Your CRM already holds revenue worth winning back

Past customers and recent leads deserve more of a sales team's time than they usually get. In a study of a telecom firm's lost customers, the ones with the strongest first relationship were likeliest to accept a win-back offer, and the best offer depended on why they'd left¹². In a separate study of 1.25 million sales leads, firms that tried to reach a lead within an hour were nearly seven times as likely to qualify it as firms that tried an hour later³.

Win-back works best when the offer fits why they left

V. Kumar, Yashoda Bhagwat and Xi Zhang followed lost customers at a large telecom firm, ran a randomized field experiment, and tracked who came back, how long they stayed and what they spent¹². Customers who had referred others, or whose problems the firm had fixed, were more likely to return. Frequent complainers were less likely².

Jacquelyn Thomas and colleagues found the best pricing in their case was a low price to win a customer back, then higher prices once they'd returned⁵. Across several industries, firms with a formal reacquisition policy won back more customers and did better financially⁶.

The famous retention numbers are hard to trace

The most repeated figure puts the odds of selling to an existing customer at 60% to 70%, against 5% to 20% for a new prospect⁴. It's usually credited to the 2006 textbook Marketing Metrics⁷, but it appeared five years earlier in the book Customer WinBack, which credits a study by a research firm also called Marketing Metrics and gives no sample or method⁴. Ipsos bought that firm in 2003⁸.

Harvard Business Review has summarized Fred Reichheld's Bain research as showing that a 5% rise in retention lifts profits by 25% to 95%⁹. The 1990 source reported single-company examples from a bank, an insurance broker and an auto-service chain¹⁰. Those businesses run on repeat accounts, so the numbers are a weak guide for a firm that sells one-off projects. The same HBR piece put the cost of a new customer at 5 to 25 times the cost of keeping one, depending on the study and industry⁹.

Then put rough numbers on a campaign. The defaults below are placeholders, so swap in your own.

Lead response time: the odds fall after the first hour

James Oldroyd, Kristina McElheran and David Elkington analyzed 1.25 million leads at 42 U.S. companies³. A lead counted as qualified when the firm had a real conversation with a decision maker³.

The same team sent a test lead through the websites of 2,241 U.S. companies and timed the replies³.

That study measured conversations, so it doesn't show how many became sales. The often-quoted 100x figure comes from a 2007 study by InsideSales.com and Oldroyd on the vendor's own platform data: the odds of reaching a lead fell 100 times between a call at 5 minutes and one at 30¹¹. It didn't measure closed sales either¹¹.

Why old leads and lapsed customers go unworked

In Salesforce's 2026 survey of 4,050 sales professionals, the average seller spent 40% of their time selling¹², and calls to former customers compete with live deals for that time. In a survey by Validity, which sells data-quality software, 76% of 602 CRM users said less than half their CRM data was accurate and complete¹³.

Closed-lost deals have less research behind them, since the win-back studies followed customers who had bought and left. The best guide is the loss reason your team recorded, because the reason shaped both who came back and which offer worked².

Can AI run a CRM reactivation campaign?

Salesforce says its own AI agents contacted 130,000 leads its reps never worked and created 3,200 opportunities in four months¹². That's a vendor describing its own product, with no baseline or revenue figure.

In a field experiment with more than 6,200 customers, undisclosed chatbots sold as well as proficient staff on outbound calls, but saying up front that it was a bot cut purchases by more than 79.7%¹⁴. With that result and the consent rules below, the safer setup has AI draft the message or brief the caller, and a person make contact.

Generic personalization has weak support. A first name in the subject line raised email opens by 20% in one study¹⁵, but a 2023 replication in German found no effect from first names¹⁶. The win-back research points to a more useful kind: an offer that answers the reason the customer left². For how follow-up compares with other jobs on strength of evidence, see which sales and operations jobs AI has proven it can do.

Consent windows for win-back email and calls run 3 months to 2 years

Who you may contact, and for how long, depends on the channel and the country. Check these with counsel before a campaign.

• Canada's CASL implies consent to email for two years after a purchase and six months after an inquiry¹⁷¹⁸. Unsubscribes must take effect within 10 business days¹⁸.
• Canada's business-relationship exemption covers calls to numbers on the National Do Not Call List for 18 months after a purchase and 6 months after an inquiry¹⁹.
• U.S. CAN-SPAM requires an honest subject line, a postal address and an opt-out honoured within 10 business days, with penalties up to $53,088 per email²⁰.
• The U.S. exemption covers Do Not Call numbers for 18 months after a purchase and 3 months after an inquiry, but not prerecorded calls²¹.
• Since a 2024 FCC ruling, AI voices count as artificial voices, so AI sales calls to U.S. numbers need prior written consent²².

How to start a win-back program

1. Pull customers who bought in the last two years, plus inquiries and closed-lost deals from the last six months. Those are the CASL implied-consent windows¹⁷.
2. Clean the list and tag each lost customer or deal with why it was lost².
3. Match the offer to the reason: a discount for price leavers, a service offer for service leavers².
4. Write the process down, with an owner, the segments and the offers each may get⁶.
5. Route new web leads to a named person with an alert, and answer within the hour³.
6. Have AI draft each message from the customer's history, and have a person check and approve it. Why staff training decides the result covers teaching staff to check AI drafts.
7. Hold back a random slice of each list. Comparing it with the contacted group shows what the campaign added beyond customers who'd have come back anyway.