What CFOs Should Settle Before Rolling Out AI in SAP Business One
Know which tool is answering the question
“AI in SAP Business One” is a loose label for several features that work in different ways. When someone demos it to you, ask which one is on the screen. SAP’s documentation for the SAP Business One Web Client, version 10.0 FP 2608, covers two of them, Ask AI and AI-assisted user-defined queries. Whether you can use either depends on the release you have installed, how it’s configured and whether the prerequisites are in place, so check those before you put a date in the rollout plan. Ask AI lets a user ask a question and supply context with it. The answer is built from general knowledge and public information on the internet. According to SAP, the feature cannot read, query, create, update or delete Business One data by itself. A user can, however, include selected field values in a question, and those are sent to the AI along with it. AI User Defined Query is for analysing your ERP data. You describe what you want in plain language and it helps turn that into SQL. Finance should review the SQL it produces as well as the numbers that come out of it. Anything a partner has built needs to be looked at separately. Depending on its design, a partner solution may connect to other systems or support transactions. Don’t judge a separate integration by the limits of standard Ask AI. Find out for each one what it can access and what it can change.Start with a report finance already understands
A receivables report is a sensible first pilot, for the simple reason that finance already has a version of it to compare against. A request like “Explain our working capital position” is much harder to begin with, because both the tool and the person reviewing it have to guess which balances and assumptions belong in the answer. Keep the first task narrow. Use one company and a reporting date everyone has agreed on, and write down the date basis, what counts as outstanding and what is excluded. Then ask the implementation team to show you the query they propose and to reconcile its results with the report you already use. SAP’s guidance on query creation says AI-generated queries are validated automatically. That helps, but passing validation doesn’t mean the calculation follows your reporting policy. Here is a made-up example. An invoice was raised for EUR 120,000. Since then a payment of EUR 30,000 and a credit of EUR 10,000 have been applied to it, so EUR 80,000 is still outstanding. A query that picks up the original invoice total reports EUR 40,000 too much. The reporting date matters as well. If that payment came in after month end, the balance you see today isn’t necessarily the balance at month end. An answer can look perfectly reasonable and still mislead you if the question was about one cut-off date and the calculation used another. So your test cases need to cover partial payments, credits, cancelled documents and transactions posted close to the reporting date. Keep the agreed definitions in the same place as the approved query. If either one changes, run the reconciliation again before anyone relies on the new output.Global reporting needs more than a translated prompt
Once the report works for the first company, the natural next step is to copy the setup to all the others. This is the point at which the differences between your entities start to matter. Say a second company has one unpaid invoice of USD 50,000. At a rate of EUR 0.90 to the dollar, which is used here only for illustration, that is EUR 45,000. Add the EUR 80,000 from the first example and the group has EUR 125,000 outstanding. A query that adds up the two original invoice totals arrives at 170,000 instead. That figure ignores the payment and the credit, and it adds euros to dollars. With two invoices the mistake is easy to see. In a report with thousands of rows it may not be. Finance has to state the reporting currency, the source of the exchange rate and the conversion date, and should not leave any of the three to assumption. You also need to know which entities the report covers. Ask which company databases feed into it and how their records are combined. Ask whether intercompany balances are included, and whether the output is meant for local collections or for group reporting. Being able to query one company doesn’t mean a group-wide analysis has been set up. Have each local finance team check the examples against its own business. A correct translation of “overdue” doesn’t tell you whether the report is based on contractual due dates or on an internal collection threshold. Agree what the term means financially, and after that test how the request is worded in each language your teams work in.Follow the information through the service
An SAP B1 data security review should begin with a diagram of how the data will flow. Ask the team to trace a single request from the user’s screen to every service that receives it, and to show any records that are kept afterwards. Your ERP database may sit in an approved location while a connected service processes information somewhere else. Knowing where the database is hosted therefore answers only part of the question. Ask for specific answers on which fields are transmitted, where they are processed, how long they are retained, whether they are used for model training and who has access to the service logs. For each answer, find out whether it is a commitment in the contract or something that depends on configuration. Where it applies, ask the same about subprocessors, and about what happens to retained information when the service ends. A question your team asks regularly will rarely need every field that is available. A request about shipping options might need the item dimensions and the destination country. The customer’s bank details would add nothing to it. Look closely at free-text notes, since they can hold personal information or confidential comments that have no place in the request. In an international group, bring in the people responsible for privacy and for contracts in each country concerned. They should assess the processing arrangement as it will really operate, against the requirements that apply there. If approval is given centrally, record which entities and which data flows were reviewed, because that makes the approval easier to defend later.Give reviewers a job they can actually perform
Saying “finance will check the answers” doesn’t amount to a review process. Whoever does the checking needs to see the underlying figures and the calculation, and needs to be told what the output is going to be used for. With a receivables query, the check might be to compare the total with an approved report and look into particular exceptions. With a written explanation of why margin has fallen, the reviewer’s task is to separate what was observed from what is being suggested as the cause. If margin went down and discounts went up in the same period, that alone doesn’t prove the discounts caused it. Someone should also test access with ordinary user accounts. A demo run by an administrator can’t tell you what a regional controller is allowed to see. SAP documents the authorisations Ask AI requires, and these include separate permissions for managing public question templates. Decide who is allowed to publish reusable questions, and then test as the users those questions are meant for. Try some requests that ought to fail. Check whether shared queries, downloaded results or technical accounts let a user see information outside their responsibilities. Remember that an account with read-only access can still reveal sensitive figures. Treat the division of work below as a practical starting point. The product doesn’t require you to work this way.| Work to be agreed | Lead responsibility | Record to keep |
| Define and reconcile the figures | Finance process owner | Approved calculation and test results |
| Configure and test access | ERP administrator | Permissions and access test outcomes |
| Explain connected services | Implementation partner | Components, data flows and limitations |
| Assess data-handling terms | Privacy or legal lead | Reviewed terms and decisions |
| Decide whether to expand | CFO or delegated sponsor | Pilot findings and open issues |
Work out whether the pilot saves useful time
Getting the first answer is only part of the job. Measure how long users spend preparing the request, checking the result and correcting it, and count the support time when something goes wrong. Suppose a task takes 40 minutes today. In the pilot, generating the output and checking it takes 25 minutes. That saves 15 minutes each time, so over 60 comparable tasks the saving is 900 minutes, or 15 hours. These numbers are invented for the example and are not a forecast of what SAP Business One AI will deliver. Set that time against the cost of setup, training and ongoing charges. Ask which subscriptions, service consumption charges or support costs apply to the arrangement being proposed, and don’t assume all of them are covered by the ERP licence. Time freed up is not the same thing as cash saved. If the finance team spends the hours it gets back on chasing overdue accounts or investigating reconciliation differences, that can be worth having even though the payroll bill stays the same. Describe the benefit honestly, and measure the work that gets better. Accuracy belongs in the same review. A process that is quicker on average and now and then produces a material error needs more work before it is used more widely. Keep a record of the failures alongside the successes, and of the effort it took to put each one right.Expand when the evidence supports it
When it comes to AI governance for SMEs, a short record of the pilot can be more useful than a long policy. It should tell the sponsor what was tested, what got better and what is still unresolved. Before access is extended, the CFO should be able to answer these questions:- Do the agreed financial test cases reconcile, and have material differences been resolved?
- Have both permitted and restricted users been tested, and did each get the expected result?
- Have the relevant data flows and supplier terms been reviewed?
- Can users explain what they are allowed to do and when a review is required?
- Is the benefit still worth having once checking and support costs are counted?
- Who can suspend access if something goes wrong?
Where Ingold Solutions can help
Ingold Solutions provides SAP Business One consulting, implementation, training and support, including on international projects. Its MyWave AI offering is a separate solution, which you should assess alongside the standard Web Client features. When you appoint an SAP Business One agency, ask for a proposal that names the tools involved and says who will test the financial results. It should separate standard features from configuration, integration work and additional services, and it should cover the ongoing responsibilities too. Bring one finance process to that conversation. It could be a report that takes too long, a question that keeps coming up or a calculation someone still does by hand. If the process is clearly defined and there is something reliable to compare against, your team can judge whether AI improves it before you commit to a wider rollout.Frequently Asked Questions
Not on its own. SAP states that the Ask AI feature it documents cannot independently create, update or delete Business One data. A partner integration may have different capabilities and permissions, so review it separately.
Yes, once they have been properly tested and reviewed. Reconcile any new or changed query against approved calculations before it goes into routine use. Passing technical validation doesn’t show that the figures are financially correct.
A single, defined use case, together with the companies, users and data it involves. Agree the financial tests and the review of data handling before you decide who else can use it.
Applicable for Package
Optional