What a person needs to know in the first moments of a handover

The situation

A policyholder calls an insurer at 11pm after a collision. They have the police report, it puts the other driver at fault, and they want to know what happens next and whether they will pay an excess. The agent takes the details, records that the customer holds the report, and explains that liability and the excess are decided by the claims team, not on the call. The customer is unhappy; they want an answer now. That is not a conversation an agent should settle, so the agent says a claims handler will call before noon, and escalates.

At 9am the customer calls again, anxious. The person who picks up sees one of two things: a claim number and a status, or a 40-turn transcript. In the first case the customer starts again. In the second, the person reads while the customer waits, and the customer starts again anyway, because silence on a line sounds like nobody is there.

Nobody in this chain did anything wrong. The failure is in what travelled between them.

What the evidence shows

Repetition costs loyalty, though less precisely than the headlines say

In a survey of about 5,100 consumers across 22 countries, run in mid-2024 and published in a 2025 report, 63% said they were willing to switch to a competitor after a single bad experience [1]. That is stated intent, not observed switching. On repetition, the foundational study, a three-year survey of more than 75,000 customers published in 2010, found 56% reported having to re-explain an issue and 62% having to contact the company repeatedly to resolve one [2]. A 2023 survey of 14,300 consumers and business buyers in 25 countries found 56% often have to repeat information to different representatives [3]. On transfers specifically, the only figures come from one North American benchmark firm with an unpublished method: 19% of callers are transferred, and transferred calls score 12% lower on satisfaction and 14% lower on first-contact resolution, the share of issues settled in a single contact [4].

There is a serious counter-finding. A peer-reviewed study of 6,649 Dutch consumers rating 93 firms, who were asked two years later whether they had stayed, found that the Customer Effort Score, the survey question asking how much effort a contact took, had "little to no predictive power" for retention [5]. Customers dislike repeating themselves. Whether that alone drives them away is less certain than the industry figures imply.

The strongest evidence for structure is clinical, and it is about completeness, not brevity

The best-studied handover format anywhere is I-PASS, a fixed order for a doctor handing patients to the next shift: illness severity, patient summary, action list, situation awareness and contingency plans, and finally synthesis by the receiver, who repeats back what they heard [6]. In a 2014 study across nine hospitals and 10,740 admissions, medical errors fell from 24.5 to 18.8 per 100 admissions, a 23% reduction, and preventable adverse events, harm to patients that better care would have avoided, fell by 30%. Adverse events a handover could not have influenced did not change, which is the control that makes the result credible. Handovers took 2.4 minutes per patient before and 2.5 after: the structure did not make them shorter [7]. A 2023 follow-up in 32 hospitals found that before the format was introduced, only 20% of verbal and 10% of written handovers contained all five elements [8].

Three cautions. The 2014 study ran in paediatric teaching hospitals in North America, compared before with after rather than against a parallel control group, and was run by the format's own developers. Every positive study tested a bundle: the format plus training, observation and coaching, never a template alone. And the wider evidence is thin: the 2024 US federal evidence review rates I-PASS as moderate certainty and the older SBAR format (situation, background, assessment, recommendation) as low [9], and a 2014 review of 29 handover studies found that the literature "does not confirm that any methodology reliably improves the outcomes of clinical handover, although information transfer may be increased" [10]. Structure reliably increases what gets passed on. Only one format has decent evidence that this reduces harm.

Safety-critical industries mandate structure because its absence was found in specific disasters

Air traffic control has a written procedure for every change of controller. The relieving controller reviews a posted status display before being briefed, the verbal briefing is reserved for what the display does not show, and the transfer includes an explicit statement that responsibility has passed [11]. The relieving controller "must be responsible for ensuring that any unresolved questions" are resolved "prior to accepting responsibility" [12]. European guidance advises watching and listening for five to ten minutes before taking over, and the US checklist and two of the three European checklists in the guidance end with the same item: live traffic, the thing the receiver must act on first [12, 13].

The purpose of structure here was never brevity. In a 1980 analysis of 52 reported handover incidents, 26 involved a briefing that was inaccurate and 21 one that was incomplete; only 4 involved no briefing at all, and the incoming controller erred about as often as the outgoing one [14]. The briefings that failed were the ones that happened.

The UK Health and Safety Executive, whose guidance cites the 1988 Piper Alpha oil platform fire, prescribes that shift handover be face-to-face, two-way, both verbal and written, and "given as much time and resource as necessary" [15]. At the 2005 Buncefield fuel depot explosion, supervisors "were confused as to which pipeline was filling which tank" because of "deficiencies in the shift handover procedures" [16]. Maritime rules add one more principle: a watch is not handed over while a manoeuvre to avoid a hazard is in progress [17].

We found no controlled study in any of these fields showing that its format cuts incidents. What they share is a prescription: complete, fixed in order, checked by the receiver, and unhurried. The customer-service problem is that the receiver cannot be unhurried.

What a receiving colleague actually needs

A warm transfer is one where the person handing over speaks to the person receiving before the customer is connected; a cold transfer simply moves the call. We found no peer-reviewed study comparing the two in contact centres. The nearest evidence is from healthcare, where a "warm handoff" means a clinician personally introduces a patient to the next clinician. In the largest study, 2,690 referrals at a Boston hospital, the introduction made no difference to whether patients attended their next appointment; what predicted attendance was getting an appointment the same day [18]. A smaller study of 328 adolescent referrals found the opposite, with 44% engaging after a warm handoff against 19% without [19]. The evidence is mixed, and the safest reading is that continuity and speed matter more than the ceremony.

What matters is what arrives. A 2022 study interviewed six customer-service specialists and then watched ten people handle simulated handovers from an automated agent. The essentials they named were the customer's identity, the object of the request, the problem category, the reason the agent stopped, and any solutions already proposed. When the agent's proposed solutions were hidden, receivers asked redundant questions, and in eight of the ten tests they repeated questions the agent had already asked [20]. A small study, but the only one that watched the moment of handover.

Two studies say something about how the handover should begin. In scenario experiments, people whose automated conversation failed rated the recovery higher when the transfer was offered automatically or with a single button, rather than when they had to ask [21]. For complaints, and after a second failure of any kind, people preferred a person to another attempt by the agent [22]. And customers who arrive angry rate an automated agent designed to seem human worse than one that is not [23]. Anger is a reason to transfer early and plainly, not to soften the voice.

A person reads about four words a second, and skims

A 2019 meta-analysis, a study pooling earlier studies, covered 190 of them and 18,573 participants and puts adult silent reading of English non-fiction at 238 words per minute [24]. Arabic is comparable: in two eye-tracking experiments at a university in the UAE, 40 young native Arabic readers per experiment read sentences silently at 248 to 261 words per minute [25]. A 100-word brief is therefore about 25 seconds of reading, while the customer waits, and on at least one major contact-centre platform every second of a pre-connection announcement counts as handle time, the measured length of the call [26].

When people skim, the pattern is predictable. Eye-tracking on 28 readers found the first half of a paragraph is read for longer than the second, and paragraphs near the top of a page longer than those below [27]. Working-memory research puts the number of separate items a person can hold at about four [28]. Whatever must not be missed belongs in the first line of the first block, and there should be about four blocks.

A summary can be wrong, and a wrong summary overwrites what the reader knew

A brief that can be checked against its source is worth more than one that cannot, because generated summaries carry errors at a measurable rate. In a 2022 evaluation, 35% to 50% of summaries produced by the language models of that time contained a factual inconsistency with the dialogue; the error types included the wrong person, the wrong pronoun and a lost negation [29]. A 2025 study in which a large language model summarised 450 clinical consultations found 1.47% of generated sentences contained content not in the transcript and 3.45% of source sentences were omitted; negation errors, where something that did not happen is recorded as if it did, were the most likely to cause harm [30]. A 2026 experiment with 328 participants found that after reading a misleading summary of an event they had watched, correct recall of the key detail fell from 83.6% to 44.8%, whether the summary was labelled as written by a person or a machine [31].

Two Gulf-specific compounders. Speech recognition on speech that mixes Arabic with another language, which is how many Gulf customers talk, reports word error rates, the share of words transcribed wrongly, between 24.8% and 53.8% across published test sets [32]. And three expert annotators labelling frustration in 555 real service conversations agreed only moderately with one another [33]. A quoted phrase or a frustration flag in a brief is evidence, not a verdict.

Commitments, emotion and vulnerability are the items that must travel

The UK telecoms regulator states that "vulnerable customers should not need to repeat themselves when they are put through to another person or department" [34], and the UK financial regulator expects staff to record a customer's needs and "know how to access and use previously recorded relevant information" [35]. We found no UAE or Saudi rule on transfers or repetition. The Central Bank of the UAE requires a unique reference number for every complaint, a final written response within 30 business days and records kept for five years [36]; the Saudi Central Bank similarly requires a reference number and an escalation route [37]. The record must exist. Nothing we found says it must be in front of the next person.

On emotion, a 2008 study of UAE bank customers found that how people were treated in the interaction, and what they received, drove loyalty, while procedural fairness, for which the study's example is timeliness, "has no impact" on either emotions or the decision to leave [38]. One older study, but Gulf evidence, and it points the same way as the wider literature: a meta-analysis of 24 studies found that a good recovery can leave customers more satisfied than before the failure, but that the effect does not extend to their intention to buy again [39], and a second found that clearly stating the waiting time and the procedure "can strengthen positive emotions" during a recovery [40]. A promise made by the first agent and not honoured by the second is, in our reading, a second failure. No study isolates that case, so we treat it as a design risk rather than a measured one.

What that means

We started from the hypothesis that handovers fail because context arrives unstructured or not at all, and that a short fixed structure fixes it. The evidence supports the first half and corrects the second. The typical failure is a briefing that happened and was incomplete or wrong, not one that was missing. The fields with the strongest practice do not prescribe brevity. They prescribe completeness on a fixed list, an order that puts the urgent thing first and the live thing last, and a receiver who checks. In customer service the receiver has seconds, so the design problem is to compress without dropping, and to give the receiver a way to check that costs the customer nothing.

Three trade-offs follow. A summary saves reading time and introduces error; the remedy is provenance, so every line can be traced to the words it came from. A frustration flag helps the receiver prepare and is noisy; the remedy is to show it as a signal with its evidence, not a label. And a transfer offered early satisfies angry customers more, at the cost of more transfers; the remedy is a brief good enough that each one is cheap for the person receiving it.

How we design for it

These are the decisions behind the handover brief our agents deliver to a person the moment a conversation transfers.

Four blocks, in a fixed order. Who: the customer, the language and dialect they used, how many times they have contacted you about this, and any vulnerability the agent noticed. What happened: the request, the facts established from your systems, what the agent tried, and why it stopped. What was promised: every commitment, with the time it was made. What's next: the one open action the customer is waiting for, last, because it is the first thing the person will say. About four is what a person can hold at once, and the order borrows from I-PASS and the air traffic checklists: severity at the top, the live item at the bottom.

The first line carries the reason for the handover. Skimmers read the top of the first block, so the brief opens with why the conversation is being handed over and the customer's state, in one sentence: "Escalated: wants a liability answer now. Frustrated, raised voice twice. Arabic, Gulf dialect, numbers in English."

Every line links to the turn it came from. Summaries can be wrong. The person receives the full conversation alongside the brief, and each block points to the moment in the transcript it summarises, so a commitment or a quoted phrase can be checked before it is repeated to the customer.

The agent writes the brief. In a human-to-human warm transfer, the transferring colleague's time is the cost. Here the agent writes the brief as the call is handed over, so the person starts informed rather than cold, and the transfer can be offered early without a colleague having to stop and write it. That is the difference between a transfer and a handover.

Frustration and low confidence trigger the handover, and both are shown as evidence. You set the rules for when a conversation goes to a person. Our agents also hand over when their confidence in an answer falls below a threshold you choose, and when tone and language suggest the customer is frustrated or distressed. The brief shows the flag with the words that raised it, because even expert annotators only moderately agree about what frustration looks like, and the person taking the call should judge for themselves.

Vulnerability can travel with the record. For clients who need it, routing is configured during implementation to recognise signs that a customer may be vulnerable and send them to a person with the full conversation, so the customer does not explain their circumstances twice.

No handover in the middle of an action. If the agent is part way through an action in your systems, it completes the step, or records plainly that it did not, before the person joins. A half-finished action is the maritime manoeuvre: the relief waits.

The person confirms before the customer is asked anything. The last element of I-PASS is the receiver saying back what they heard. We recommend the receiving colleague open with one sentence that does the same job: "I can see you called last night about the police report and were promised a call before noon. Shall we pick it up from there?" It costs five seconds, it lets the customer correct the brief, and it is the only moment the customer will feel the handover worked.

What we don't know yet

Nobody has measured how long a person actually has between a transfer landing and their first word to the customer, or how long they spend reading before speaking. The 25-second figure above is arithmetic from reading rates, not observation.

No study has tested whether a structured brief reduces customer repetition or improves satisfaction in a contact centre. The clinical evidence measured clinical errors; the customer research measured stated intent. The link between them is reasoning, and we say so.

No study has shown that escalating on detected frustration improves outcomes, and a 2026 evaluation of eight language models found their thresholds for handing over vary by 25 to 38 points within a single model family, with about two thirds of settings overconfident [41]. Escalation thresholds have to be set and tested per deployment, not assumed.

We found no Arabic-language research on handover or repeat contact in Gulf firms, and no Gulf regulator text on transfers. The UAE reading-rate, code-switching (mixing Arabic and English), bank-customer and regulator evidence above is the extent of what is Gulf-specific.

And none of this has been tested by us. This article is a synthesis of other people's evidence and our reasoning about how to design for it. When we have measured results from deployments, with a method and a sample, we will publish them.

Sources

1. 1.Zendesk, "Zendesk 2025 CX Trends Report: Human-Centric AI Drives Loyalty", press release, November 2024.⁠https://www.zendesk.com/newsroom/articles/2025-cx-trends-report/
2. 2.Dixon, M., Freeman, K. and Toman, N., "Stop Trying to Delight Your Customers", Harvard Business Review, July to August 2010.⁠https://hbr.org/2010/07/stop-trying-to-delight-your-customers
3. 3.Salesforce, "State of the Connected Customer", sixth edition, 2023.⁠https://www.salesforce.com/research/customer-expectations/
4. 4.SQM Group, "Call Transfer & Hold Performance Impact on Csat and FCR", 2021.⁠https://www.sqmgroup.com/resources/library/blog/call-transfer-hold-performance-impact-csat-and-fcr
5. 5.de Haan, E., Verhoef, P. C. and Wiesel, T., "The predictive ability of different customer feedback metrics for retention", International Journal of Research in Marketing, 2015.⁠https://doi.org/10.1016/j.ijresmar.2015.02.004
6. 6.Starmer, A. J. et al., "I-PASS, a Mnemonic to Standardize Verbal Handoffs", Pediatrics, 2012.⁠https://doi.org/10.1542/peds.2011-2966
7. 7.Starmer, A. J. et al., "Changes in Medical Errors after Implementation of a Handoff Program", New England Journal of Medicine, 2014.⁠https://doi.org/10.1056/NEJMsa1405556
8. 8.Starmer, A. J. et al., "Implementation of the I-PASS handoff program in diverse clinical environments: A multicenter prospective effectiveness implementation study", Journal of Hospital Medicine, 2023.⁠https://doi.org/10.1002/jhm.12979
9. 9.Agency for Healthcare Research and Quality, "Use of Structured Handoff Protocols for Intrahospital Within-Unit Transitions", Making Healthcare Safer IV, 2024.⁠https://www.ncbi.nlm.nih.gov/books/NBK613742/
10. 10.Robertson, E. R. et al., "Interventions employed to improve intrahospital handover: a systematic review", BMJ Quality & Safety, 2014.⁠https://doi.org/10.1136/bmjqs-2013-002309
11. 11.Federal Aviation Administration, Order JO 7110.65, Air Traffic Control, Appendix A, "Standard Operating Practice for the Transfer of Position Responsibility", current edition.⁠https://www.faa.gov/air_traffic/publications/atpubs/atc_html/appendix_a.html
12. 12.Federal Aviation Administration, Order JO 7210.3, Facility Operation and Administration, paragraph 2-2-4, current edition.⁠https://www.faa.gov/air_traffic/publications/atpubs/foa_html/chap2_section_2.html
13. 13.EUROCONTROL, "Safety Survey Report: Handover-Takeover in ATS", undated, hosted by SKYbrary.⁠https://skybrary.aero/bookshelf/books/3800.pdf
14. 14.Grayson, R., "Problems in Briefing of Relief by Air Traffic Controllers", NASA Aviation Safety Reporting System, 1980.⁠https://asrs.arc.nasa.gov/docs/rs/13_14_Problems_Briefing_Relief_byATC_and_Altimeter_Errors.pdf
15. 15.Health and Safety Executive, "Shift handover", human factors guidance, current; and Lardner, R., "Effective Shift Handover: A Literature Review", HSE Offshore Technology Report OTO 96/003, 1996.⁠https://www.hse.gov.uk/humanfactors/topics/shift-handover.htm
16. 16.Health and Safety Executive, "Buncefield: Why did it happen?", 2011, paragraphs 42 and 55.⁠https://www.icheme.org/media/10706/buncefield-report.pdf
17. 17.International Maritime Organization, STCW Code, Section A-VIII/2, Part 4-1, "Taking over the watch", paragraph 22. The Code is published by IMO and is not freely available online.⁠https://www.imo.org/en/OurWork/HumanElement/Pages/STCW-Convention.aspx
18. 18.Pace, C. A. et al., "Warm Handoffs and Attendance at Initial Integrated Behavioral Health Appointments", Annals of Family Medicine, 2018.⁠https://doi.org/10.1370/afm.2263
19. 19.Anand, P. and Desai, N., "Correlation of Warm Handoffs Versus Electronic Referrals and Engagement With Mental Health Services Co-located in a Pediatric Primary Care Clinic", Journal of Adolescent Health, 2023.⁠https://www.sciencedirect.com/science/article/abs/pii/S1054139X23001428
20. 20.Poser, M., Hackbarth, T. and Bittner, E. A. C., "Don't Throw It Over the Fence! Toward Effective Handover from Conversational Agents to Service Employees", HCII 2022.⁠https://doi.org/10.1007/978-3-031-05412-9_36
21. 21.Lu, Mo, Luo and Min, "How to initiate human-agent intervention for recovery when chatbots fail: A social information processing perspective", Decision Support Systems, 2026.⁠https://www.sciencedirect.com/science/article/abs/pii/S0167923626001041
22. 22.Lu and Min, "Self-recovery or human intervention? Understanding the role of task type and failure frequency in chatbot failure recovery", Journal of Retailing and Consumer Services, 2025.⁠https://www.sciencedirect.com/science/article/abs/pii/S0969698925002231
23. 23.Crolic, C., Thomaz, F., Hadi, R. and Stephen, A. T., "Blame the Bot: Anthropomorphism and Anger in Customer-Chatbot Interactions", Journal of Marketing, 2022.⁠https://doi.org/10.1177/00222429211045687
24. 24.Brysbaert, M., "How many words do we read per minute? A review and meta-analysis of reading rate", Journal of Memory and Language, 2019.⁠https://doi.org/10.1016/j.jml.2019.104047
25. 25.AlJassmi, M. A. et al., "Effects of word predictability on eye movements during Arabic reading", Attention, Perception, & Psychophysics, 2022.⁠https://doi.org/10.3758/s13414-021-02375-1
26. 26.Genesys, "Set Whisper Audio action", Genesys Cloud documentation.⁠https://help.genesys.cloud/articles/set-whisper-audio-action/
27. 27.Duggan, G. B. and Payne, S. J., "Skim reading by satisficing: evidence from eye tracking", CHI 2011.⁠https://doi.org/10.1145/1978942.1979114
28. 28.Cowan, N., "The magical number 4 in short-term memory: A reconsideration of mental storage capacity", Behavioral and Brain Sciences, 2001.⁠https://doi.org/10.1017/S0140525X01003922
29. 29.Wang, B. et al., "Analyzing and Evaluating Faithfulness in Dialogue Summarization", EMNLP 2022.⁠https://aclanthology.org/2022.emnlp-main.325/
30. 30.Asgari, E. et al., "A framework to assess clinical safety and hallucination rates of LLMs for medical text summarisation", npj Digital Medicine, 2025.⁠https://www.nature.com/articles/s41746-025-01670-7
31. 31.Sim, M., Eiger, Y. and Kohno, T., "AI-Enabled Human Memory Manipulation: Misleading AI-Generated Summaries Distort Human Memory", arXiv preprint, 2026.⁠https://arxiv.org/abs/2609.28820
32. 32.Hamed, I. et al., "A Survey of Code-switched Arabic NLP: Progress, Challenges, and Future Directions", COLING 2025.⁠https://aclanthology.org/2025.coling-main.307.pdf
33. 33.Hernandez Caralt, M. et al., "'Stupid robot, I want to speak to a human!' User Frustration Detection in Task-Oriented Dialog Systems", COLING 2025 Industry Track.⁠https://aclanthology.org/2025.coling-industry.23.pdf
34. 34.Ofcom, "Treating vulnerable customers fairly: a guide for phone, broadband and pay-TV providers", 2020, updated 2022, section 5.3.⁠https://www.ofcom.org.uk/__data/assets/pdf_file/0024/244473/2022-treating-vulnerable-customers-fairly.pdf
35. 35.Financial Conduct Authority, FG21/1, "Guidance for firms on the fair treatment of vulnerable customers", 2021, paragraphs 3.17 to 3.18.⁠https://www.fca.org.uk/publication/finalised-guidance/fg21-1.pdf
36. 36.Central Bank of the UAE, Consumer Protection Standards, Article 8, "Complaint Management and Complaint Resolution", 2021.⁠https://rulebook.centralbank.ae/en/rulebook/article-8-complaint-management-and-complaint-resolution
37. 37.Saudi Central Bank (SAMA), "Financial Consumer Protection Principles and Rules", 2022, Rule 19. Published in the SAMA Rulebook.⁠https://rulebook.sama.gov.sa/
38. 38.Dayan, M., Al-Tamimi, H. A. H. and Elhadji, A. L., "Perceived justice and customer loyalty in the retail banking sector in the UAE", Journal of Financial Services Marketing, 2008.⁠https://doi.org/10.1057/palgrave.fsm.4760085
39. 39.de Matos, C. A., Henrique, J. L. and Rossi, C. A. V., "Service Recovery Paradox: A Meta-Analysis", Journal of Service Research, 2007.⁠https://doi.org/10.1177/1094670507303012
40. 40.Valentini, S., Orsingher, C. and Polyakova, A., "Customers' emotions in service failure and recovery: a meta-analysis", Marketing Letters, 2020.⁠https://doi.org/10.1007/s11002-020-09517-9
41. 41.DosSantos DiSorbo, M. and Ju, H., "Act or Escalate? Evaluating Escalation Behavior in Automation with Language Models", arXiv preprint, 2026.⁠https://arxiv.org/abs/2604.08588