cannot agree on what a Stage 3 deal is and gives it a more expensive way to disagree — plus a six-month implementation that consumes the exact political capital the real fix requires.

The same logic applies to AI. An LLM applied to an unstable process produces confident output about unreliable data, which is strictly worse than no output, because people believe it.

So the sequence is: establish truth, then stabilize the process, then automate it. Each phase is only possible because the one before it happened. Doing them in any other order is how transformations fail loudly in month five.

What I would not do in the first 90 days

Explicitly deferred — and why

I would not replace Salesforce. The problem has not been established as a platform problem. Every migration I've seen that started before the data model was agreed on simply moved the same ambiguity into a new system, at a cost of nine months and all credibility.
I would not buy a forecasting platform. Not in the first 90 days, possibly not at all. Fix stage architecture and close-date discipline first, then re-measure variance. If it's still above 10%, it's a tooling conversation — and by then I can prove it.
I would not deploy AI into the GTM workflow. AI on unstable processes manufactures confident nonsense. It comes in month three, on the stable parts, where the inputs are trustworthy.
I would not redesign the comp plan. The plan probably isn't the issue; the crediting data is. Redesigning comp mid-year while reps already distrust their numbers costs more goodwill than any plan improvement returns.
I would not rip out the tool stack. Rationalization in month one removes things people depend on before I understand what they depend on them for. Inventory first, consolidate at renewal dates.
I would not reorganize the team. I don't know who the strong operators are yet. Structure decisions made in week two are made on the basis of who talks most in meetings.

A Head of Revenue Operations isn't valuable because she can list what's broken. Everyone in the building can do that. She's valuable because she can defend the order of operations to a CRO who wants the tool bought this quarter.

Month 1 — Establish truth

Days 1–30

Nothing gets fixed until the numbers mean the same thing to everyone

Objective: one definition per metric, one forecast baseline, one operating cadence — and zero new systems.

Listening tour with a deliverable. 25 conversations across sales, marketing, CS, finance and RevOps — but the output is a written "current state and contradictions" memo, not a summary of opinions. Circulating the contradictions in writing is what makes adjudication possible.
Adjudicate the metric dictionary. ARR, bookings, pipeline, qualified opportunity, MQL/SQL handoff, churn, NRR. Written definitions, one named owner per metric, signed off by CRO and CFO in the same room. This single artifact resolves most of the funnel argument between marketing and sales.
Reconcile the three ARR numbers. Walk CRM to billing to the finance workbook, document every variance, agree on the system of record. Finance stops maintaining a parallel truth.
Establish a forecast baseline. Rebuild the last six quarters: what was called, what landed, where the variance came from — slippage, categorization, or deals that were never real. Variance becomes attributable instead of mysterious.
Install the operating cadence. Weekly forecast call with a fixed agenda, monthly business review, quarterly planning. Same inputs, same format, every time. Cadence is what converts analysis into decisions.
Data integrity triage. Required fields at stage gates, ownership rules, duplicate cleanup on open pipeline only. Not a data project — the twenty fields that touch the forecast.

By day 30: a signed metric dictionary, one ARR number, an attributed variance history, and a forecast call people prepare for.

Month 2 — Stabilize the process

Days 31–60

Make the pipeline mean something and give commercial decisions a home

Objective: stage architecture, pipeline governance, Deal Desk, and comp credibility restored.

Rebuild stage architecture on exit criteria. Every stage defined by verifiable buyer behavior, not seller sentiment — economic buyer identified, technical validation complete, mutual close plan agreed. Stage-based probability becomes defensible, which is most of the forecast variance fix.
Forecast categories separated from stages. Commit, Best Case and Pipeline get written criteria and an inspection standard. Deals in Commit that fail the criteria come out on the call, publicly, until the behavior changes.
Pipeline governance. Aging thresholds, close-date-push policy, next-step requirements, and a standing hygiene report that goes to managers — not to reps — so it becomes a coaching input rather than a nag.
Stand up Deal Desk. Approval matrix, discount thresholds, non-standard terms path, and a 24-hour SLA. This is the function that makes revenue predictable at the margin and stops one-off exceptions from quietly becoming policy.
Fix comp crediting, not the comp plan. Reconcile the disputed quarters, automate crediting off the now-trustworthy CRM data, publish a statement reps can audit themselves. Trust in the number is worth more than any plan tweak.
Move renewals out of spreadsheets. Renewal opportunities auto-created in CRM at contract start with dates, owners and risk fields. Churn becomes forecastable instead of discovered.
Tool inventory with renewal calendar. Every contract, cost, owner, actual usage and renewal date. No cancellations yet — just the map that makes month-three decisions fast and defensible.

By day 60: stage exit criteria live, Deal Desk operating, renewals in the system, comp disputes closed, and a forecast whose inputs are now stable enough to model.

Month 3 — Automate and apply AI where the process holds

Days 61–90

Now — and only now — leverage

Objective: remove manual work, rationalize spend, and deploy AI on the stable surfaces.

Automate the highest-friction workflows. Lead routing, opportunity creation from renewals, approval routing, quote generation, CRM hygiene nudges. Target: give each rep back the Friday afternoon.
Deploy AI pipeline-risk scoring. Deterministic signals — stage aging, stale activity, close-date slippage, missing next steps, forecast-category inconsistency — feeding an LLM layer that writes the weekly executive brief. Possible in month three precisely because stage criteria and activity data became trustworthy in months one and two. This is the system I build.
AI on call and email data. Theme extraction across customer conversations for churn signals and product feedback. High value, low risk: it augments judgment rather than replacing a number.
Rationalize the stack. Using the inventory, consolidate the four overlapping tools at their next renewal dates. Fund the savings into the two capabilities that are actually missing.
Self-service reporting. A small set of certified dashboards with owned definitions, so leaders stop requesting analysis for questions the system should already answer.
Publish the operating model. What RevOps owns, what it advises on, what it doesn't touch; the request intake path; the quarterly planning calendar. Ambiguity about the mandate is the most common reason RevOps leaders stall in year one.

By day 90: forecast variance measurably down, renewals forecast rather than discovered, Deal Desk load-bearing, AI deployed where it compounds, and a documented operating model with a team plan behind it.

Priority stack — the five that matter

Of the 37 broken things, these are the ones I fund first. Everything else is sequenced behind them or deliberately left alone.

Priority	Fix	Why it goes first	Measured by
1	Metric dictionary & single ARR source	Every other fix depends on agreed definitions. Nothing downstream is defensible without it.	Zero variance between CRM and finance ARR
2	Stage architecture & forecast categories	The actual root cause of 25% variance. Not a tooling fix.	Forecast variance under 10% by Q2
3	Comp crediting integrity	Rep trust is the currency every later change spends. Disputes must stop before process change lands.	Zero disputes for two consecutive quarters
4	Renewals into CRM	Churn is a third of the ARR bridge and currently invisible until it happens.	90-day forward renewal visibility
5	Deal Desk	Makes commercial decisions consistent and shortens late-stage cycles immediately.	24-hour approval SLA, discount variance down
Later	Tool rationalization, AI expansion, comp plan redesign, team restructure	Each requires stable process or information I don't yet have.	Sequenced to renewal dates and Q2 planning

The operating model I'd publish on day 90

RevOps owns

Forecast process and accuracy · metric definitions and data integrity · CRM architecture · Deal Desk · territory and quota design · comp crediting · GTM systems roadmap · pipeline governance.

RevOps advises

Segmentation and ICP · pricing and packaging · headcount planning · marketing attribution · CS health scoring · board reporting narrative.

RevOps does not own

Sales management and coaching · quota acceptance · marketing programs · product roadmap · renewal conversations. Owning execution destroys the neutrality that makes the function credible.

Team shape

Role	Focus	When
Head of Revenue Operations	Operating model, forecast, exec partnership, planning	Day 1
Systems lead (existing)	CRM architecture, integrations, automation	Reassigned month 1
Analytics lead (existing)	Reporting, metric governance, pipeline analytics	Reassigned month 1
Deal Desk manager	Commercial approvals, pricing exceptions, contracting	Hire month 2
GTM data / AI engineer	Pipeline scoring, workflow automation, data pipelines	Hire month 4, once the process is stable

Four to five people at $75M ARR. The AI hire deliberately comes last — hiring that role into an unstable process is how companies end up with an expensive engineer building dashboards.

On the scenario: the company is fictional and composed from patterns common at this stage. The sequencing logic, the deferral reasoning and the operating model are how I would actually approach it.