How to Re-Check Your ICP Every Quarter Before It Drifts

At FieldAssist, where I was CMO from April 2023 to September 2024, mid-market and enterprise qualified leads grew 2.25x. A count like that only means something while the definition of "qualified" holds still. If the profile underneath it moves and nobody writes the move down, the same number starts describing a different market.

A new region opens and its first deals look different. An acquisition brings a buyer the ICP never named. Those are decisions, and nobody carries them into the written page. Meanwhile the lead score, the ad audiences and the first-call script keep pointing at last year's picture, so marketing qualifies one profile while sales wins another.

Start with a test I have written about. Ask five people who your ICP is. "Five different answers means the motion was never agreed on" (LinkedIn, 17 Jun 2026). Settle that first.

The method is a blind read, a fit test and a set of trials. A spreadsheet does most of it. Each tool gives you a first draft to check. In these runs, the outputs still needed the manual fit and money checks below.

The ICP review skill sits in the B2B ICP and Customer Insight plugin, in review for the Claude directory and not yet installable. Until then, write its table by hand. ICP Intelligence's gap analysis sizes the same metric gaps.

Guide 03 builds the ICP. Guide 23 takes a confirmed change into the lead score.

Step 1. Read what this quarter's wins share

About this example. Protectt.ai is a real company. Protectt.ai has not reviewed this guide. Facts about Protectt.ai come from public sources, linked alongside with the date I checked them. Every number marked illustrative, and every ICP line, deal, customer type, loss reason and metric in the example, is an input I chose to show the method. None of it is Protectt.ai's data, and the tool outputs are exercise results, not Protectt.ai's real strategy.

What is public: Protectt.ai, a mobile app security company, raised a Series A of INR 76 crore (about $8.7 million) in March 2025, with expansion focused on the US, the Middle East and the Asia-Pacific region. In September 2026 its Middle East head said the company started in the region "just over a year ago" and has "double-digit clients" across the Middle East and Africa (company-reported), and that the UAE Central Bank has mandated mobile application threat protection for listed financial institutions. On 27 October 2026 it announced a majority stake in AIShield, an AI security platform covering red teaming and runtime protection, adding to its existing AI security work.

The mandate claim above comes from the company interview; ask the bank to confirm the applicable notice before using it to qualify an account.

That Middle East move was a decision, made in public, and nothing here says Protectt.ai's own ICP missed it. The exercise asks: if you ran go-to-market at a company like this and the written ICP had never caught up, what would a check on Q3 2026 show? The illustrative ICP, pasted in exactly these words: "Banks, NBFCs and insurers in India with a customer-facing mobile app of 1 million or more installs. Trigger: a regulator mandate on mobile app security. Signer: CISO. Champion: Head of Digital Banking."

Input to the pattern tool: company "Protectt.ai", product_category "mobile app security and fraud control for banks and insurers", and the eight illustrative rows alongside, four in India and four in the UAE. What came back: banking as the largest industry with 3, the Head of Fraud Risk as champion in "4 deals (50%)", and two wins triggered by the launch of an AI assistant inside the app.

How to read it: this is a fast pivot, no more. The tool has no region field, so count that yourself. In these illustrative rows, 4 of 8 wins came from the UAE, including the three largest contracts, and by my arithmetic UAE wins averaged $115,000 against $56,250 in India.

Step 2. Test the written ICP against wins and losses

Input to the evolution tracker: the same company, product_category and ICP, time_period "Q3 2026 (July to September)", and the lists alongside: eight win lines, four loss lines covering six illustrative losses, and two market changes.

What came back: eight candidate additions, four candidate disqualifiers, two market implications and a draft ICP. That is one candidate per line, so read it as your own list, not as analysis. In this run, the tracker missed a match: the Indian private bank win, which matches the ICP's segment, place, trigger and champion, came back "not in your current ICP". Do not adopt its draft. "Add on trial" and "Qualify out on trial" hold the first three lines of each list, so here the second qualifies out an Indian bank on a single loss. Keep its CRM pull list as a checklist.

A repeat with the same inputs on 2 November still missed that bank match. It asked for the deciding loss reason before creating a disqualifier, but kept the same deals under "Qualify out on trial". Resolve that conflict in the review before excluding an account.

Now the test no tool here runs: would the written ICP have picked these deals? Part by part, 5 of 8 wins fit the segment, 4 the place, 3 the trigger and 1 the champion. Installs and signer were never recorded. Take segment and place across all 14 illustrative closed deals: inside the written ICP 3 of 7 won, 43%, and outside it 5 of 7, 71%. An ICP that works wins more inside than outside. These rows do the reverse.

It is still 14 deals, and one deal moves the total win rate 7 points. My suggested rule: under about 20 closed deals a quarter, pool the last four quarters. This exercise shows one quarter to keep the rows short.

Step 3. Put gaps in money, cut by segment and read who stays

Four illustrative metrics against target: contract value $85,625 against $100,000, cycle 118 days against 100, win rate 57% (8 of 14) against 50%, and churn 8% against 5%. The ICP review skill's gap script printed churn as the largest percentage gap, "38% reduction needed", with contract value at "+17% increase needed". Percentages of different bases do not compare, so use money. On an illustrative $6 million of recurring revenue, 3 points of churn is $180,000 a year. At 8 wins a quarter, $14,375 more on each of 32 wins a year is $460,000. On these inputs, contract value leads.

In these rows, UAE wins at $115,000 against $56,250 looks like place. Cut by segment, alongside, and most of it is banks: a bank pays about twice what a non-bank pays in either country, while inside a segment the UAE pays about 1.3 to 1.4 times India. Six banks in eight, with two non-banks at $48,750, reach about $97,200 with no new region. Six UAE trial wins at their $103,333, with two Indian wins at $56,250, reach about $91,600.

A win shows who buys, and only a renewal shows who stays. If contracts run a year, none of the eight has reached one, so the illustrative 8% describes customers signed earlier. Pull the customers signed 12 to 24 months ago, split those who stayed and expanded from those who left, and read them against the ICP. Higher churn outside the ICP is a reason to investigate fit, not proof that drift caused the losses. Similar churn rates do not settle the cause: compare the loss reasons, contract age and onboarding experience before changing the ICP. Once a year, rewrite the ICP from this cohort.

Step 4. Write each change as a trial that can fail

The skill proposed the five changes alongside, and its marked draft leaves "Signer: CISO" Open, because the wins never recorded who signed. The table is that first draft with the tests and disproofs rewritten by hand. It is the agenda for the quarterly ICP meeting, not the new ICP. Five suggested rules, mine and not the skill's, decide whether a trial can fail.

1. Test on quarters that did not raise the idea. Q3 raised it, so the back test runs on Q1 and Q2 2026. If they hold too few deals, rely on the forward read.
2. Count only deals that start after go-live. With the illustrative 95-day cycle, a trial starting in mid November closes its first new deals around mid February 2027, so a Q1 read mostly measures deals already open.
3. Set a floor and a date in advance. Illustrative floor: 8 closed trial deals. At Q3's pace of 3 a quarter, that arrives about October 2027. Miss it and the trial ends unproven, and the line stays out of the written ICP.
4. Keep a control for anything you stop selling to, or nothing can disprove the cut. Here a random half of the small NBFCs stays on the old treatment: at half of Q3's 3 deals a quarter, about 4 closed deals by October 2027, again illustrative.
5. Hold no more than two unread trials at once. A new change waits for a slot.

My two picks, on this illustrative data; neither goes live until its back test agrees. UAE banks and regulated fintechs go on trial, though not on contract value, which is mostly segment. The case is the mandate and win rate: 4 of 5 UAE deals won against 4 of 9 in India. That is a lead, not a result: one deal moves the UAE figure 20 points, and without the three small NBFC losses India is 4 of 6. So compare like with like. If bank against bank shows no UAE edge, the trial stops and the candidate becomes banks first, in either country.

All three wins the trial covers had the Head of Fraud Risk as champion, so the lists name that role, and a result reads place and role together. A new region is a cost decision too: add the cost to win and serve, and count the listed institutions left, because a mandate wave ends once they have bought. The churn half of the disproof needs a first renewal, so book it for about February 2028.

NBFCs under 1,000 employees qualify out on trial, on the same bar: three deals are a lead. First call the three buyers, because a CRM reason is what the seller typed, and "price" may be a packaging fault. Then pull their installs, because the ICP sizes accounts by installs. The cut also flatters you: dropping them lifts the illustrative win rate from 57% to 73% (8 of 11) with no change in selling, so report both.

I would hold the other three. The Head of Fraud Risk has four wins, but three are UAE deals, so that signal cannot yet be told apart from the place. The AI assistant trigger has two wins, both UAE, and the slowest cycle at 150 days. A second owner rests on one stalled deal. No win here bought the AI security stake.

Version the score: back-score the last two quarters and report old and new side by side for a quarter, or your count of "qualified" stops meaning anything.

This is the point where I would want a second pair of eyes: quarterly evidence conflicts before an ICP change. Bring me your wins and losses by segment, retention, money at risk and proposed trial, and we will work it through in 45 minutes.

1. Ask five people who your ICP is, before anyone opens it.
2. Export closed won and lost deals with country, champion, signer and the reason the buyer gave. Pool four quarters if you close under about 20 deals a quarter.
3. Build the pivot of wins, by hand or with the pattern tool.
4. Work out win rate inside and outside the written ICP, and pull the customers signed 12 to 24 months ago.
5. Write each change with its back test, and a disproof with a floor and a date.
6. Hold a 45-minute ICP meeting with sales, product marketing and customer success, with one named person who decides. Keep to two unread trials, and send each to scoring, targeting and messaging the same week.

1. Reading the ICP before the wins. Look at what the wins share first, so the old profile does not frame it.
2. Changing the ICP on one big deal. Put it on trial and write what would disprove it.
3. Recording the champion but not the signer. The ICP's signer line then goes untested.
4. Reading a trial on the quarter that raised the idea, or on deals that started before it.
5. Confirming a change and leaving the score alone. It belongs in the lead score and the lists that week.

• How to Size Your B2B Market Bottom-Up and Score Target Accounts Without Inflated Analyst Numbers
• Lead Scoring and MQL Rules That Sales Will Trust
• Win-Loss Interviews That Change the Message and the ICP
• All 30 guides