Return to overview
AI in Recruitment
Remo Vloet6 min.

Budgeting recruitment AI usage

Separate the ATS subscription from the provider bill. Budget around your recruiters' work, using your own AI key and identifiable usage costs.

On this page

Your agency pays for an ATS subscription, but AI usage may arrive on a separate bill. With Simply’s bring-your-own-key route, you connect your own API key and the AI provider bills that usage directly. You don’t buy Simply credits for that route. The amount depends on your chosen model and the work performed.

Budget from recognisable tasks: processing CVs, asking questions about records or drafting text. Recruiter headcount alone tells you little about consumption. One colleague processing an archive can trigger more AI calls in a day than the team normally makes in a week.

Key takeaways

  • Separate the Simply subscription from directly billed AI usage.
  • Estimate recurring tasks, including repeat generations.
  • Keep one-off bulk processing outside the ordinary monthly average.
  • Give one colleague responsibility for the provider account and configured limits.

Which costs go to Simply and which go to your provider?

You pay Simply for the selected ATS subscription. When you use your own AI key, the provider charges that usage to your own account. Simply does not sell an intermediate credit bundle for this route. Include both costs in the budget: AI usage isn’t part of the subscription and is billed separately.

Simply’s pricing page describes the own-key arrangement. For budgeting, the useful point is separation: you know where the subscription sits and where to check consumption. Add both amounts for the functions you use and the workload you expect.

Record which provider, account and model the agency uses. A colleague’s personal account makes usage harder to follow when they leave or replace the key. Someone acting for the agency should have access to invoices and responsibility for the settings.

By default you work with a key Simply manages. Once you add your own key in the settings, all AI traffic runs through it and your provider bills you directly.

Why don’t all AI tasks cost the same?

Tasks differ in the amount of input, length of output and selected model. A brief question about one record is different work from processing a long document. Group similar tasks together so their average cost continues to describe something useful about how your team works, rather than mixing every action into one figure.

Language-model providers commonly charge using tokens, the units in which the model processes text. Anthropic’s pricing documentation, for example, distinguishes input, output and caching. One AI action is therefore not a fixed pricing unit. Use costs from the route actually chosen, including any separately charged components.

Asking for a lengthy response when you need one field value can create unnecessary work. Repeatedly regenerating an entire document may also consume more than a focused correction. Give AI a defined task that matches what you intend to do with its result.

The lowest cost per call isn’t the only consideration. If a model repeatedly drops important conditions, the team spends time checking and repairing the output. Keep that human effort separate from provider charges. A cheaper answer you cannot use is not a completed recruitment task.

How do you turn tasks into a monthly budget?

Create a worksheet with expected execution counts and average cost per task. Establish the costs from a small, recognisable amount of your own usage. Separate longer documents and repeat attempts where they materially change the average. This gives you an estimate you can later compare with the actual bill and work completed.

Here is an example monthly workload. Fill in your measured cost for each task:

TaskExpected monthly countCost to supply
Process new CVs120Average cost per CV with your selected model
Run a longer record analysis40Average cost per analysis
Draft a text30Average cost per draft
Regenerate a draft10Average cost per repeat generation

Multiply each row and add the results. If c is measured CV cost, a is analysis cost and t is text-generation cost, estimated usage is 120 × c + 40 × a + 40 × t. The final count includes thirty initial drafts and ten repeats. Add the ATS subscription as a separate line.

Obtaining those figures needn’t become a large project. Ask the administrator to collect relevant calls from a defined period and connect them to the tasks performed. A provider account serving other business applications cannot be attributed entirely to Simply.

OpenAI describes usage and cost reports, including project filters and exports. Use the equivalent records from your own provider to reconcile the bill. You don’t need to build a billing system to prepare this budget.

How should you budget for a peak month or historical records?

Create a separate line for work outside normal incoming activity. Processing an old candidate archive is a one-off task; processing those same records again is another operation. Decide which information you want processed and why before adding the whole archive to a bulk run, so the budget reflects a deliberate choice about the work.

In the fictional example, a campaign doubles only incoming CVs. The estimate becomes 240 × c + 40 × a + 40 × t. Don’t double the whole monthly total when the other tasks stay unchanged. If the team also plans more analyses, add those counts separately.

For historical records, selection determines the workload. Some records may no longer be needed, others already have the required information and others lack only one field. Agree with the administrator what requires ordinary import and what needs AI processing. Moving data and asking AI to interpret its content again are different tasks.

Allow time to investigate partial failures too. Which records were processed and which need another attempt? Don’t immediately rerun an entire bulk job when only some records need attention; you could repeat both work and consumption unnecessarily.

How do you manage costs while the team is using AI?

Agree who may start larger processing runs, who reviews consumption and when the team needs to discuss extra spending. A named owner is more useful than a cost report nobody opens. Check the provider settings with that person: a budget notification, a technical rate limit and a hard spending cap have different consequences.

Choose a notification that leaves time to act. Specify the response: explain additional work, stop an unexpected repeat or approve more budget deliberately. Don’t assume every warning blocks further processing; that depends on the chosen provider setting.

When changing models, review the content as well as the bill. The same task can use a different amount of input or output with another model. Check a few tasks that matter to the agency before using that choice for a large bulk run.

With Simply’s AI-native ATS, provider and model selection form part of the setup. Make that choice around the job: which model gives usable results for your record processing, and what does it cost at the expected volume?

What should the team review at month end?

Compare the estimate with completed tasks and the provider bill. Look for a change in the work first: more incoming records, longer documents, extra generations or a one-off processing run. That lets you revise the next budget without treating every higher bill as a price rise or every lower bill as an improvement.

Also note which tasks people prefer to finish manually. A recruiter can correct one sentence without regenerating the whole document. A large recurring task may, conversely, justify changing the setup. The budget helps make that choice explicit.

Bring your expected workload to a Simply demo. We can discuss how your own AI key, the selected functions and agency activity fit together. You can then assess the subscription and variable usage as separate costs.

Simply is an AI-native ATS and CRM for recruitment teams.

Share this post

About the author

Remo Vloet

Remo Vloet is a co-founder of Simply, the AI Operating System for recruitment agencies: inbox, meetings, sourcing, CV parsing, search and matching, documents and automation in one system. With a background in building complex software, he contributes to the technical vision behind Simply.

LinkedInArticles by Remo Vloet

Frequently asked questions

Do you need to buy AI credits for Simply?

For the own-API-key route, you don't buy Simply credits: your provider bills that usage directly. By default you use a key Simply manages; ask about its terms in that case.

Is AI usage included in the ATS subscription?

No. AI usage is billed separately. With your own key you pay your provider directly, so budget the subscription and the usage as two lines.

Can you quote one cost per CV in advance?

A useful estimate requires the selected model and recognisable inputs. Measure your own CV processing and separate unusual documents; one example amount is not a fixed tariff for every file.

Are failed or rejected answers free?

Don't assume so. A technically completed generation can be unusable in practice. Check actual consumption and your provider's billing terms.

Is a budget warning a hard limit?

Not necessarily. Check whether the setting only notifies or also stops processing. Agree who receives the alert and decides what happens next.

More in AI in Recruitment

Chase less,place more.

Find out how Simply can completely evolve your workflow. No slides, just product.

Cookie settings

Choose which optional cookies we may use; our cookie policy explains each one. Optional categories are off until you turn them on, and you can change this at any time from the footer.

Necessary

Keeps the site secure and remembers your language and these choices. These cannot be switched off.

Always on
What this includes
  • simply_lang · Simply

    Remembers the language you picked in the footer.

    Stores: Cookie, 1 year

  • simply_consent · Simply

    Remembers the choices you make here.

    Stores: Local storage, 12 months

  • Cloudflare Web Analytics · Cloudflare

    Counts page views and measures loading speed, without identifying you.

    Stores: No cookies

  • Cloudflare Turnstile · Cloudflare

    Checks the demo request form is sent by a person, not a bot.

    Stores: No cookies

Analytics

Shows us how visitors use the site, so we can improve it.

What this includes
  • Google Analytics · Google

    Counts visits and shows which pages are used.

    Stores: _ga, _ga_*: 2 years

  • Contentsquare · Contentsquare

    Analyses clicks and scrolling, with form input masked.

    Stores: _cs_id, _cs_c: 13 months; _cs_s: 30 minutes

Marketing

Measures our ads and campaigns, and lets us show relevant ads on other sites.

What this includes
  • Google Ads · Google

    Measures which ads lead to a demo request.

    Stores: _gcl_au: 90 days

  • Meta Pixel · Meta

    Measures conversions and supports retargeting on Facebook and Instagram.

    Stores: _fbp: 90 days

  • LinkedIn Insight Tag · LinkedIn

    Measures conversions and supports retargeting on LinkedIn.

    Stores: bcookie, li_sugr, lidc and others on linkedin.com: 1 day to 1 year

  • Factors.ai · Factors.ai

    Identifies visiting companies and attributes visits to campaigns.

    Stores: _fuid: 400 days

Google Tag Manager loads the Google, analytics and marketing tags above only after you allow them.