How to scope an account-growth analytics pilot

A pilot exists to settle one question with a number. Eight success criteria, four of them about whether sellers actually used it, because an unused signal scores zero however accurate it was. Plus the decision rule for what happens at the end, written while nobody is invested in the answer.

A pilot exists to settle one question with a number. Most account growth pilots fail because they were scoped to demonstrate a capability instead, and a capability demonstration has no failing grade.

What belongs in the scope

What belongs in a pilot scope

A pilot exists to settle one question with a number. Anything that does not help settle it belongs in phase two.

A pilot that cannot name the segment, the owner and the two measures before it starts is a data project with a sales label on it.

Source: Thinklytics case library, published engagement scopes and outcomes.

Six items in, six deferred.

One named segment and one named motion. Renewals in one region. Expansion in one product line. Churn risk in the enterprise book. Narrow enough that whatever happens is attributable to this work rather than to the quarter.

The signal source, instrumented or connected. If the signal does not exist today, instrumenting it is the pilot, and that is a legitimate pilot. One B2B SaaS company had no usage tracking at all, so the work was mapping 140 user actions before any scoring existed.

A resolved account identity across those sources. Only across the sources the pilot touches. A signal attributed to the wrong record is worse than no signal because it looks like information.

Delivery into the tool the owner already uses. Covered below, because it is the most commonly broken rule.

A stated threshold, and a named owner per alert. Not a team. A person, and their manager knew before the pilot started.

An action rate and a revenue measure, agreed up front. The two numbers the pilot will be judged on, written down before anyone can see which way they will land.

Deferred: every segment at once, a full data platform, enterprise-wide master data, a new user interface, an approval workflow, and attribution modelling across all channels. All real work, none of it helps settle the question, and each one adds weeks while diluting the attribution.

Why the action rate comes first

Four of the eight success criteria are about adoption rather than accuracy, and the action rate leads.

The action rate is the share of flagged accounts a named owner actually worked, within a stated number of days. It goes first because everything else is downstream of it: a perfectly accurate signal nobody works returns zero, and more model quality does not change that.

It also makes the pilot diagnostic, which is the property that makes a pilot worth running at all.

A high action rate with a weak revenue result means the signal was wrong, and phase two is about the signal. A low action rate means the constraint is ownership or workflow, and phase two is about the motion rather than the analytics. Those are different projects with different owners and different budgets, and a pilot that only measures model accuracy cannot tell you which one you are looking at.

The uncomfortable evidence on follow-through is in lead scoring: why scores do not move revenue.

Success criteria

Pilot success criteria, and the adoption plan behind them

Agreed before the pilot starts. Four are about whether sellers used it, because an unused signal scores zero regardless of how accurate it was.

• Action rate: the share of flagged accounts a named owner actually worked, within a stated number of days. The first criterion, deliberately. Everything else is downstream of whether anyone acted.
• A revenue or retention figure attributable to the flagged cohort. $1.8M from 84 flagged accounts in six months. 340 accounts retained in a quarter. State the measure and the window.
• Precision measured on the band the team actually works, not overall. A high headline accuracy can hide poor precision in the top decile, which is the only part a seller ever sees.
• Signal latency: time from the event occurring to the owner being alerted. Renewal flag 90 days before expiry rather than a reminder 45 days after. Latency against window size is the whole game.
• Delivered inside a tool the owner already had open. Mobile app for field reps, Slack for customer success managers, the advisor's existing review workflow. Never a new portal for a pilot.
• A named owner per alert type, agreed with their manager. Not a team. A person, and their manager knew before the pilot started.
• An escalation rule for alerts nobody works. One distributor escalated to a manager if the rep had not followed up within 30 days. That rule is part of why the renewal numbers moved.
• A decision rule for what happens at the end of the pilot. What result extends it, what result stops it, written down while nobody is invested in the answer.

The last one protects you from the most common pilot ending, which is neither success nor failure but an inconclusive result everybody argues about for a quarter.

Source: Thinklytics case library, published delivery outcomes and acceptance measures per engagement.

Eight, agreed before the pilot starts. Three are worth expanding.

Precision on the band the team actually works, not overall. A healthy headline accuracy can hide poor precision in the top decile, and the top decile is the only part a seller ever sees. Name the band in the criteria.

Signal latency, from the event occurring to the owner being alerted. This is the criterion most pilots omit and it decides whether the signal is actionable at all. One distributor's renewal reminders had been arriving 45 days after the contract expired; moving the flag to 90 days before expiry took the renewal lag from 45 days to 6 and quarterly renewals from 128 to 160.

An escalation rule for alerts nobody works. The same distributor escalated to a manager if the rep had not followed up within 30 days. That rule is part of why the renewal numbers moved, and it is usually left out because it reads as a trust issue rather than as a control.

Delivery has to land where the owner already is

Salesforce's State of Sales, seventh edition, is an anonymous survey of 4,050 sales professionals across 22 countries conducted August through September 2025 with third-party panelists.

It found reps spending 40% of an average workweek meeting with customers and more than half their time on nonselling work, with 42% saying they are overwhelmed by too many tools and 51% of sales leaders with AI reporting that tech silos delay or limit those initiatives.

Seller attention is the scarce input in this whole exercise. A pilot that requires a new login is competing for the least available resource in the company, and it will lose to the account knowledge the rep already has.

What has worked in our engagements: a mobile app for 180 field reps carrying equipment details, service records and contract status. Slack alerts to named customer success managers when an account crossed a churn risk threshold. A client risk profile inside the review workflow 42 advisors already ran, which took portfolio review from three hours to 22 minutes and surfaced $14.2M of rebalancing opportunities.

None of those was a new portal.

What nine pilots took and returned

Nine account growth pilots, what they took and what they returned

Published outcomes. Ten to 18 weeks, and the ones that returned the most were not the most sophisticated.

The $2.8M came from a rule and a mobile app. The $14.2M came from putting a profile in the workflow an advisor already ran. Neither needed a new platform.

Source: Thinklytics case library, published delivery durations and outcomes per engagement.

Ten to 18 weeks across nine engagements, and the duration tracked how much signal had to be created from nothing rather than the sophistication of the analytics.

Ten weeks to instrument 140 product events and flag 84 accounts ready to expand, worth $1.8M in six months, with the churn warning window moving from 60 days to 14. Ten weeks to get a churn model into daily production with alerts to named CSMs, which flagged $2.6M of at-risk ARR and retained 340 accounts in the first quarter. Twelve weeks for the field rep app and the renewal rule, which returned $2.8M of cross-sell in year one. Twelve weeks for a medtech territory design delivered eight weeks before commercial launch, worth $2.4M of territory optimisation and 1.34x revenue against the original plan.

The two highest-returning numbers in that list came from a rule plus a phone, and from putting a profile into a workflow somebody already ran. Neither needed a new platform, which is the thing worth remembering when a pilot proposal arrives with an architecture diagram.

The decision rule

Write down, before the pilot starts, what result extends it, what result stops it, and who decides.

This exists because the most common pilot ending is neither success nor failure. It is an inconclusive result that gets argued about for a quarter while the team that ran it moves on and the signal quietly stops being maintained.

Pick thresholds on the two agreed measures. Something like: an action rate above a stated percentage and an attributable revenue figure above a stated number extends to the next segment; an action rate below a stated percentage stops the work and triggers a conversation about ownership rather than about analytics.

Agreed in advance, the end-of-pilot meeting takes 20 minutes. Agreed afterwards, it takes a quarter.

The account growth pilot success criteria worksheet is what we fill in with clients before a pilot starts, including the sales adoption plan, because the person approving the pilot is often not the person whose team has to work the alerts.

Five questions before signing

Show me an action rate from a previous pilot, not just a model accuracy. Which segment and which motion, by name, and who are the owners. Where does the alert land, in which existing tool. What signal latency are we targeting, from event to owner. And what is the decision rule at the end.

A firm that has run these answers the first question with a number and the last one without hesitating. A firm that has not will answer the first with an accuracy figure, which is the tell.

What we would do first

Build the rule by hand for one segment, in a spreadsheet, and send it to the owners. Three weeks later, count how many were worked.

That action rate costs nothing and it is the most informative number available before any spend. It tells you whether the pilot you are about to scope should be about the signal or about the motion, and that distinction is the difference between a pilot that settles something and one that produces a deck.

Which approach the signal calls for is in rules-based prioritisation vs predictive scoring, and why the opportunity is being missed in the first place is in why expansion opportunities get missed in existing accounts.

Delivery sits in pipeline and revenue analytics for the account view, forecasting and optimisation where a model is warranted, analytics and BI for the delivery surface, sales and CRM AI automation where the alert has to land in the seller's existing tooling, and team enablement where the adoption plan needs more than an email. The full set of work in this area sits under we react instead of predicting.

Frequently asked questions

What should an account-growth pilot include?

Six things. One named segment and one named motion, narrow enough that the result is attributable. The signal source, instrumented or connected. A resolved account identity across those sources. Delivery into the tool the owner already uses rather than a new portal. A stated threshold with a named owner per alert. And an action rate plus a revenue measure agreed before the pilot starts. A pilot that cannot name the segment, the owner and the two measures up front is a data project with a sales label on it.

What are the success criteria?

Eight, and four concern adoption rather than accuracy. Action rate, meaning the share of flagged accounts a named owner worked within a stated number of days. A revenue or retention figure attributable to the flagged cohort. Precision measured on the band the team actually works rather than overall. Signal latency from event to alert. Delivery inside a tool the owner already had open. A named owner per alert type agreed with their manager. An escalation rule for alerts nobody works. And a decision rule for what happens at the end.

Why is the action rate the first criterion?

Because everything else is downstream of whether anyone acted. A perfectly accurate signal that nobody works returns zero, and more model quality does not change that. Putting the action rate first also makes the pilot diagnostic: a high action rate with low revenue means the signal was wrong, and a low action rate means the problem is ownership or workflow rather than analytics. Those two outcomes point at completely different second phases.

How long should a pilot run and what does it cost in time?

In our case library these ran 10 to 18 weeks, and the shortest were the narrowest. Ten weeks to instrument 140 product events and flag 84 expansion-ready accounts. Ten weeks to get a churn model into daily production with alerts to named customer success managers. Twelve weeks to put equipment and contract data on 180 field reps' phones. The duration tracked how much signal had to be created from nothing rather than the sophistication of the analytics.

What should be deferred to phase two?

Every segment at once. A full data platform. Enterprise-wide master data. A new user interface. An approval workflow. And attribution modelling across all channels. Each is real work and none of it helps settle the pilot's one question, so each one adds weeks and dilutes the attribution the pilot exists to produce. Resolve identity only across the signal sources the pilot touches, not across the enterprise.

Why does delivery have to be in an existing tool?

Because seller attention is the scarce input. Salesforce's State of Sales, seventh edition, surveying 4,050 sales professionals across 22 countries between August and September 2025, found reps spending 40% of an average workweek meeting customers and more than half their time on nonselling work, with 42% saying they are overwhelmed by too many tools. A pilot that requires a new login is competing for the least available resource in the company. Mobile app, CRM field, Slack alert, or the review workflow the owner already runs.

What is the decision rule and why write it first?

A written statement of what result extends the pilot, what result stops it, and who decides, agreed while nobody is yet invested in the answer. It exists because the most common pilot ending is neither success nor failure but an inconclusive result that gets argued about for a quarter. Pick the thresholds on the action rate and the revenue measure in advance, and the end-of-pilot meeting takes 20 minutes.

What should I ask a firm before signing?

Five questions. Show me an action rate from a previous pilot, not just a model accuracy. Which segment and which motion, by name, and who are the owners. Where does the alert land, in which existing tool. What is the signal latency we are targeting, from event to owner. And what is the decision rule at the end. A firm that has run these answers the first question with a number and the last one without hesitating.

The work behind this

Nine account growth engagements in the case library published a measured result, running 10 to 18 weeks. Each states the segment, the delivery surface and the outcome, including the two highest-returning ones that used no model and no new platform.

Lead scoring and churn prediction, 6 engagements.