An AI Support Report Your Finance Director Will Read

An AI support report your finance director will read counts contacts that stayed resolved, not contacts that were handled. Most dashboards do the opposite, and the gap between the two numbers is where the budget argument happens.

I’ll build the whole thing here, including the spreadsheet, so you can rebuild it from scratch. The example numbers are invented for illustration. The structure is what matters. It follows on from two earlier pieces on what Klarna’s reversal teaches and who owns what a support bot says.

What does the spreadsheet look like?

You need one workbook with two sheets. The first sheet is called Contacts. It has one row per customer contact, with headings in row 1 and the data starting in row 2. The columns are these.

Column A holds the contact ID. Column B holds the date the contact started. Column C holds the customer problem ID, which stays the same if the same customer comes back about the same problem. Column D holds who handled it, using exactly one of three labels: AI only, AI then human, or Human only. Column E holds the outcome, using Resolved, Escalated or Abandoned. Column F holds Yes or No, answering whether the customer contacted you again about the same problem within seven days.

If you receive 2,000 contacts in a month, they fill rows 2 to 2001. Every formula below points at that range. The second sheet is called Report. It’s where the numbers appear, one per row, with the label in column A and the figure in column B.

How do you count what the AI really handled?

Start with the figure everyone quotes, the containment rate. This is the share of contacts the AI dealt with without a person. On the Report sheet, in cell B3, type: =COUNTIF(Contacts!D2:D2001,"AI only"). This counts how many cells in the Handled by column say exactly AI only. In cell B2, type =COUNTA(Contacts!A2:A2001), which counts the non-blank contact IDs, so it gives you the total number of contacts. In cell B4, type =B3/B2 and format it as a percentage.

With our invented month, 1,400 of the 2,000 contacts are AI only. B4 shows 1,400 ÷ 2,000 = 0.70, so 70 percent. That’s the number that goes in the vendor’s slide.

It is not the number the finance director cares about. Containment tells you the AI was the last to speak. It doesn’t tell you the customer was helped.

How do you count what was actually resolved?

In B5, count the AI-only contacts that ended as Resolved: =COUNTIFS(Contacts!D2:D2001,"AI only",Contacts!E2:E2001,"Resolved"). COUNTIFS counts rows where every condition is true, here Handled by equals AI only and Outcome equals Resolved. Say it returns 910. In B6, type =B5/B3 to get the AI’s resolution rate on its own contacts: 910 ÷ 1,400 = 0.65, or 65 percent.

So of the 70 percent of contacts the AI kept, only 65 percent of those were fixed. Put differently, the AI fixed 910 ÷ 2,000 = 45.5 percent of all contacts, not 70. The other 490 AI-only contacts were abandoned. Those customers gave up, and they’re the ones nobody on the dashboard can see.

Now add the seven-day check. In B7, type: =COUNTIFS(Contacts!E2:E2001,"Resolved",Contacts!F2:F2001,"No"). This counts every contact, whoever handled it, that ended Resolved and where the customer did not come back within seven days. In our invented month, 1,450 contacts were marked Resolved overall, 200 of them came back, and B7 shows 1,450 − 200 = 1,250.

How do you turn it into a cost figure?

Type the month’s total cost in B8. Include the AI platform fee and the people who work the queue. Say it’s $41,000, made up of $3,000 for the platform and $38,000 for people. In B9, type =B8/B2, the cost per contact: $41,000 ÷ 2,000 = $20.50. In B10, type =B8/B7, the cost per contact that was resolved and stayed resolved: $41,000 ÷ 1,250 = $32.80.

That’s a 60 percent jump between the two figures, from $20.50 to $32.80, and neither is wrong. The first answers “what does a contact cost us?” The second answers “what does it cost to actually fix a customer’s problem?” Give both. The second is the one that survives a budget meeting.

The same logic sits behind what I wrote about First Call Resolution a while back. The number is hard because someone has to decide what counts as resolved, and your report is only as honest as that definition.

What goes on the page the finance director reads?

Six lines, in the same order every month. Contacts received. Containment rate. Resolution rate for AI-only contacts. Contacts resolved and stayed resolved. Cost per contact. Cost per resolved-and-stayed-resolved contact.

Add one sentence under the numbers saying what changed since last month and why. Add one customer story, with permission, because a story makes the numbers matter. Then add the line I wish more reports carried: what we’re going to do about the weakest number.

Keep the Contacts sheet behind it. When someone asks how you got 65 percent, you can open it, filter the Handled by column to AI only, and show them the 910 rows. A report people can check gets trusted. A report people have to take on faith gets argued with.

How do you see the human side on its own?

The same sheet answers this. On the Report sheet, in cell B11, type: =COUNTIFS(Contacts!D2:D2001,"Human only",Contacts!E2:E2001,"Resolved")/COUNTIF(Contacts!D2:D2001,"Human only"). The top half counts rows where Handled by says Human only and Outcome says Resolved. The bottom half counts every Human only row. Dividing one by the other gives the human resolution rate.

In our invented month, assume 150 contacts were Human only and 135 were resolved. That’s 135 ÷ 150 = 0.90, or 90 percent. Compare it with 65 percent for AI-only contacts. The humans resolve more of what they touch, but they also cost far more per contact, which is why the cost per resolved-and-stayed-resolved figure matters more than either rate alone.

You can repeat this for the AI then human group by changing the label in both places. That group tells you whether the hand-off works. If its resolution rate is below the Human only rate, customers are losing something in the transfer, and you should look at what the person sees when the bot hands over.

What if you use Google Sheets instead?

Everything above works the same. COUNTIF, COUNTIFS and COUNTA behave identically in Google Sheets. Use two tabs named Contacts and Report. The only change is that, as new rows arrive, you can write the range as D2:D to leave it open at the bottom, instead of typing a new end row each month.

Whichever tool you choose, keep the labels exactly consistent. A cell that says “AI Only” with a capital O is counted the same in Excel but may be mistyped elsewhere. A drop-down list on columns D, E and F removes that problem, and it takes a minute to set up.

How often should the report change?

Rarely. The six lines should stay the same for at least two quarters, so that trends mean something. If you swap a measure every month, nobody can tell whether things improved or the ruler moved.

When you do need to change one, run the old and new definitions side by side for a full month and say so on the page. Add a short note to a third sheet, called Changes, with the date, what changed and why. When someone asks in a year why a figure jumped, the answer is in one place.

And give the report a single owner with a name. That person can say no to a flattering change, which is the most valuable thing a report owner does.

How do you keep the numbers honest?

Three habits. Write down the definition of Resolved and keep it fixed. Read 30 contacts from the Resolved group every month to check they really were. And keep the old numbers when you change the method, so the change is visible rather than hidden.

Tell your boss before the first honest report goes out that the numbers will look worse than the dashboard they’re used to. In the example, containment of 70 percent became effective resolution of 45.5 percent, and then 37.5 percent once repeat contacts were removed (750 ÷ 2,000). That drop is the finding, not an error.

It’s the same argument I made in Support Is Not a Cost Center. Support earns its budget by showing what it produces, and what it produces is problems that stay fixed.