The Accounting Technology Lab

Brian F. Tankersley, CPA.CITP, CGMA: Welcome to the accounting

Technology Lab sponsored by CPA
practice advisor, with your

hosts, Randy Johnston and Brian
Tankersley, hi everybody. It's

Brian, working hard at the lab
today, but I thought I would let

you know that do have a quick
announcement? We have a

giveaway. We're going to be
giving away one Rico, scan, snap

IX, 2500 receipt edition. We're
going to draw. We're going to

make the drawing right after
April 30. No purchase required.

Suggested Retail value, $599
this is a one has one touch

simplicity, and our friends at
Rico graciously provided it to

us with automatic invoice OCR,
detachable receipt guide for

small and curl documents and
lightning fast, 45 page a

minute, duplex scanning. It
transforms paperwork into clean,

searchable digital data in
seconds. If you want to

register, it's CPA te.ch/scan
snap. The rules are at CPA

te.ch/rules and the privacy
policy is at CPA, te.ch,

forward, slash, scanner, dash,
privacy. And for those of you

attending the video, attending
the video version of this, I've

actually put on the screen now a
QR code that you can capture and

use to register for the for the
again, for the scanner. Giveaway

again. We're giving one away.
We're going to give it away

again right after April 30 and
again, you must, you must be an

accounting professional when all
the rules again are at CPA te

CH, CPA te.ch/rules so anyway,
with that, thanks for joining

us, and here we are back in the
lab.

Randy Johnston: Welcome to the
accounting Technology Lab. I'm

with my co host, Brian
Tankersley, and I'm Randy

Johnston. Wanting to talk to you
today about generative AI in

your practice now, generative AI
has had a lot of traffic over

the last three and a half years,
or something like that. And what

we wanted to do was just discuss
where AI is appearing and how

it's appearing, and how you
might make decisions going

forward for your use of AI. And
you're going to discover in our

discussion with you today that
we think there's multiple ways

that you can consider this and
not everything. Is a large

language model. In other words,
it's not all open AI chat, GPT

or Claude or copilot or Gemini
or whatever you think is the

greatest thing since sliced
bread. So Brian, how would you

like to proceed with our
listeners today?

Brian F. Tankersley, CPA.CITP, CGMA: Well,
I think, I think it's important

to understand here that that
this is going to be an evolution

that everybody's going to go
through and how they work with

things, just like when you think
about how we went from paper

Working Papers, where we had the
reinforcers around the two hole

punch, things at the top and all
of that, you know, it's, you

know, we used to compete. We
used to try to figure out what

the best pencil was and what the
best eraser was. Now I don't

know how. I don't know that I
have any mechanical pencils left

in my in my desk anymore, so
it's, it's really a different

tool set, but it took quite a
bit of evolution to go through

there. So you know the thing
here is that there are things

that AI can do for you today.
And we encourage you to try to

check those things out. You
know, drafting emails,

summarizing, summarizing large
documents where you can go back

and have human in the loop to
check them. You can go through

and and again, use it for
optical character recognition.

You can, you know, in
particular, in marketing, in

areas where you know, you can
use generative systems to clean

up the writing and things you
know you and I've used Grammarly

for years now. You can also use
some of these tools with bank

feeds and update information,
although we got to be very

careful with privacy here. Okay,
so, so we don't, we need to. We

need to, again, be very careful
there, and we but we have, we

have these bots in here that
again we set up and again this,

we look at Claude co work and
the forthcoming, the forthcoming

co pilot bots, as well as open
claw and and these tools can

actually take actions inside of
things. I was actually looking

at an extension for open claw
yesterday that would actually go

in and do do maintenance on Unix
systems, like installing updates

and things like that. So I found
it, you know, it's very

interesting, some of the things
that can be automated now that

that we again, haven't had now

Randy Johnston: before you,
before you leave that thought

behind, I was actually
reflecting on self maintenance

in Star Trek or Star Wars, and
realistically, agents are

probably one of the key things
that you'll have to watch for.

We began talking to you over a
year. Ago about how agents could

be used. You're going to find
the agents you see this year

more sophisticated than the
agents of the past. So all this

maintenance and updates again,
we've got to watch what we allow

the agents to do. Because
they're they are capable of

doing real transactions like
transferring

Brian F. Tankersley, CPA.CITP, CGMA: money,
yeah. So, so they can create

real damage, if we you know with
this, but there are certain

things that you can have them.
You can have them, do you can

have them go through and
summarize your inbound emails in

the morning, like I do with
Claude every every morning, at

eight o'clock, it goes in and
summarizes the inbound emails

that have come and the ones I
haven't responded to from

yesterday, and and those things
to try to help me manage this.

We also, though we're seeing a
revolution in tax research,

whether it's co counsel, or
whether it's blue jay or whether

it's, you know, the Walters,
Clewer tools we've got, we've

got a wide range of tools in
here that are that are taking

new approaches with this, you
know, research, where we used to

have to have a senior tax
manager would take, you know,

or, or again, a technical tax
tax director. We would have

those people doing heavy duty
research, and then they would go

in and and, you know, if you
wanted to get an answer to a

question, it would take, you
know, half a day and cost you a

couple $1,000 at standard rates.
Now, now, when we're looking at

this, we can go in and have the
AI do the first cut at it, and

then we can have those people
look at it and see if that makes

sense to them. And then they can
check the things that don't seem

right and make sure that there
are, there are big holes in

this.

Randy Johnston: So one of the
tricks when you're doing that is

to actually cite original
resources, and you can ask for

that, but we've noticed,
particularly as I just watched,

the evolution in Thomson Reuters
checkpoint edge co counsel to

their latest release of CO
counsel tax, how much more

sophistication is in that
platform. 18 months later, it's

it's a stunning amount of
progress, but to me still going

back to original IRS sightings,
original regulation sightings

and so forth. That helps
minimize some of the

hallucination. But this human in
the loop check is still in my

mind, mandatory. I don't think
any of these tools are good

enough today. You can turn them
loose. Now, the vendors are

trying to tell me they are. I'm
just not quite that comfortable

yet.

Brian F. Tankersley, CPA.CITP, CGMA: Yeah,
the vendors don't have to pay

the malpractice premiums after
you make a mistake. So there's

that. Okay, so the next one is
going to be, we're going to talk

a little bit about audits. And
you know the audit there are a

lot of audit tools out there
that will, again, like data

Snipper and teammate, document
managing that will match things

up, or tools that will scan the
whole ledger, like mine bridge,

or you have data binding tools
like salonis that you can use

against against the enterprise
processes, the automation that

people have set up. You can
again, you can get, you can get

robotic process automation as
well, as well as AI to do this,

to do the data entry again, you
can do, you can also do real

time insights. You know, you
look at what's happened with

ACL, and ACL is has repurposed
itself from an audit tool to now

a GRC tool, where it's doing
continuous monitoring so that

you can catch things as they're
happening, before they turn into

something that that's to the
point where it causes you to

have a qualified opinion, or to
have, again, to have findings,

audit findings, you can catch
them before they've they've

grown to that level of
magnitude. We can also use, we

can also use, use things to do,
to send better client notes. So

again, the you can take tax law
and use tools like ready to

advise, to to make things for a
non expert. You can have AI,

write financial analysis and
again, explain things in human

terms. You know, one of the one
of the most dangerous things

about client accounting services
is our clients may not

understand their financial
statements, and so we really

have to teach them in here. You
know, co pilot, one of

Randy Johnston: the comments
I'll make here, which you can

hear in a separate accounting
technology lab from the

accelerator is there are new
generation products like ping

assistant that not only do
transcription and take notes,

but also provide insights. And
you know, one of the most common

questions that I've answered in
the last 90 120 days is talk to

me about the security of the AI
transcription tools. What can we

use? And of course, Brian and I
both use transcription tools.

This podcast, for example, has
been processed by otter for some

time. We recognize that it
doesn't always get all the words

right, but it's sure be. It's
trying to sit down and

transcribe it like you you did
in the old days, but tools like

ping assistant really do provide
better client notes, much

faster, in a much superior way.

Brian F. Tankersley, CPA.CITP, CGMA: Yeah,
yeah. So, you know, there are a

lot of things here, but again,
the these, you still have to put

the human in the loop, because I
find that, you know, as good as

otter is, you know, when we put
this thing out, it will get

things wrong, and I will correct
some of them, and some of them I

will just let sail through
knowing that it's machine

transcription, but it's, it is,
it does. It takes something that

would take, easily, take an hour
per episode to create a

transcript off of this, and it
does it pretty much instantly in

the background while I'm doing
other things. You also do want

to take the busy work out of
your out of your practice, so

this auto nag thing that we talk
about, you know, tools like

trust and KVA can handle this
document chasing for you. You

know, lisio does that most of
the tools will do, will do the

follow up in here. You can also
have agents to file your

selected engagement letters as
they come in. And we're seeing

more and more products come out
with tools that will

automatically file those things.
You know, we have dashboard

statuses to show you what's
happening, so you can see what's

on fire in here. And And again,
there's a when a client uploads

something, you can have it
notify you immediately and set

off some triggers.

Randy Johnston: And Brian, we're
lucky enough earlier this week

to have a in depth discussion
about how things are classified

and filed and what makes sense
and what doesn't make sense. And

you know, what we're trying,
trying to do with vendors who

are creating these products is
to give them enough insights

that they can understand what
you need do your work in a

practical way, day after day.
And I think the early versions

of these will be good, but the
later versions will be even

better. But I don't think it's
worth waiting for the later

versions based on the
productivity that you can gain

right now. So we're going to go
through a pretty radical

transition in my mind, on both
document management and

workflow, and realistically,
this taking the busy work out of

the practice is such a big
advance, agreed.

Brian F. Tankersley, CPA.CITP, CGMA: We
also do want to keep our data

safe, though. So we do want to
keep our PII away from the

public chat bots. Again, the
enterprise plans typically will,

will exclude the again, the AI
data getting trained in here.

The other thing to remember
here, by the way, is that is,

again, with the legal
profession, the courts have said

that the data that you put into
these large language models

publicly is now discoverable.
And so there's a whole new

compliance challenge here. And
so the question is, do you want

to, you know, do you want to put
these things on proprietary

models, or proprietary instances
of models, like those you can do

in in Azure. You know, again,
the the thing about this is that

is that it is important that you
understand the risks and the

regulatory risks, and that you
that, again, you don't make

catastrophic mistakes that cause
things to be discoverable,

either in litigation support or
in, God forbid, in litigation

against you.

Randy Johnston: Yeah. So one of
my learnings in the past few

weeks, but I kind of knew this.
I just want to call it out for

you. Trying to keep the data
safe from the AI engines is one

of the advantages of Microsoft's
co pilot. It does take some

setup, and there's some tagging,
and there's misunderstanding

between advanced levels of
Microsoft, 365, like e5 being

required for purview and the
tagging versus Business Premium,

sophisticated enough to really
handle most firms needs further

the tagging efforts are labor
intensive today and very error

prone. So one firm that I had
the privilege of talking to in

the past week or so said, Look,
we went to all this effort to

tag everything, and when we
turned it on for tax season,

nobody could get any work done,
and so we had to turn it all

back off. So I want you to
understand that there are

vendors who will try to come to
you to talk about AI strategies

and data safety, and they talk a
good game, but could put you at

risk on the other side of
productivity when they're trying

to get safety. You know, we've
had that discussion in pre prior

podcast with you. There's always
a compromise between ease of use

and security and generally when
it's. Easy to use, it's not very

secure. And if you make it too
darn hard to use, it's secure.

But nobody can get anything
done.

Brian F. Tankersley, CPA.CITP, CGMA: But
we also do need to, you know,

you've said for 15 or 20 years
now that if you have an

accountant keying in data, you
have a broken process and and I

think you know, this is one of
the areas that that we've got to

really hammer on now, because a
lot of you are still keying in

data. So you know, again, this,
again, you can let makers hub,

AI or Dex or others get the
information. In there are

general ledger platforms like
like digits that have that will

actually ingest invoices,
extract all the information. If

you haven't seen the payables
infrastructure in digits, then

the bill pay in there. It's
actually very, very elegant the

way it's been done, you know.
Again, letting the you know,

again, focusing on the errors
that the system flags and

letting the you know. Again, the
things were far enough into this

journey with OCR and data entry
that that we can almost start to

let some of the systems run on
on near autopilot, where the

system is checking itself. But
again, you can do this on bills

and and again you can try this
and see how it works. You know

many of you already do hands off
processing on bills with your

with your light bill or your
water bill, where you just set

it up to automatically draft
your account every month, as

opposed to going in and having
to transmit the money to them in

here.

Randy Johnston: Yeah. So on this
point, though, most of you are

aware now that unfortunately,
bot keeper did fail, and that

was really sad in my mind. But I
believe that there is a piece of

learning from Enrico Palmero,
you know, again, I like Enrico

well, but he said, Look, our
product was working. We put a

lot of money to make it into the
product to make it work, but we

started having a lot of merger
acquisitions, and when people

started to consolidate, the
number of licenses and

subscriptions that we had
dropped precipitously. And one

of the things I asked you to
begin considering as far back as

2019 was to control your SaaS
subscriptions. And so you'll

continue to hear about
saspocalypse, and I'm not sure

that bot keeper didn't really
fall prey to merger acquisition

and this constriction of SAS
licensing. But was it working?

Yes, and were they able to make
it work? Well, yes, and when I

look at other vendors, like
makers hub.ai, their product is

working beautifully. So the
question will become, how much

automation can you throw at this
and how much of your regular

labor can you take out of your
transaction processing, in tax

and in audit and in client
accounting services Brian has

laid out so far.

Brian F. Tankersley, CPA.CITP, CGMA: Now
we also do need to think about

this. You know, we we just had,
we just had one of the big four

actually asked for a reduction
in fees because of the

automation that was being
deployed by their external

auditor. So, so we do need to
think about value pricing in

here. You know, ed class and Ron
Baker had been beating the drum

against against the billable
hour for 20 or 30 years that I

know of. And and as we, as we
think about this now, switching

to flat fees is, is again a way
to keep things, keep things

steady. As you, as you implement
more and more technology, you

also want to sell your
expertise, not your calendar,

and again, reward your team for
getting things right, as opposed

to just hours. There are things
other than ours that are

important, that that need to
again, need to be rewarded. And,

you know, again, when we're
looking at AI and these, these

technological changes, people
are going to make mistakes,

okay? And we have to make sure
that we don't crush people that

make mistakes if they're if
they're doing them in good faith

and their thought process is
correct. You know, we still want

to do an after action analysis
to see what we got right and

what we didn't. But again, if
you, if you come down too hard

on people that make those
mistakes,

again, to you want to talk, have
your AI talk like a tax partner

the industry context is going to
come first. So again, you want

to make sure you've got the
right wording in here. You also

want to set length limits so
that things get boiled down to a

certain number of things don't
take the first draft. You know,

again, I've used multiple
engines to clean up analyzes and

things in here. You. Again, if I
if I don't like, if I get two or

three things I don't like out of
one engine, I quite often will

switch. You also can upload to
PDFs or other things to keep

things factual. So I will tell
you that I often upload upload

documents for context, like when
I'm coming up with names for

courses, I'll quite often upload
an outline for the course and

then let it come up with with
something. But after it's

analyzed, the text or Word
document that has the things

we're going to talk about and
and it generally does a better

job as a result. So again,
that's that's one way to put in

here. You do want to have that
human in the loop. The tax code

is your source of truth, so you
want to, you want to, you want

to verify every site in here.
You can use, you know, again.

You can use tools like, like
count, co counsel, tax and

answer, connect again. You can
use them to double check claims

that you have other places. But
you do need a human being to

look over things first and
again, there are places where

things make mistakes. You know,
I was, like I mentioned in a

previous podcast. One of the
things that that I was actually

working on yesterday was I was
setting up an instance of a

libre chat, which is an open
source tool that you can use to

work with multiple large
language models. And it kind of

gives you an infrastructure
where you can, you only have to

key the data in once, and you
can send it to multiple models

and try it out different places
and do more sophisticated

things. So I was actually
working on this, and I asked a

simple question, what were the
what were the final placements

in the DC, in the drum corps
International Finals of valley

fever for all seven years of its
existence, from 1979 to 1986

Okay, so there. And so I asked
that notice, there's eight years

in there, but they didn't feel
the core in 1985 and so there

were seven years that they
existed. Well, it every single

engine made up things in there.
Nobody got me the right answer.

The best answer I got was, I
can't figure this out, and

you're going to have to look up,
here's some sources where you

might go to find this, because I
can't do this. But it's, it's,

it's important here, because
that's a very arcane piece of

knowledge that I had about,
about some of that information.

And I think it's, again, I think
it's important for you, for

everybody, to understand that
this, you know, this, this thing

that we're doing here, it will
make especially if you're

dealing with art, with the less
information there is on it, the

more likely it's going to blow
it and blow it big time. Three

different things made up facts
in front of me, and I was

looking at it, going, Wow, that
that's hallucination in action.

Randy Johnston: Well, Brian,
this is too much information for

our listeners, but I just can't
resist today, because we had the

good fortune of attending
grandson's birthday this week.

And I'm going to be specific,
which maybe is also something I

shouldn't do, but 10th birthday
this week, and I pulled him

aside and said, you know, 10th
birthday is a big deal. You

know, obviously Artemis launched
on your 10th birthday. You'll

remember that forever. But I
said, I personally remember my

10th birthday because I got to
go to the National Guard Armory

and learn to twirl a rifle with
the sky riders drum corps. So

that's what I was doing on my
10th birthday. And he kind of

looked at me and said, What's
twirling a rifle? But for some

of our listeners that have seen
these performance groups

twirling batons and rifles and
so forth. I actually got to

learn to do that on my 10th
birthday. So how many people,

though, would know sky riders
and they are performing this

summer here in my hometown?

Brian F. Tankersley, CPA.CITP, CGMA: Well?
And those rifles are dangerous.

I mean, if you hit yourself in
the head, you can, really, you

can go to the hospital. It's,
it's it's pretty

Randy Johnston: serious stuff.
It explains a lot, doesn't it?

Yeah, well,

Brian F. Tankersley, CPA.CITP, CGMA: I
say, you know, I played drums

so, you know, I, I hit my head
different ways. But anyway,

cool. So again, you want to, you
also want to keep your firm

running smoothly. So you can use
AI to kind of monitor things. In

here, you can check bi or pro
staff to catch those log log

jams. Or, you know, again, we
can, we can look at, you know,

there's, there's all manner of
tools. You can look at

historical data. Again, is your
crystal ball, you know, you can

look at the AI scheduling. You
know, there's the good

scheduling tools that are out
there, like Empire suite and and

others that are that are in the
that are in the space, many of

them use linear programming and
other approaches. So AI is

helpful. But you know, again,
you look at the tax scheduling,

for example, that that that you
have in some of these engines,

and, and again, AI can help, but
it may. Not be a replacement.

It's also important here to look
at, look at crunch times, and

see who's struggling, and see
those tsunamis that people are

about to experience, that that
nobody else knows about. So you

can, you can, you know again,
avert the, avert the disaster

that's coming.

Randy Johnston: And we're
looking for AI to assist with

these things. But there is some
tone of knowledge that the

scheduling products still
struggle with. And I think it

might be, you know, there's too
many variables for them to

process through AI. They're
getting better. Not saying they

aren't, but you know, again, you
think about what you know and

what you do when you're
scheduling tax or audit teams,

Brian F. Tankersley, CPA.CITP, CGMA: you
know, Randy, I think, I think,

you know, this really brings us
to, you know, we've got some

additional things we could talk
about, but I think this kind of

brings us to a, again, to kind
of a wrap up on this episode. So

again, the to get the tools in
your daily workflow, it's like

exercise. You have to make
yourself do it, and you will not

like it at the beginning. Like
exercise, okay, but you have to

fit it in. Let copilot draft
your emails. Let quick run

QuickBooks. Ai sorting on a few
clients. You know, try a client

or two on digits or will it, or
will it, or some of the other,

some of the other AI based
engines, ask your tech savvy

staff for honest feedback before
you buy licenses. So try it out

with some people. See what
happens. Are your clients going

to like this or not? I don't
know. You know, if you have,

you're doing litigation support,
they may not like the AI use of

AI, or if they're in a
particularly litigious industry,

like handgun manufacturing or
something this. This may not be

a place where, where it fits
again, check in every few

months, see what's going on and
again, if tools not saving time,

stop using it similarly. Talk to
other people in the industry and

and see where you can make
progress. You know what kind of

approaches they're using,
they're making things, and with

that, Randy, I'll turn it back
over you to bring us home. Well,

you know,

Randy Johnston: this is pretty
straightforward, because you

know your comment is, you know,
don't just stand there, go do

something. And of course, Brian
and I are both running ergonomic

devices that monitor us, called
deep care, and my deep care

ergonomic device is saying it's
time to stand up. So in any

case, we'll talk about deep care
in another episode, but we

wanted to give you an overview
of what we believe is the state

of AI adoption. But we have more
to come in future accounting

Technology Labs, we'll talk to
you again real soon. Good day.

Brian F. Tankersley, CPA.CITP, CGMA: Thank
you for sharing your time with

us. We'll be back next Saturday
with a new episode of the

technology lab from CPA practice
advisor. Have a great week.