---
title: "Deloitte Says 84% of AI Investments Pay Off. Prove It on Your December Gift Backlog First."
description: "Deloitte found 84% of AI investors report positive ROI. Here is how a development office proves it on December gift entry, with baselines and worked arithmetic."
canonical: https://callsphere.ai/blog/deloitte-says-84-of-ai-investments-pay-off-prove-it-on-your-december-g
category: "Business & Strategy"
tags: ["nonprofits", "fundraising operations", "gift processing", "donor acknowledgment", "ai roi", "year-end giving"]
author: "CallSphere Team"
published: 2026-06-22T08:31:49.000Z
updated: 2026-07-25T23:21:41.964Z
---

# Deloitte Says 84% of AI Investments Pay Off. Prove It on Your December Gift Backlog First.

> Deloitte found 84% of AI investors report positive ROI. Here is how a development office proves it on December gift entry, with baselines and worked arithmetic.

Eighty-four percent. That is the share of organisations investing in AI that report a positive return, according to Deloitte's State of AI in the Enterprise 2026. It is a genuinely good number and it is also the least useful number in the room, because your board is not going to fund a percentage. They are going to ask what it did here, in this office, to something they can see.

So pick something they can see. In a fundraising shop there is one process that is measurable, painful, seasonal, and directly tied to revenue, and it is not the one people usually reach for. It is gift entry and acknowledgment.

## Why the thank-you letter is the right thing to measure first

Every organisation says it thanks donors within 48 hours. Very few can produce the report. In December they definitely cannot, because between the 26th and the 31st a year's worth of mail arrives at once: lockbox checks, Classy and Givebutter exports, donor-advised fund checks from Fidelity Charitable and Schwab Charitable, stock that appears in the brokerage account with no name attached, and IRA qualified charitable distributions that arrive as a check from a custodian on behalf of someone whose record is under a different address.

The gift processing clerk keys those into Raiser's Edge NXT, Bloomerang, DonorPerfect or Salesforce Nonprofit Cloud one batch at a time. Roughly half of them do not match cleanly to an existing constituent record. Somebody has to decide whether "R. Alvarez, 14 Linden St" is the Rosa Alvarez who gave in 2019 at a different address, whether the check from a donor-advised fund should be soft-credited to the individual whose fund it is, and whether the $500 from a matching gift portal belongs to the employee or the employer. That decision work is why acknowledgment slips from two days to three weeks, and why the letter dated January 19 arrives after the donor has already made their tax appointment.

**Proving AI paid for itself, in a fundraising office, means showing that the same December gift volume was acknowledged in fewer days by the same number of people — with the error rate measured, not assumed.**

## Capture the baseline before you switch anything on

This is the step everyone skips and the only one that makes the argument stick later. Before you change a single process, pull four numbers out of your database for last December. Not this year — last year, when nothing was different and nobody was watching.

- **Median days from gift date to acknowledgment mailed**, for December gifts only, and separately for gifts of $250 and up, which the IRS requires a contemporaneous written acknowledgment for.
- **Staff hours per 1,000 gifts processed**. Have the gift processing clerk and whoever helps her keep a simple tally for two weeks. You want hours, not a feeling.
- **Percentage of gifts requiring manual matching** — anything that did not auto-match to a constituent record and needed a human to resolve.
- **Correction rate**: how many gift records were edited after the acknowledgment went out. Wrong name, wrong amount, wrong designation, duplicate constituent. This is your error baseline and it is the one that protects you, because a faster process that creates more corrections is not a win.

Write those four numbers on one page, date it, and give a copy to the board treasurer. You have just made it impossible to argue about the result in March.

```mermaid
flowchart TD
  A["Lockbox checks"] --> E["Daily gift batch"]
  B["Classy and Givebutter exports"] --> E
  C["Donor-advised fund checks"] --> E
  D["Stock transfers and IRA distributions"] --> E
  E --> F["Match to constituent record"]
  F --> G{"Clean match?"}
  G -->|Yes| H["Acknowledgment queued same day"]
  G -->|No| I["Exception queue reviewed by gift clerk"]
  I --> H
```

## What actually changes at the exception queue

The 2026 development that matters here is not a fundraising product. It is that assistants like Claude Cowork, which launched in January and expanded to web and mobile in July, and ChatGPT Work, which launched on 9 July, take a goal and work across your files and systems rather than answering questions in a chat box. You can hand one the day's unmatched gift file, your export of existing constituents, and your written matching rules, and get back a proposed match for each line with the reason stated — same last name and same ZIP, matches a spouse record, matches a fund at a donor-advised sponsor whose advisor gave in 2023 — plus a flat "no confident match" for the ones it should not guess at.

The important word is proposed. The gift clerk sees a screen of suggestions with reasons, accepts the obvious ones in a few seconds each, and spends her actual attention on the genuinely ambiguous fifteen percent. Same for the letters: the draft acknowledgment comes back already carrying the correct Publication 1771 language, the quid pro quo disclosure where a gala ticket over $75 was involved, and the designation wording pulled from the campaign record — and a human reads them before they go out.

That is the whole pattern Deloitte found behind the 84%: one messy process, a human in the middle, and a measured before-and-after. Not "adopt AI." One process.

## The arithmetic your finance committee will actually check

Assume a mid-sized organisation: 11,000 gifts a year, 4,200 of them in December, one full-time gift processing clerk at $26 an hour loaded, plus 60 hours of temp help in the December crunch at $22.

|  | Last December (baseline) | This December (measured) |
| --- | --- | --- |
| December gifts | 4,200 | 4,200 |
| Requiring manual matching | 52% (2,184) | 52% (2,184) |
| Average handling time per exception | 3.1 minutes | 1.2 minutes |
| Exception hours | 113 | 44 |
| Temp help needed | 60 hours | 10 hours |
| Median days to acknowledgment | 14 | 4 |
| Post-acknowledgment corrections | 2.9% | 1.6% |

The labour saving is 69 exception hours plus 50 temp hours — call it $2,894 at the rates above. The subscription for two staff seats runs a few hundred dollars for the season. Those are illustrative figures, and the honest point is that the labour saving alone barely justifies the exercise. What justifies it is the ten days. First-time donors who are thanked promptly renew at a meaningfully better rate than those who are not, and in a file with 900 first-time December donors, a few percentage points of retention at an average second gift of $85 is real money that shows up next year. Measure it: tag the December cohort and pull renewal rates twelve months later. That is the number that settles the argument, and nobody else in your sector can hand it to you — it only exists in your own database.

## The parts of gift processing that should stay slow

Do not let anything automatic touch four things.

**Merging constituent records.** An unmerge in Raiser's Edge is miserable and sometimes impossible. Suggestions yes, automatic merges never.

**Anonymous and sensitive gifts.** Donors who ask for anonymity, gifts from board members, gifts that will land on Schedule B, and anything from a donor in a contested estate — those go to a named person, always.

**Noncash gifts.** If a donor needs Form 8283 signed for a noncash contribution, that is a finance director's signature and a real conversation, not a queued letter. Same with anything where you might be seen to be valuing the gift for the donor, which you must not do.

**The major gift thank-you.** The $25,000 gift gets a call from the executive director and a handwritten note from a board member. If your new process makes that letter faster, you have improved the wrong thing.

One more limit worth saying plainly: none of this fixes a dirty database. If you have 4,000 duplicate constituents from a 2018 import, you will get faster, more confident suggestions built on top of a mess. Clean the duplicates first or you are just accelerating the wrong answers.

## Frequently asked questions

### Our data lives in Raiser's Edge NXT. Do we have to move it somewhere?

No. The practical version of this works on exports and imports — you pull the unmatched batch out as a file, the review happens against it, and approved matches go back in through the import tool your database administrator already uses. Direct connections exist and are getting better, but do not make a system change a precondition for testing the idea this December.

### What do we tell donors about their information?

Two things matter. Use a paid business account, not a free consumer one, so your data is not used to train anything, and confirm that in writing with the vendor. And do not put Social Security numbers, full card numbers or bank details into any of this — those live in your payment processor and nowhere else. Then update your donor privacy policy to say service providers may assist with processing, which is what it probably already says about your lockbox bank.

### Is this worth it for an organisation with 900 gifts a year?

The labour case, no. The acknowledgment-speed case, maybe, if December is currently a three-week backlog with one person doing everything. Small shops usually get more out of using the same approach on grant reporting, where a single funder report can eat two full days.

### How long before we know if it worked?

One December. That is the honest answer. You can see the exception hours change inside two weeks, but the retention number that convinces a skeptical treasurer takes a full year to arrive, so start the baseline now rather than waiting for a perfect plan.

One related note: a great deal of December work is not gift entry at all, it is the phone — donors calling to ask whether their gift arrived, whether a December 30 credit card charge counts for this tax year, and how to make a donor-advised fund grant before the cutoff. [CallSphere](https://callsphere.ai) builds AI voice and chat agents that answer those calls and web chats around the clock, answer the routine questions from your own approved wording, and pass real conversations to a person with the details already captured, so the development office is not choosing between the phone and the batch.

---

Source: https://callsphere.ai/blog/deloitte-says-84-of-ai-investments-pay-off-prove-it-on-your-december-g